This is Part 2 of a 3-part series on building an AI-powered marketing team. Last week I covered the individual trap of AI and how to build your first shared system. This week: how to tell if that system is actually paying off, and whether it's safe to trust.

80% of marketers feel pressure to adopt AI right now, and 89% of that pressure comes from the C-suite or the board, per Supermetrics’ 2026 Marketing Data Report.

Here’s the other half of that same report: 52% of marketers say external teams define their data strategy and measurement, not marketing, and nearly four in ten still can’t prove ROI across channels.

Put those two together and you get a familiar shape. Marketing is told to move fast, without owning the controls that make it safe or the measurement that proves it’s working. That’s not two problems. It’s the same problem twice: responsibility without ownership.

Once a workflow moves off one person’s laptop and into shared infrastructure — which is what Part 1 was about, a mistake stops being one person’s mistake. It’s the whole team’s exposure.

Why “is it safe” just got a deadline

The EU AI Act reaches full enforcement in August 2026 — five weeks from this issue — and it mandates documented controls for high-risk AI systems. Marketing content generation, personalisation, and targeting can fall inside that scope depending on how they’re used.

Many of us aren’t based in the EU, the practical effect is the same: “we’ll figure out governance later” is no longer a safe default, because “later” now has a date attached to it.

But the deadline is really just forcing a question that shared AI workflows already raised in Part 1: once a workflow has one owner and touches shared data, who decides what needs a human check before it ships? Without an answer, you get one of two failure modes — either everything gets manually reviewed (which kills the speed AI was supposed to buy you), or nothing does (which is how a hallucinated stat or an off-brand claim ends up in a client-facing deck).

The three-tier guardrail test

Not every AI output carries the same risk, so it shouldn’t get the same review. Sort outputs into three tiers before you decide what needs a human in the loop:

1. Tier 1: Internal and exploratory. Brainstorming, first-draft outlines, research summaries for your own use. Nobody outside the team sees this. Light or no review needed, and the cost of an error is simply a wasted five minutes.

2. Tier 2: Team-facing but not public. Internal briefs, sales enablement, drafts that other teammates will build on. These need a peer review, not a formal sign-off — someone else on the team should read it before it becomes the basis for someone else’s work.

3. Tier 3: External, high-stakes, or regulated. Anything that makes a claim, cites a number, mentions pricing, touches personal data, or goes external (ads, client decks, published content). This tier requires a named human sign-off before it ships, and a record of who reviewed it and when.

Most teams don’t need a review board, just this triage step done once, written down, and applied consistently — the same discipline Part 1 recommended for documenting a prompt: treat “what needs review” as a process, not a judgment call made fresh every time.

Why “is it working?” is not that easy to answer

I wish I could just answer “the team likes it” and “we’re faster now” when finance asks what a workflow is worth, but feeling faster isn’t the same as proving value.

It’s easy to know a workflow saves time, but it’s much harder to know whether it’s actually moving a number the business cares about, and the data backs this up: despite the pressure to adopt AI, nearly four in ten marketers still can’t prove ROI across channels, and over half report pressure to cut costs while maintaining results.

Teams are being asked to do more with AI and prove it’s worth the investment, often without ever having set up a way to measure it. The usual failure here isn’t a bad tool. It’s that “time saved” became the only metric anyone tracked, and time saved doesn’t survive a budget conversation.

A measurement framework you can actually run

• Pick one business metric it should move. Not “productivity”, but something finance already tracks: conversion rate, cost per lead, time-to-first-draft on a deliverable that has a deadline, win rate on proposals. If you can’t name the metric, you’re not ready to claim ROI yet.

• Baseline it before you scale the workflow. You can’t prove lift without knowing the starting point. If the workflow’s already team-wide, use the last quarter before rollout, or a comparable team/segment that hasn’t adopted it yet.

• Instrument the actual touchpoint, not the team’s general sense of speed. Track the specific step the AI workflow touches — first-draft turnaround, review cycles, time-to-send — not a vague team-wide productivity feeling.

• Set a kill-or-scale review date. Same cadence as the quarterly workflow review from Part 1, but with a decision attached: if the metric hasn’t moved by the review date, the workflow gets fixed, downgraded back to a personal tool, or retired. Infrastructure that nobody’s allowed to kill isn’t infrastructure, just a sunk cost with a Slack channel.

Where this leaves you

Safe and working aren’t two workstreams. They’re one question asked twice. A workflow with guardrails but no proof is safe and unaccountable, nobody can tell if it’s worth keeping. A workflow with ROI but no review tier is measured and safe, right up until an unreviewed claim ships under your name. Once it’s shared infrastructure, you don’t get to pick one.

Part 3 covers adoption: getting the team to actually use what you’ve built, once it’s proven and safe. See you then!

The AI;DR

Elsewhere in the AIverse

OpenAI launches Presence, its first real enterprise agent platform. Presence deploys voice and chat agents scoped to one job at a time — billing, claims, IT support — with OpenAI’s own engineers doing the implementation and a Codex-powered loop proposing updates as gaps surface in production. This is a services play as much as a product one — OpenAI’s field team, not a self-serve dashboard.

Black Forest Labs puts one model behind images, video, and robots. FLUX 3 generates up to 20-second video clips with synced audio, and the same backbone is already running on Audi’s productiohgjb mn line handling soft-material assembly. FLUX 3 Video is in early access now.

Washington and Beijing both start reaching for the export-control lever. Beijing is reportedly weighing new controls that would block foreign users from downloading Chinese open-weight models onto local servers, while the Trump administration is debating a reciprocal move and weighing sanctions over alleged IP theft via model distillation. If your stack includes any Chinese open-weight model (Kimi, DeepSeek, Qwen), access and hosting terms could shift with little warning.

I’ve been putting this one off for weeks now. Every time “AEO” showed up in a LinkedIn post or a summit deck, I’d nod, save the link, and go back to whatever tool review was due that week. It felt like the marketing equivalent of “I should really start flossing.”

I don’t think I get to do that anymore. “AEO” — answer engine optimisation, or GEO if you’re on that side of the acronym war — has been showing up in every AI marketing recap for months, including a recent talk I attended where marketers from the likes of HubSpot, Estee Lauder, and DoorDash threw around numbers like “4-6x higher conversion from AI-chat journeys versus classic search” and claims like “early signs of organic traffic eroding as AI overviews take over clicks.”

Those are figures from large US brands, several timezones away from us in APAC, but the mechanic underneath is real: AI models build answers from third-party sources — Reddit, review sites, publishers, forums — and if you're not in that pool, you don't exist to an entire discovery channel that's only getting bigger.

Here’s the step-by-step, without the consultant markup.

Step 1: Find out where you already stand

Before you build anything, check your current visibility. Pick 5-10 questions your buyers are asking — the ones that would normally send them into a Google search and toward you. Not “what does [your company] do,” but the actual problem language: “best project management tool for remote teams,” “how to reduce SDR outreach time with AI.”

Run each one cold through Claude, ChatGPT, Perplexity, and Gemini, logged out or without custom memory if you can manage it, so you’re seeing what a stranger sees. For each query, note three things: do you show up, does a competitor show up instead, and is whatever gets said about you actually accurate.

This takes under 30 minutes and costs nothing. It’s also the single most-skipped step, because it’s more satisfying to jump straight to “fixing” things than to find out if there’s anything to fix yet.

Step 2: Treat it as two separate jobs, not one

AEO isn’t SEO with a new coat of paint. SEO rewards ranking. AEO rewards being the source a model decides to cite or paraphrase — a different game with different inputs.

Job one is earning third-party authority in the places models actually crawl: Reddit threads, niche communities, comparison sites, tier-two publishers. This turns PR and community work into answer-coverage work, not just brand-awareness work. If nobody’s talking about you anywhere a model can see, no amount of on-site optimisation fixes that.

Job two is restructuring your own content to be citable. Less “ultimate guide to X,” more a direct, specific answer to a specific question, near the top of the page, with the kind of stats, quotes, and clear structure a model can lift cleanly. FAQs and schema markup help here, but they’re the finishing touch, not the strategy.

Step 3: Fix the citation pool problem first

Most teams jump to Step 2’s second job — rewriting content — because it’s the part they control. But if nobody’s referencing you anywhere, a perfectly restructured page still has nothing to be cited from. Start with one or two genuinely relevant places you could realistically get mentioned: a comparison post, a community thread, another practitioner’s newsletter or blog. Not favour trades — actual relevance. This is slower and less controllable than rewriting a page, which is exactly why most people skip it and why it matters more.

Step 4: Don’t copy the playbook wholesale if you’re building for APAC

A lot of AEO advice assumes Reddit is where your audience’s conversations live, that Google sits underneath every AI overview the way it does in the US, and that everything worth optimising is in English.

None of that holds evenly across Southeast Asia. A meaningful share of real discovery happens inside WhatsApp groups, Telegram, Line, and closed communities that no model is crawling, which means “get mentioned on Reddit” might optimise for a citation pool your actual audience never touches.

If that’s your situation, Step 3 needs a genuinely regional answer: figure out where your buyers’ real conversations happen, and treat that as your citation target, even if no AI model indexes it yet. Building visibility there still compounds, just not necessarily inside the systems these AI tools are trained on today.

Step 5: Re-test and go deep on one thing at a time

Don’t try to run all four jobs — audit, third-party mentions, content rebuild, regional adaptation — simultaneously. Pick the one query or page where you’re weakest, fix it, and re-run your Step 1 audit in 60 days to see if anything moved. If it did, you’ve got a repeatable process. If it didn’t, you’ve learned something cheap instead of something expensive.

That’s the whole playbook: audit manually first, split the two jobs, fix the citation pool before the content, adapt for where your audience actually is, only bring in a tool once manual tracking gets tedious, and re-test before you scale it. No agency retainer required.

Bonus: Once you outgrow manual checks, here’s what to track with.

The Step 1 audit works fine by hand for a while, five to ten queries, a spreadsheet, 30 minutes a month. But once you’re tracking more prompts than you can reasonably copy-paste, a few tools have emerged specifically for this:

Otterly.AI: the cheapest entry point, starting around $29/month for 15 prompts across ChatGPT, Perplexity, Google AI Overviews, and Copilot, with citation analysis. Good first tool if you just want ongoing visibility tracking without committing to a platform.

Peec AI: starts around $95/month, built specifically for AI search analytics, tracks prompt-level visibility across models. Pricing scales with how many prompts and engines you track — Claude sits behind an enterprise tier, and adding extra engines like Gemini costs more per month, so check what’s actually included before you commit.

Profound and Scrunch: more enterprise-grade, pairing visibility monitoring with auditing and optimisation workflows. Worth a look once AEO becomes a standing line item rather than a quarterly experiment, less so for a first test.

The AI;DR

Elsewhere in the AIverse

• The line between “design tool” and “dev tool” just got blurrier. Canva Code 2.0 lets you vibe-code websites and apps from a plain-language prompt, then edit the result like any other Canva design — starting from a template, HTML, or an existing project. It’s live for all members, including free plans.

• Kimi K3 is Moonshot AI’s most capable model yet, and it’s not a modest jump. It’s even topping Anthropic’s Fable 5 and OpenAI’s GPT-5.6 on Arena’s web design leaderboard, and Moonshot is reportedly raising a round that’d value it at $31.5B. The frontier model conversation isn’t a two-horse US race anymore — Chinese labs now belong in that discussion.

• The World AI Cooperation Organization, a new intergovernmental AI body headquartered in Shanghai, just launched with 29 founding members — China, Russia, Serbia, Belarus, Cuba, Brazil, and a mix of African and Asian nations. Seems like AI governance is quietly splitting into parallel tracks. so if your stack, compliance posture, or vendor list assumes one global standard is coming, it’s worth watching which bloc your markets end up aligned with.

This is Part 1 of a 3-part series on building an AI-powered marketing team. First up, I explore the individual trap of AI and how to build the first shared system.

I was recently checking out Supermetrics’ 2026 Marketing Data Report, and it turns only 6% of marketers say they've fully implemented AI across their team. Meanwhile 75% are stuck in what the report calls "performative AI": tools everywhere, real change nowhere. Adoption isn't the problem, but something else is broken.

What I think it is: AI is inherently individual. It runs on one machine, inside one context window, shaped by one person's prompts and one person's judgment about what "good" looks like.

That's a feature when you're the one using it, and genuinely great for personal productivity. But it's a liability the moment you try to turn it into a team capability, because the value never leaves the laptop it was built on.

You might have seen this play out at work. Someone on your team builds a useful AI workflow — maybe it turns raw sales call notes into a competitive intel brief, or drafts first-pass ad copy from a brief in a fraction of the usual time. It’s good and saves them hours a week.

Yet it lives entirely in their head and their prompt history. Nobody else can use it. When they’re out sick or leave the company, the workflow leaves with them, and everyone else goes back to doing it the slow way.

That’s the 75% stuck in performative AI. Lots of individual wins, zero organisational capability.

Why solo workflows don’t scale on their own

It’s not laziness or hoarding. It’s structural.

A workflow that lives in one person’s head has no shared version of the prompt, no shared version of the underlying data, and no owner responsible for keeping it working when the model updates or the data source changes.

Without those three things, “sharing” a workflow means sending a Slack message with a prompt pasted in, which decays the moment the recipient’s context, tools, or data differ even slightly from the original.

The four-part test for what’s worth centralising

Not every individual workflow deserves to become team infrastructure though.

Building shared systems has real overhead someone has to own it, maintain it, and keep it accurate. Before you invest in that, run the workflow through four questions:

1. Do multiple people need this same thing? If you’re the only one who ever needs this output, keep it personal. Centralising a one-person need is wasted effort.

2. Is the underlying data shared, not personal? A workflow built on your personal notes app doesn’t generalise. One built on your CRM, your shared drive, or your team’s call recordings does.

3. Does consistency across the team actually matter here? Some tasks benefit from everyone doing it their own way. Others, like positioning language, competitive claims, brand voice, actively suffer when five people each have their own version.

4. Is there a maintenance burden worth centralising? If keeping the workflow accurate requires real, ongoing work (updating data sources, adjusting prompts as the model changes), that’s a strong signal it should have one clear owner instead of five people quietly maintaining five broken copies.

If a workflow clears two or more of these, it’s a candidate for becoming shared infrastructure. If it clears zero or one, leave it as a personal tool and move on — not everything needs to scale.

How to actually build the team system

Once you’ve identified a workflow worth centralising, the build is less about the AI and more about the plumbing around it:

• Name an owner. Not a committee — one person accountable for the workflow staying accurate and useful. This is the single biggest predictor of whether your shared AI workflow survives six months.

• Move the data to a shared source. If the workflow pulls from someone’s personal notes, migrate it to the team’s actual system of record first. The AI layer is only as good as what it’s reading from.

• Document the prompt like it’s a process, not a trick. Write down what the workflow does, what inputs it needs, and what a good output looks like. Treat it the way you’d treat any other repeatable process on the team.

• Put it where people already work. A brilliant workflow that requires opening a separate tool nobody remembers to use doesn’t scale. Build it into Slack, your CRM, or wherever the team already spends time.

• Review and re-test quarterly. Models change, data sources change, and a workflow that was accurate this month can quietly drift by November. Someone — that’s the owner — needs to be checking.

What this looks like in practice

Say someone on your team built a solid workflow for turning social listening data into content angles (if you haven’t seen last week’s issue on that exact process, it’s a good pairing with this one). Individually, it’s great. As a team system, it needs: a shared social listening data source everyone can pull from, one prompt template documented and version-controlled, an owner who checks quarterly that the output still matches what content and sales actually need, and a place it lives that isn’t buried in one person’s saved prompts folder.

That’s the difference between a clever trick and a capability. One disappears when the person who built it moves on. The other compounds.

Reply and tell me: what’s the one AI workflow on your team that’s still trapped on somebody’s laptop?

This is the first part of a three-part series. Part 2 will cover ROI, governance, and measurement — is it safe, and is it working. Part 3 covers adoption: getting the team to actually use what you’ve built.

The AI;DR

Elsewhere in the AIverse

• Meta pushes AI into your feed, not just your prompt box. Meta’s Superintelligence Labs shipped three new models: Muse Image, now ranked #3 on Arena’s text-to-image leaderboard, is live inside Meta AI, Instagram Stories, and WhatsApp. Muse Video is teased for a wider release, and Muse Spark 1.1 — a low-cost agentic coding model — landed on Meta’s new Model API. Worth watching: image gen embedded directly into Stories means AI-generated visuals are about to become a native, one-tap option for anyone posting on Meta’s platforms, not just something you export from a separate tool.

• Chinese labs keep closing the gap on price, not just performance. Tencent released Hy3, a smaller open-source model that outperforms its size class, and Meituan fully open-sourced LongCat-2.0 — billed as the first trillion-parameter coding model trained without Nvidia GPUs. The trend holds: Chinese labs are matching frontier US performance at a fraction of the cost, which keeps pushing the price of “good enough” AI toward zero.

• SpaceXAI (formerly xAI) and Cursor’s first joint model just landed. Grok 4.5 is the company’s smartest model yet and its first release since acquiring Cursor. It’s comparable to leading frontier models and excels at coding, agentic tasks, and office work — it plugs directly into Word, PowerPoint, and Excel. Office-suite integration is the detail to note.

Fable 5 has had a strange few weeks. It launched June 9 as the most capable model Anthropic has ever released to the public. On June 12, the US government hit it with an export-control directive over a reported jailbreak technique, and Anthropic pulled access for everyone worldwide.

Fast forward to two days back, when Fable 5 came back online with a new deadline: included for up to 50% of your weekly usage limits on Pro, Max, Team, and select Enterprise plans through July 7. After that, it's usage credits: $10 per million tokens in, $50 per million out. In a sense, it bills you by the word, and it's not cheap.

Is it even worth paying for, or should you stick to Opus for now?

First, what kind of tool is it?

What is a model this expensive for, if you’re a marketer and not a venture-funded code factory?

It’d be a mistake to treat Fable like “Opus, but better.” It’s not built to be a smarter version of the same job, but built for a different kind of job entirely. Give it a huge pile of source material, let it work through the whole thing on its own without you checking in, and it hands back something finished.

It can actually work with a massive amount of information at once — not just skim it, but reason across all of it. And it plans out multi-step work itself, catching its own mistakes along the way, so you’re not babysitting it through each step.

You describe what you want, then leave it alone. The catch: it costs about twice as much as Opus and burns through tokens twice as fast, so it only pays for itself when the job is big enough to need all that.

Three marketing jobs I’d point it at

1. Positioning from raw evidence. Let’s take B2B SaaS companies as an example. Feed it competitor pages, G2 reviews, your own site, and real sales-call transcripts, and let it work in phases. First a teardown of every competitor’s homepage, pricing, and “vs you” attack page: verbatim quotes, links required, no hand-waving. Then an audit of your own messaging the same way. Then a self-critique pass, where it attacks its own draft positioning and kills anything that contradicts your actual reviews.

What lands at the end is a full framework: a positioning statement, message pillars with proof, rewritten homepage copy, objection scripts for sales, and a flagged list of problems that need a business decision, not better copy.

The APAC bonus: point it at your customers’ transcripts and it works from the vocabulary your buyers in specific markets use. It’s the kind of job that eats a strategist’s afternoon across a dozen tabs, done in one pass.

2. A full-site messaging audit from one prompt. Point it at your entire website and ask where the value prop drifts page to page, where the audience silently shifts from founders to enterprise and back, where a case-study page is still wearing a “start free trial” button, where one page’s claim quietly contradicts another’s.

It crawls the lot and returns a single prioritised report: biggest gaps up top, a page-by-page breakdown with the offending copy quoted, and a fix list ordered by likely traffic and conversion impact.

Opus can do a slice; Fable holds the whole site at once, which is the entire point. If your messaging has quietly fractured across landing pages over two years of “quick edits” nobody documented, this is what finds it.

3. Building a real, working thing. Not just a mockup, but a fully functioning, interactive tool from a prompt or two, with the logic, the UI, and deploy-ready code handled in one autonomous run.

The example worth stealing is a “Spotify Wrapped” for your customers: a personalised year-in-review showing how each one used your product, with nudges to get more out of it. But the shape generalises: an interactive ROI calculator for your pricing page, a self-scoring assessment or quiz as a lead magnet, a report generator that turns someone’s inputs into a tailored PDF.

It’s the sort of asset that used to mean a developer, a designer, and a sprint, now sketched and built in an afternoon. You’ll still definitely want a human QA pass before it goes live, but the distance from “idea” to “working” all but collapses.

The tell across all three: if a job would take a sharp person half a day across multiple sources, and the output is something more complex than a doc, that’s a Fable job.

What it’s not for

The drafting, the campaign angles, the ad variants, the email sequences, the rewrite-this-paragraph — the work that actually fills a marketing week — runs fine on the models you already have, and probably runs better. Newer and pricier hasn’t moved the needle on everyday marketing work for a couple of cycles now. Save Fable for the half-day heavy lift, not the Monday-morning email.

The cost underneath the price

Here’s the part underneath all of this. Every Claude model before Fable was just included in your subscription — you paid your monthly fee and used it. Fable breaks that pattern. Once your included usage runs out, you start paying per word, through the API.

And that bill is bigger than it looks. SemiAnalysis — a firm that specializes in figuring out what AI companies actually spend to run these models — has spent years digging into the real cost of serving something like Fable: the chips, the electricity, the data centers, all of it. Their finding is blunt: the price you’re quoted is a business decision layered on top of a much bigger real cost, and running the most powerful models at scale is expensive in a way that doesn’t show up on your bill until the free ride ends.

To be fair, it’s not all bad news — AI in general keeps getting cheaper. Older, less powerful models drop in price by roughly 10x a year as the tech improves. But that’s the cheaper, “good enough” tier. The frontier — Fable-level stuff — sits at the very top, priced close to what it actually costs, because nobody’s covering the difference for you there yet.

Which is why cost matters more for this tool than anything else we’ve covered. That’s fine if you’re a well-funded US startup that’ll never really feel the bill. It’s riskier if you’re a marketer in Jakarta or Manila or Ho Chi Minh City, working in a currency that isn’t the dollar, where $50 per million tokens turns into real money fast.

The rule that’s held all year still holds: don’t build your workflow around something you can’t afford once the discount ends. The win isn’t having access to the best model. It’s knowing exactly which slice of your work is actually worth paying for it.

The AI;DR

Elsewhere in the AIverse

• AI hardware gets a serious new player. Etched came out of stealth with $5B in valuation and $1B in signed contracts for chips built to run AI cheaper than general GPUs, shipping this summer. If this delivers, it’s the biggest lever for bringing prices like Fable’s back down.

• A new AI browser wants to just handle things. Aside, a YC-backed browser, claims #1 on three browser-agent benchmarks and shows itself cancelling unused subscriptions in its launch video. The benchmarks are self-reported, so treat with caution — but browsers are clearly the next agent battleground.

• Gemini now touches your Mac’s actual files. Spark can sort a Downloads folder or turn local invoices into a spreadsheet, with phone-to-Mac task handoff coming soon. It’s beta, US-only, and gated behind $99/month Google AI Ultra — but every major lab is clearly racing toward agents that act on your desktop, not just chat.

Last week, the skies above Cannes became a canvas as drones lit up the French Riviera above the Croisette — giant words and shapes floating in the night air, including, inevitably, the letters “AI”.

At the same time, the beachfront filled with lavish activations all courting the same brand leaders, and creators talked openly about what AI brand deals were now worth. And somewhere between the rosé and the panel sessions, Cannes Lions rolled out a new AI Craft subcategory across its craft Lions: proof that AI now has a permanent seat at the industry’s biggest festival.

The bar for AI creative just got raised — and most brands aren’t ready

Cannes added an AI Craft subcategory across its craft Lions — Design, Digital Craft, Film Craft, Industry Craft and Creative Data, but it’s no surprise that nearly all Grand Prix awards still went to works made by hand.

Apple TV's Design Grand Prix? A new logo built entirely from glass, filmed in-camera with coloured light reflected through it. Jury president Greg Quinton noted pointedly that "what could have been made far quicker … was made by hand." Coinbase's Film Craft Grand Prix, "Your Way Out," recreated a PlayStation-era video-game world on practical sets — printed costumes, low-poly masks, shadowless lighting — under a zero-CGI, zero-AI rule that director Oscar Hudson and production house MJZ treated as the entire point. De'Longhi won Industry Craft with "Tiny Coffee Shops," a painstaking set of handbuilt miniature cafés overseen by Simon Weisse, the model-maker behind The Grand Budapest Hotel.

The new AI Craft category signals something real, though. In 2026 the standard shifted: did AI change the outcome? It honours work where human creativity and AI combine to produce something neither could make alone, with craftsmanship and intent as the entry requirements, not just technical novelty.

The one clear AI winner: Google’s “Project Genie” Digital Craft Grand Prix — illustrates exactly what that bar looks like. Project Genie lets users generate interactive, explorable worlds from a text prompt or a photo. The jury picked it because the AI unlocked a creative possibility that literally didn’t exist before.

That's the distinction hiding inside all the Cannes AI noise. AI didn't lose at Cannes 2026, average creativity did. Scott Galloway put it neatly from the Pivot stage: AI, he said, "takes everyone to the middle" — which is exactly why creativity matters more now, not less. The handmade work won because it was extraordinary, not because it was handmade. And the one AI campaign that won showed what the tool can do when it's treated as a genuine creative unlock rather than a production shortcut.

ChatGPT pitched itself as an ad channel — and the “consideration phase” framing is the most interesting part

This time last year, OpenAI had no ads business and no official interest in Cannes Lions. This year it arrived with a pitch advertisers will find hard to ignore — and, if they’re honest, harder to fully believe.

VP of monetisation Benji Shomair made the case: “this isn’t social; this isn’t display; and this isn’t search.” What ChatGPT offers, he argues, is access to a part of the purchase journey that’s never been properly addressable: the consideration phase, where people are actively researching options, comparing features, and deciding.

The ad unit itself, for now, looks like… an ad unit. But Shomair’s advice to early advertisers is telling: make it conversational, and use questions, because questions are what users are actually asking. That’s not a tweak to your existing creative. It’s a different brief entirely.

The skepticism is warranted. It’s been 19 weeks and the platform is still technically in pilot; launch pricing has already slid from a $60 CPM to as low as $25 in ten weeks. But the underlying behaviour shift is real — if people are genuinely deciding while they’re on ChatGPT, that’s a different kind of attention than social or search — and it deserves a different kind of creative.

AI agents won't kill brands. They'll just make weak ones easier to ignore.

This is where it gets interesting, and where Mark Ritson shared one of the most useful thinking of the week.

At Cannes, Ritson asked his AI agent for a drink recommendation. He got a confident answer — start with a Boulevardier, darker and warmer than a Negroni — and happily followed it down a rabbit hole of Campari-adjacent cocktails. The point wasn’t the drinks. It was the dynamic: a machine making a confident brand recommendation, and a person trusting it.

This triggered an old argument. In 2014, Simonson and Rosen published Absolute Value, predicting that perfect information — reviews, ratings, price comparison — would kill brand loyalty. Once consumers could objectively evaluate anything, brands as proxies for quality would fade. Marketing’s commentariat loved it. The death of the brand was confidently scheduled.

It didn’t happen. Reviews turned out to be gameable. People did not read 14 reviews before buying toothpaste — they bought the brand they’d heard of. Big brands dominated the very review averages meant to bury them.

Scott Galloway has been making a version of this case for years, and the line lands the same way every time: “the era of brand is over.” Agentic AI, the argument goes, is perfect diligence fully automated — no typing, no reading, no thinking. The brand premium evaporates, and the best actual product wins.

But here’s the problem: LLMs are not brand-blind. They’re brand-obsessed. When researchers test what large language models actually recommend, the same pattern keeps falling out — they favour the well-known. The brands that appear most often, most favourably, in the most places across the web are the ones the model has effectively been taught to trust. Fame isn’t a soft asset to an LLM. It’s the data.

A whole industry is already forming around this. “Share of model” is the new “share of voice” — where do you rank when the machine names a brand unprompted? And the arithmetic is more brutal than search. Search gave you ten links and a scroll. The agent gives you one answer, maybe three. You’re the answer, or you’re invisible.

The implication is uncomfortable but clarifying: AI won’t punish strong brands. It’ll punish weak ones — the me-too products that survived on shelf position and media noise. If your brand isn’t famous enough to show up in an LLM’s training data with weight and credibility, the agent is indifferent to you.

Which means the boring, long-game stuff — content, earned media, consistent presence, reviews — is now effectively training data for the next generation of purchase decisions. Brand-building just got a new ROI argument.

The AI;DR

Elsewhere in the AIverse

• ChatGPT is about to stop being just a chatbot. OpenAI is reportedly building its biggest overhaul since launch — a “superapp” that bundles Codex, AI agents, image generation, and third-party tools into one ChatGPT, landing in the coming weeks.

• The image-model leaderboard got a reshuffle. Reve 2.0 ditches the usual prompt-and-pray approach for a code-based layout you can fine-tune detail by detail and the model just knocked Nano Banana 2 down a spot on Arena rankings, making it the No. 2 image model overall.

• Oracle’s earnings are a good news/bad news AI story in one chart. Revenue hit $19B for the quarter and the backlog exploded 363% to $638B — both ahead of expectations. The stock still dropped earlier this month, because Oracle also said it needs another $40B in financing to keep building data centers. Translation: even the companies winning the AI infrastructure race are starting to spook investors with how much the race costs.

No coding required. No new subscriptions. Just a bit of structured thinking, and some prompts you can steal.

1. Your marketing check-in partner aka The Chief of Staff

The problem: You’re making decisions in a vacuum. Campaign performance, messaging pivots, competitive moves, most of this gets processed in your head during a commute, not in a structured way that sharpens your thinking.

What to build: A standing “check-in” prompt you run every Monday morning or even every other day. You paste in your week’s context — what happened, what you shipped, what’s on your mind — and the AI plays thinking partner: pushing back, spotting blind spots, asking the questions a good colleague would.

Sounds like journaling? Not really, this is a structured debrief with something that doesn’t get bored, doesn’t have its own agenda, and likely has access to more marketing case studies than anyone on your team.

The prompt snippet:

You are my senior marketing thinking partner. I’m going to share my week’s context: campaigns running, results I’m seeing, decisions I’m sitting with. Your job is not to validate me. Push back where my reasoning is thin. Ask clarifying questions. Flag assumptions I haven’t tested. Here’s what’s on my mind this week: [paste context]

Where to run it: Claude works well here — use a Project so it retains context week over week.

2. Your search content assistant aka The SEO Strategist

The problem: SEO content you wrote 18 months ago is losing ground — not to competitors, but to AI-generated answers. Google’s AI Overviews, Perplexity, ChatGPT search: they’re all pulling from sources that are structured to be cited, not just ranked. Most marketers haven’t updated their content strategy for this yet.

What to build: A prompt workflow that takes any topic you’re writing about and produces content optimised to be cited by AI search, not just found by Google. The key shift: AEO content answers questions directly and completely, with clear structure, before it tries to persuade anyone.

The prompt snippet:

I’m writing a piece about [topic] for B2B marketers. Before I write the persuasive version, help me build the AEO layer: What are the 5 most common questions someone searching this topic would ask? For each, write a direct, complete answer in 2–3 sentences — the kind an AI search engine would pull as a cited snippet. Be specific. Avoid fluff.

Bonus move: Run this before you write anything, not after. Use it to structure your H2s. Your SEO and AEO strategy then live in the same piece.

3. Your repurposing engine aka The Content Multiplier

The problem: You write a solid article. It gets published. Maybe you share it once on LinkedIn. Then it disappears into the archive while you start the whole production cycle again from scratch. Most marketers are chronically under-distributing content they’ve already done the hard work on.

What to build: A prompt that takes any long-form piece and systematically breaks it into distribution-ready formats — without it sounding like a robot summarised your work.

The key is giving the AI your voice constraints upfront, not just the content.

The prompt snippet:

Here’s an article I’ve written: [paste article]. I need you to repurpose it into 5 formats: (1) A LinkedIn post — first-person, punchy opener, no hashtag spam, ends with a question or provocation. (2) An email intro paragraph — warm, direct, makes the reader want to click through. (3) A carousel outline — 6 slides max, each slide as a headline + one supporting line. (4) A short-form video script — 60 seconds, hook in the first 5 words, no filler. (5) Three standalone social captions — each one works without reading the article. Match this tone: [paste 2–3 sentences of your own writing as a reference].

Why the tone reference matters: Without it, you get competent-but-generic. With it, you get something closer to your actual voice, which means less editing time on the back end.

4. Your brief generator aka The Brief Writer

The problem: Bad briefs are a tax on everyone: your agency, your freelancers, your designers, your future self. Most marketing briefs are either a vague one-pager or a 20-slide deck that nobody reads. The output suffers, and the blame gets misplaced.

What to build: A prompt workflow that turns a messy brain dump into a clean, structured brief. You answer five questions conversationally; the AI structures it into something an external partner can actually work from.

The prompt snippet:

I need to create a brief for [campaign / piece of content / creative asset]. Help me build it by asking me these five questions one at a time, then compiling my answers into a structured brief at the end: (1) What is this, and what’s it trying to achieve? (2) Who is the audience, specifically, not generically? (3) What do we want them to think, feel, or do after seeing it? (4) What are the hard constraints — format, timeline, budget, brand rules? (5) What does ‘great’ look like here — is there a reference or benchmark? Once I’ve answered all five, write the brief.

The move here: The conversational Q&A format forces you to actually think it through, not just paste in a strategy doc and hope for the best. The brief becomes a byproduct of the thinking, not a chore on top of it.

None of these are actually magic

A quick caveat: the quality of output is almost entirely a function of what you put in. A lazy context dump gets a lazy response. But if you treat these as actual workflows: consistent format, a weekly rhythm, and real input, they will compound quickly.

The AI;DR

Elsewhere in the AIverse

• ElevenLabs can now dub your videos without killing the vibe. Dubbing v2 translates across 90+ languages while preserving the original speaker’s tone and delivery — not just the words. For marketers producing video content across APAC markets, this is the “finally” moment. Localisation that actually sounds human is no longer a six-figure production problem. Try it or watch it in action.

• Anthropic shipped Opus 4.8 and the valuation is now almost a trillion dollars. The model leads on coding and knowledge work, but the more interesting update is how it handles honesty: it’s been trained to avoid making confident claims when the evidence is thin. That’s a meaningful fix for anyone using AI in research or content workflows where hallucination risk is real. Also ships with a Fast mode for cost-sensitive use cases. Meanwhile, Anthropic closed $65B in funding at a $965B valuation — because apparently that’s just a number now. Learn more.

• Trajectory just came out of stealth with $15M and a thesis called Continual Learning — using real-world product usage to continuously retrain agentic models post-deployment. The team includes researchers from DeepMind, OpenAI, Apple, and Meta. For marketers, this is the direction AI agents are heading — systems that get better the more you use them, rather than static tools you prompt around.

This is the second installment of Tested — a series where I put AI tools through their paces in real marketing workflows. First up was Perplexity. This time: Canva AI 2.0.

Last month, Canva held its annual Canva Create 2026 event in Los Angeles and announced what it’s calling its “most significant evolution” since the company launched in 2013.

The pitch: Canva AI 2.0 isn’t just a design tool with AI bolted on. It’s a “conversational, agentic platform” — one that can interpret your intent, coordinate its own tools, and take an idea all the way to a finished campaign without you doing the template-hunting, layout-nudging, copy-tweaking dance.

Before I get into it, a bit of context. I’m a marketer now, but I actually had a (very short) freelance design career for about a year before I made the switch. I’d like to think I come at this with an eye for what good design looks like. And what’s clear to me, watching Canva AI 2.0 land, is that design — like many other areas of skilled work — is changing in ways that can’t be ignored. Businesses, marketers, and designers are all going to need to evolve. The question is how fast, and on whose terms.

First, a quick orientation on the headline features of Canva AI 2.0:

Conversational design: Generate fully structured, editable designs from natural language prompts, with context maintained across iterations.

Agentic orchestration: AI that interprets your intent and coordinates Canva’s tools automatically; give it a brief, get a full campaign across formats.

Living memory: Persistent knowledge of your brand preferences, guidelines, and past work that adapts outputs over time.

Object-based intelligence: Granular edits to individual elements without disrupting the whole design.

Connectors: Integrations with Slack, Notion, HubSpot, Gmail, Google Drive, and more.

Web research: Real-time research pulled directly into design workflows.

That’s… a lot of surface area, so I’m focusing on three things most relevant to my actual workflow.

Use case 1: Conversational design

The task: Build an Instagram carousel for How Marketers AI — topic, layout, copy — without touching a single template.

This is the flagship 2.0 feature. Instead of browsing templates and customising, you describe what you want and Canva generates a fully structured design.

I typed: “Instagram carousel for a B2B marketing newsletter called How Marketers AI. Topic: four AI tools worth trying in 2026. Tone: direct, no hype. Five slides including a cover and a CTA.”

What came back in about 30 seconds was a five-slide carousel with layout, placeholder copy, colour blocking, and stock graphics. It looks fine, but it wasn’t quite the style I was looking for, so I made some iterations.

I asked it use the general colour palate for my newsletter and to change to a sans serif font that I like to use called Poppins. The output is more in line with my initial brand guidelines but I thought it looked rather dull — nothing a few more prompts couldn’t fix. 2.0 held the thread well throughout the chat, and the context-persistence across edits is genuinely new behaviour, while old Canva AI would just generate and forget.

Verdict: The time saving is real, and conversational design gets you 70% of the way there faster than templates. But the quality ceiling matters, and the other 30% is still on you. Think of it as skipping the blank page and the template hunt — not skipping the design judgment.

Use case 2: Agentic orchestration

The task: Give Canva a single campaign brief and see what it produces across formats. For this task, I requested assets for a fictional fintech company called PayFlow, with this prompt:

Product: PayFlow — a B2B payment automation platform for SMEs in Southeast Asia
Positioning: PayFlow eliminates the manual work behind business payments — reconciliation, approvals, multi-currency transfers — so finance teams can close the month in hours, not days.
Audience: Finance managers and CFOs at mid-sized SMEs in Singapore, Malaysia, and Indonesia. They're managing cross-border payments manually, tired of chasing approvals over WhatsApp, and sceptical of solutions that promise simplicity but require six months of implementation.
Key message: Your payment stack shouldn't slow you down. PayFlow automates the entire payment workflow — from invoice to reconciliation — so your team stops doing finance admin and starts doing finance.
Tone: Direct, no fluff. Confident without being salesy. Peer-to-peer, not vendor-to-customer.

Outputs needed:
- LinkedIn social post announcing PayFlow's launch
- Email header for a campaign targeting finance managers
- One-pager overview for a sales meeting leave-behind

One of the most ambitious claims in the 2.0 announcement: describe a campaign and the AI coordinates across Canva’s tools to produce outputs. I tested it with a brief for a fictional fintech event — positioning statement, audience, key message — and asked for a social post, an email header, and a one-pager.

The results were mixed in an instructive way. It gave me four different design styles for the social post and asked me to give feedback on which I liked best before creating the rest of the assets. The socials were generally solid, the email header was generic but usable with some work, and the one-pager had good visual hierachy, wasn’t too busy, and had decent copy too.

The coordination was generally good: three different asset formats from one brief, without me switching between templates or rebuilding from scratch each time. The assets felt like a campaign, even when it had to output multiple formats.

Verdict: The whole task took maybe 13 minutes including refinements on the social post. Genuinely useful for first-draft campaign assets and rapid internal ideation, but probably not ready to hand to a client or run as-is. The orchestration works, but the polish doesn’t at this stage without manually editing. That said, it’s so much better than building each asset separately from templates.

Use case 3: Web research in workflow

The task: Generate a research-backed presentation for PayFlow about fintech payment automation trends.

Canva 2.0’s Web Research feature pulls real-time information directly into designs, so I asked it to pull recent stats on SME financial services trend for a deck.

It surfaced five data points with source attributions, planned them out for me and asked if I was happy with the presentation structure before formatting them into the slide.

The sources were real — I checked — and the stats were accurate, but the APAC specificity was limited, as expected. It’s a similar limitation to what I flagged in the Perplexity piece, that AI still seems to have issues with pulling out quality and comprehensive data, so the deck ended reading quite generic.

That said, Canva’s real value is in the design output. It took just a few minutes to pull out stats, do research, and whip up layouts for 10 pages that would honestly take me at least a few hours to design from scratch.

There’s also a lot of value in Canva 2.0’s living memory and brand intelligence feature — which learns your preferences, guidelines, and past work and applies them automatically. You’ll have to upload your brand colours, fonts, and three sample assets first, but it learns with every new design task you prompt.

Verdict: The compounding effect is real but slow. If you produce assets once a month, Living Memory won’t change your life. If you’re in Canva three to four times a week, the time saving stacks. It’s a long-term investment in a faster baseline — not an instant fix.

Verdict: The best surprise of the 2.0 update. Not groundbreaking, but eliminating the tab-switching between research and design genuinely compresses the workflow significantly and will probably save hours on your day. Data pulling and context input is definitely still needed from your end. The living memory is where it gets interesting, the longer you use it the better it gets.

What I’m learning after a day with 2.0

Conversational design kills the template hunt. That 90-minute asset build is now closer to 15. Not because the AI is doing everything, but because you’re no longer spending half your time browsing templates and nudging layouts.

Agentic orchestration oversells the output quality. Multi-format campaign generation from a single brief is genuinely impressive as a workflow, but the outputs still need significant editing. Canva is calling this agentic, but I’d call it a very good first draft machine.

Web research is the sleeper feature. Nobody’s talking about this one as much as agentic orchestration, but for marketers who regularly build decks or pitches with supporting data, eliminating the research-to-design context switch is quietly valuable. That said, always check the data and research output before pushing anything out.

Living memory gets better with use. Set it up, use it consistently, and your brand baseline improves over time. It’s not magic — it’s compounding.

It’s still Canva’s ecosystem. Every 2.0 feature is powerful inside the platform. The moment you need outputs that live somewhere else: editable source files, assets that need to go through a separate production workflow, the magic stops at the export button.

The verdict

Canva AI 2.0 is a meaningful upgrade. The features I tried generally all deliver real workflow compression for marketers who are already in Canva regularly.

If you’re already on Canva Pro, update your brand guidelines in the living memory today. The investment is less than 10 minutes and the returns are compounding. If you’re not on Pro yet, Canva AI 2.0 is the strongest argument yet for upgrading, but only if design is a regular part of your work, not an occasional one.

Next in the Tested series: TBD. What AI tool do you want me to put through its paces? Reply and let me know.

The AI;DR

Elsewhere in the AIverse

• Google goes all-in on Gemini as your AI operating system. At I/O 2026, Google unveiled a wave of Gemini updates that signal where it’s headed. Spark is a 24/7 personal agent that proactively handles tasks across Google Workspace — with local computer access coming this summer. Omni generates cinematic video from any input. There’s also a new macOS app, a Daily Brief agent, and a redesigned interface, all running on Gemini 3.5.

• Creatify Agent ships finished ads from a URL. Paste a product URL, and Creatify’s new agent handles the entire ad production loop — competitor research, video and image generation, and self-checking for quality. No brief, no back-and-forth, no production queue. If you’re running paid social and spending hours briefing a creative team for straightforward performance ads, this is worth a look.

• Reactor wants to be the infrastructure layer for AI-generated worlds. The AI lab just launched a platform that lets anyone explore real-time AI-generated environments directly in the browser. They’re positioning themselves not as a consumer product but as the delivery infrastructure for real-time AI content. If real-time world generation becomes a content format — think interactive brand experiences, immersive product demos — the infrastructure layer is where the leverage sits.

Meme of the week

I was recently in a conversation with a group of tech marketing leaders. The discussion was supposed to be about AI strategy, but it quickly became something more uncomfortable: a candid reckoning with what’s happening to marketing talent — and what marketing teams actually need to do about it

The split is already happening

One thing that came through clearly: marketing teams are at wildly different stages of the AI journey — from basic LLM prompting to agentic workflows with embedded brand voice. And the gap between early adopters and everyone else is widening fast.

That matches what I see more broadly. 88% of marketers are using AI tools in their daily work, but only 21% of B2B leaders feel confident they’re using it effectively — and the #1 barrier is lack of in-house skills, not budget.

Sounds like we’ve got near-universal adoption with near-zero confidence.

The people I spoke to put it more directly: AI is a paintbrush, but the painter still needs to learn how to use it. Most teams haven’t done that learning yet. And the ones who have are pulling ahead.

So… what’s happening to the role of the marketer?

Fundamental roles are being questioned. Will there still be designers and content marketers in five years? Graphic designers and copywriters are already being asked to prompt and train the models that may eventually replace them. Obviously, people aren’t exactly motivated to do that.

It’s a real culture problem. You can’t build AI fluency across a team if people feel like they’re building the thing that’ll end their job.

As AI takes on more execution, humans are expected to focus on strategy, direction, and outcomes — while intelligent systems handle activation, testing, and optimisation. On paper, that’s a promotion, but in practice, it means the execution skills many of us spent years building: writing copy, pulling reports, setting up campaigns, are being commoditised faster than we can replace them with something better.

The consensus from that conversation: the most valuable hire right now is someone who can adapt to change. “Multi-threaded” generalists and T-shaped specialists who can act as orchestrators of AI agents. Not someone who’s great at one thing someone who can direct systems across multiple functions.

The winning profile is depth in something real, combined with enough breadth to work across the AI toolchain.

The APAC opportunity for marketers

Skills shortages affect 77% of APAC employers, with sales and marketing ranking as one of the hardest categories to fill. We’re already in a scarcity environment. Add AI disruption and it’s not just a skills gap — it’s a skills chasm.

But something interesting came up in the conversation: localisation and market expansion emerged as one of AI’s clearest advantages in the region. Campaigns can be adapted and scaled across markets faster than ever. That’s a genuine unlock for APAC roles where you’re often running across five languages, ten regulatory environments, and wildly different consumer behaviours simultaneously.

That said, the marketer who’s genuinely valuable in APAC isn’t just AI-fluent in general. They’re AI-fluent with regional judgment.

The thing AI still can’t do

One common consensus is that AI isn’t quite ready to own branding just yet. Human insight, voice of customer, and taste remain the necessary foundation, and cutting agencies in favour of AI-generated branding isn’t yet sustainable at scale.

There’s also a concern worth naming: over-reliance on AI risks atrophying the craft skills that make great marketing possible. Critical thinking. Creative instinct. The ability to read a room. If junior marketers skip the learning phase because AI can do the output, they never build the judgment that makes the output good.

Marketers with strong strategic thinking skills now command 34% higher salaries than tactical specialists. That gap is new. The market is already pricing in what AI is replacing.

What adapters might look like

I want to resist painting a neat before/after, because the reality is messier. But some patterns are clear:

They’ve changed what they spend time on. Not just added AI to an existing workflow, but actually cut things. Moved from producing content to directing it. From executing campaigns to designing the logic behind them.

They’ve developed a POV on quality. Anyone can generate copy with AI. Fewer people can tell immediately when it’s wrong: when the tone is off, the nuance is missing, the brand voice has slipped. That editorial judgment is a skill, and it compounds.

They’ve stopped waiting for their company to train them. This is perhaps the most important one. Only 25% of workers receive formal AI training from their employers: even as skills in AI-exposed roles are evolving 66% faster than other jobs. The adapters aren’t waiting for a curriculum. They’re running experiments and building their own framework.

The non-adapters, by contrast, share one trait: they’re treating AI as a productivity tool rather than a structural shift. Using it to do the same job faster, not to do a fundamentally different job.

The new job description

The marketer’s role is now evolving toward orchestration. Setting the creative and strategic direction that AI executes. And that means the whole team needs to follow, because you can’t orchestrate a system you don’t understand.

If you ask me, the priority right now is building AI fluency across teams. Fluency means everyone in the function understands enough to work with AI intelligently, not just alongside it.

AI amplifies great marketers. It doesn’t replace the need for taste, judgment, and human insight. But it does raise the floor. Average execution is being automated. What’s left, and what’s actually valued, is the stuff that was always hard to teach.

The AI;DR

Elsewhere in the AIverse

• OpenAI puts its coding agent in your pocket. Codex — OpenAI’s agentic coding tool — is now built directly into the ChatGPT mobile app, syncing across devices so you can kick off a task and walk away. The so what: for marketers building or managing AI workflows, the barrier to running agentic tasks just dropped again.

• Tavus launched Image-to-Replica. Drop in a still image, brand mascot, or illustration, and it becomes an interactive digital human that watches, listens, and responds in real time. The creative bar for “good enough” video is moving fast — which is either exciting or terrifying depending on how much your brand relies on authentic human presence.

• Ineffable Intelligence just landed an engineering partnership with Nvidia. Hot off a $1.1B seed round at a $5.1B valuation. The startup founded by DeepMind alum David Silver is building AI that learns through trial and error rather than human-written training data.

About a year ago, I called AI search “the new marketing battleground.” I was right about the battlefield, just wrong about the timeline.

When I wrote that piece in June 2025, AI Overviews were still that weird little box at the top of Google that people kind of trusted… but not completely. GEO was a concept marketers had bookmarked, nodded at, and then done nothing about.

I recently watched Robby Stein — Google’s VP of Product for Search — on the Superhuman AI podcast, and it really got me thinking about where search is going. So this is my attempt to update the original piece with what Google has actually shipped, and what Stein let slip about where it’s heading.

What I got right, and didn’t

The shift from keyword optimisation to answer engine optimisation is real. Zero-click is real. Brands that structure content for machine readability are winning. GEO is now something SEO agencies actually pitch with a straight face.

What I underestimated? The sheer speed of Google’s moves. When I wrote this in mid-2025, AI Overviews were the main event. By November, Google had launched Gemini 3 Pro. By December, AI Mode was delivering faster, smarter answers — tackling complex questions, comparisons, and planning with stronger reasoning Search Engine Land globally. By January 2026, Personal Intelligence for AI Mode went public, connecting Gmail and Google Photos to deliver context-aware search responses.

The Robby Stein reframe: search isn’t shrinking, it’s expanding

The most counterintuitive thing Stein said is that AI isn’t cannibalising search. It’s making people search more.

Google is seeing more searches than ever as people ask harder, more conversational, and more visual questions powered by AI. AI hasn’t really changed the foundational needs: there’s just more curiosity being fulfilled now that AI makes it possible.

This is a meaningful reframe for marketers. We’ve been operating from a scarcity mindset — fewer clicks, less traffic, zero-click apocalypse — when the actual picture is closer to: the pie is getting bigger, but the slices are being cut differently.

What’s changing is the kind of query. Simple lookups still go straight to 10 blue links. But Stein described how people are increasingly asking harder, longer questions in natural language — and once you have a really specific question, AI Mode triggers with an overview and a way to dive into a back-and-forth conversation Limitless Podcast. You can now have 10, 20 turns with Google before landing on an answer.

For B2B marketers, this matters. Complex buying decisions — “which enterprise SaaS solution fits a hybrid team across Singapore, Manila, and Jakarta” — are exactly the kind of multi-variable queries where AI Mode thrives. Your content either shows up in that reasoning chain or it doesn’t.

Reasoning-aware search: a new approach to AI search

Here’s something I didn’t fully appreciate in my original piece: AI Mode doesn’t just retrieve information. It reasons. Stein described how Gemini-powered models issue their own internal Google searches behind the scenes — multiple simultaneous queries — to construct a single response.

And with Gemini 3 in the picture, the leap is significant. Gemini 3 in AI Mode can dynamically create the ideal visual layout for responses on the fly — featuring interactive tools and simulations — tailored to the specific query. When the model detects that an interactive tool will help, it uses its generative capabilities to code a custom simulation in real-time and adds it into the response.

Ask about physics? You get a live simulation. Ask about a financial comparison? You might get an interactive graph, not a table.

On the model side: Gemini 3 Flash is the default free model powering AI Mode globally, while users in the US can toggle “Thinking with 3 Pro” for complex, multi-step reasoning tasks — at slower speeds but deeper analysis. ALM Corp Think of it as economy versus business class reasoning, now built into the search bar.

Agentic search: “Done, I’ve booked it for you”

This is the part that should make every marketer pause.

In the podcast, Stein used a restaurant analogy that stuck with me. You’re looking for a date night spot in San Francisco. You get your AI-assisted shortlist. But then — Google has now implemented a live way for an agent to browse OpenTable and the web for reservations, then bring back not just great restaurants with reasoning and analysis, but also the specific times tables are available.

This is what agentic search actually means. Google is moving beyond providing links to executing complex, multi-step tasks by deeply understanding personal context. The “consideration phase” of buying — which is where B2B marketers spend most of their budget — is now being navigated by an AI agent on behalf of a human.

The implication for demand gen and content marketing is uncomfortable but worth sitting with: if an AI agent is doing the shortlisting, you need to be in the AI’s knowledge base, not just on page one of search results. Different problem. Same urgency.

Personalised search and the privacy question

The newest development, and the one with the most long-term consequences, is Personal Intelligence for AI Mode, which connects users’ Gmail accounts and Google Photos to let search reference personal emails, travel bookings, photo memories, and transaction histories when generating responses.

But the architecture it represents is the trajectory: search that knows your context, not just the world’s.

For B2B marketers in APAC, the honest answer is: this isn’t your immediate problem. But watch how it plays out. The brands that win in a personalised-search world are the ones already building trust signals into their content — being cited by AI as a credible source, not just ranking for keywords.

What this means for you now

My original piece was mostly about the shift from SEO to GEO. That still stands. But I’d add three updates:

1. Think about the reasoning chain, not just the citation. AI Mode constructs answers by reasoning across multiple sources. Your content needs to be usable in an argument, not just discoverable. That means being specific, factual, structured, and opinionated — not fluffy.

2. Agentic search is already here for transactional queries. If you’re in hospitality, travel, financial services, or retail, your booking/conversion funnel is being touched by AI agents right now. Figure out what that looks like for your category before your competitors do.

3. Visual search is growing faster than most APAC marketers realise. Stein mentioned a 70% year-over-year increase in visual searching on Google. Lens, circle-to-search, screenshot queries — this is an APAC behaviour pattern (anyone who’s watched how people shop in Southeast Asia on their phones knows this). If your visual assets aren’t optimised for machine readability, you’re invisible to this mode entirely.

The AI;DR

Elsewhere in the AIverse

• Gemini can now create files on Microsoft 365 and Workspace. No more copy-pasting AI output into a doc like it’s 2022. Gemini now generates Docs, Sheets, Slides, Word, Excel, PDFs, and more directly in chat. For marketers churning out reports, briefs, and decks, this is a legitimate time-saver.

• Twitter’s former CEO is building the search layer for AI agents — and investors are paying attention. Parallel Web Systems, founded by Parag Agrawal, just hit a $2B valuation after raising $100M — nearly tripling its valuation in five months. The startup builds internet and research APIs specifically designed for AI agents to search, extract, and monitor the web.

  Vibe coding with Lovable just went mobile. Lovable — the no-code app builder that lets you spin up websites and apps from a text prompt — is now on iOS and Android. Kick off a build from your phone, let the agent run, get a notification when it’s ready.

👋 Quick-catchup: In Part 1, I ran a visibility test: four prompts asking AI to recommend the best APAC B2B marketers, and my name came up all of zero times. That result became the brief for a personal branding agent built on Claude Code: a workflow that takes raw reflection and routes it through a structured identity layer before generating anything.

If you missed Part 1, read it here first. If you’re caught up — here’s the full build.

🗂️ The workflow: layer by layer

The agent runs across four layers. Think of them less as sequential steps and more as nested dependencies, and each layer only works because the one before it exists. I’ll tell you what’s built, what’s still in progress, and what’s planned.

🧬 Layer 1: The Identity Foundation

Everything starts here. Before the agent generates a single word, it reads a file called brand-dna.md — a persistent context file that makes every output yours rather than generic.

I kicked it off with this prompt in Claude Code:

Create a file called brand-dna.md with the following sections:
- Core belief (the one thing I want to be known for e.g. APAC B2B marketing leader in fintech, tech, SaaS)
- Contrarian positions (what I believe that most in my field don't)
- Niche language (specific terms my audience uses)
- Content pillars (topics I cover and why)
- Voice notes (what I sound like vs what I don't)
- Anti-patterns (topics, tones, formats to avoid)

Claude Code scaffolded the file. I filled it in manually, which is important — you don’t want AI inventing your positions and skills for you.

Pro tip: you won’t want to skip this step, even if takes some time to think about. Most people have a vague sense of their personal brand but have never written them down as explicit, falsifiable statements.

“I believe AI is changing marketing” is not a position. The agent is only as opinionated as what you give it.

The file never gets auto-updated. You edit it yourself when your thinking evolves. That’s a feature, not a limitation, since your identity wouldn’t and shouldn’t shift every time you have a new idea.

📥 Layer 2: Story Capture

The input layer. One prompt, no polish required:

What happened this week? What did you notice, hear, experience, 
or think about that you haven't seen anyone else write about yet?

Anything you experienced at work or saw on LinkedIn. Stream of consciousness is fine. The agent takes it from there.

After you submit, it does four things:

• Classifies your input into one of four types: Observation (pattern you noticed), Experience (something that happened to you), Reaction (response to something external), Contrarian Take (position against prevailing wisdom in your field)

• Writes a one-line summary

• Flags which positions in brand-dna.md the story connects to

• Flags if it doesn’t connect to anything: either a new position worth adding, or content that’s off-brand

Here’s the prompt I used to build this layer:

Build a simple CLI input that captures free-text from me. After I submit, 
classify it into one of four types: Observation, Experience, Reaction, 
or Contrarian Take. Output the classification and a one-line summary. 
Read brand-dna.md before classifying — the classification should be 
informed by my stated positions.

Classification sounds like a small thing, but it really isn’t. Reactions are timely with have a shelf life. Observations are evergreen and they compound. Knowing which you’re working with changes how you write the output and when you publish it, and getting that call right early saves a lot of second-guessing later.

✍️ Layer 3: Output Generation

One story, four assets. From a single classified input, the agent generates:

• LinkedIn post draft — under 150 words, first-person, no filler openers, hashtags optional

• Newsletter angle — one paragraph on how this becomes a full issue, including which content pillar it fits.

• Content hook — one punchy line that could open a carousel, Reel, or cold LinkedIn post

• Follow-up prompt — one question to sit with before publishing anything

The prompt structure:

Given the classified story and brand-dna.md, generate the four outputs below:

LinkedIn post draft — under 150 words, first-person, no filler openers, add hashtags

Newsletter angle — one paragraph on how this becomes a full issue, including which content pillar it fits.

Content hook — one punchy line that could open a carousel, Reel, or cold LinkedIn post

Follow-up prompt — one question to sit with before publishing anything

Constraints: Every output must reflect at least one position from brand-dna.md. 
If it doesn't, flag it and explain why. Do not soften my positions to make 
the content more palatable. Do not use "I'm excited to share" or any variant. 
Add hashtags for LinkedIn.

That flag constraint is the most useful thing in the whole system. In testing, about a third of first-draft outputs get flagged as not reflecting any stated position, which is the agent telling you the story you captured wasn’t specific enough to carry your identity. Better to know before publishing than after, amiright?

What I’m still finishing: a push-back function. If a generated output actively contradicts something in brand-dna.md — not just ignores it, but goes against it — the agent argues back rather than just flagging. It’d work when the contradiction is explicit with subtler inconsistencies, but it’s not reliable enough yet.

🔁 Layer 4: The Visibility Feedback Loop

The slowest layer — and the most important one to get right.

On a weekly or fortnightly schedule, the agent reruns the four visibility test prompts from Part 1 and logs the results to a file. Over time you’re tracking whether:

• Your name starts appearing in AI-generated expert lists

• The language AI uses to describe your niche matches your own niche language

• Your stated positions are being reflected back in any form

I’ll be real, though. You won’t see movement in a week. Probably not in a month. But it’s the right signal to track if the goal is AI recognisability, not just content volume or follower count.

Technically, this is the simplest layer to build. The harder part is deciding what “progress” looks like in a way that’s honest rather than self-flattering. A name appearing once in one tool is good, but not exactly a signal. Your name appearing consistently across tools, in response to different prompts, over time, that’s what you’re looking for.

🗺️ The Full Flow

You capture a raw story or thought
          ↓
Agent reads brand-dna.md
          ↓
Agent classifies the story
(type + brand alignment check)
          ↓
Agent generates output bundle
(post + newsletter angle + hook + follow-up prompt)
          ↓
Agent scores outputs against brand-dna.md
(flags anything generic or off-position)
          ↓
You edit, publish, or discard
          ↓
[On schedule] Visibility test reruns
Results logged for tracking over time

3 things I didn’t expect

Yup, writing my identity file was the hardest part. Writing clear, falsifiable positions forces a clarity that most personal branding exercises actively avoid. Vague positioning feels safer, but your agent won’t work with vague, not well at least. If you take nothing else from this series, take that.

Classification is a more useful editorial instinct than I expected. When the agent tells you your story is a Reaction rather than an Observation, it shifts how you write the output and when you publish it. I’ve started applying the same classification mentally even when I’m not using the agent. That’s a sign the framework has some legs beyond the tool itself.

Generic outputs are information, not failure. When the agent flags a draft as not reflecting any stated position, most of the time it’s right. The story wasn’t specific enough to carry your identity. The flag is the feature.

Where this lands

Thanks for following along! This started as a build log and ended up being as much about the thinking behind personal branding as the technology inside the agent.

My build isn’t finished, but the core of the framework works. Capture, classify, generate, score. 10 minutes of raw input, a usable draft on the other side, calibrated to positions I’ve actually stated rather than generic best practices. That gap I talked about in Part 1 — between the thinking and the publishing — this closes it. Not perfectly, but consistently enough to matter.

Want to create one of your own?

You don’t need Claude Code to begin — not exactly. The most valuable part of this entire system is the one that requires no technology at all.

Open a blank document right now and answer these:

• What’s the one thing you want to be known for in your field?

• What do you believe that most people in your industry don’t?

• What topics, tones, or takes do you actively want to avoid?

That’s your brand-dna.md. Even if you never build the agent, having those answers written down and visible changes how you write, what you publish, and what you say no to. Everything else in this framework is just automation built on top of that clarity.

The AI;DR

Elsewhere in the AIverse

• Poke raises $10M to build AI agents for your personal life. Text it to plan your schedule, track fitness, edit photos, manage your day. Fresh funding at a $300M valuation suggests there’s a real market for the personal side of agentic AI.

• Meta introduces Muse Spark from its new Superintelligence Labs. The first in the Muse family of models, built as a natively multimodal reasoning model with tool-use, visual chain of thought, and multi-agent orchestration.

• Claude now lets anyone build and deploy custom agents. Claude Managed Agents launched in public beta this week. Define your agent’s tasks, tools, and guardrails; Claude handles the production infrastructure.

It's a sharp framework. And it got me thinking about what AI fluency looks like on the ground in the day-to-day reality of how marketing teams adopt and build with AI.

Because the gap between “I use AI” and “I build with AI” is enormous. And most teams are still on the wrong side of it.

After watching marketers across the industry grapple with this, I’ve started to see a clear pattern. When it comes to using AI for marketing, there are four distinct levels, each one requiring a stronger skill set, a different mindset, and a different kind of patience.

You can’t skip the climb. But trust me, once you see the steps, it gets a lot less daunting!

🪜 Level 1: Reliable Prompting

What it looks like: Your team uses ChatGPT or Gemini like a slightly smarter Google. Results are inconsistent. There’s no shared practice. Some people swear by it; others quietly gave up after a few bad outputs and went back to writing everything themselves.

What it actually is: The foundation. And most people rush through it without really building it.

Level 1 isn’t just “using AI.” It’s learning how to get AI to give you useful, consistent outputs eight or nine times out of ten. That’s harder than it sounds — especially for marketers who are used to tools that behave predictably. AI doesn’t work like that. It’s non-deterministic, which is a fancy way of saying: same input, different output, every single time.

Which means the right question isn’t “is this output correct?” It’s “did this output make my thinking better?”

The way to get better at this level isn’t to try more tools. It’s to take one task — rewriting a campaign brief, drafting a nurture email, summarising a competitor’s positioning — and iterate on it inside a single conversation until the output is dramatically better than where you started. Add context. Push back on bad answers. Figure out that the AI responds differently depending on how you frame the problem.

What good looks like at Level 1:

• You’ve stopped asking “is this right?” and started asking “did this make my thinking better?”

• You know how to layer in context — brand tone, audience details, past examples — to meaningfully change the output

• You don’t quit after a bad first response (the marketing equivalent of not scrapping a campaign because the first concept missed)

The self-check: If your team’s AI usage is mostly one-off prompts with no shared practices, you’re here. That’s fine. Own it. Everyone starts here.

🔧 Level 2: Custom Mini-Apps

What it looks like: Someone on the team has built a custom GPT, a Claude Project, or a Gemini Gem that others actually use. There are emerging patterns — “this is how we handle X.”

What it actually is: The most underrated level in the stack. Also the one where the real paradigm shift happens.

If Level 1 is learning to fish, Level 2 is building a rod that casts itself.

At some point, you’ll notice you’re typing the same instructions over and over. Same brand context. Same output format. Same three caveats about your audience every single time. That’s your cue. Stop copy-pasting yourself and save those instructions into something reusable.

Custom GPTs, Claude Projects, Gemini Gems — these aren’t just productivity hacks. They’re mini-apps with your thinking baked in, which means every conversation starts from a higher baseline. Every person on your team who uses that tool gets the benefit of the work you put in once.

It’s tempting to skip this and jump straight to building dashboards and interfaces. Don’t. The real goal of Level 2 isn’t the tool you build — it’s the systems thinking you develop in the process. That’s the muscle that makes everything else work.

What good looks like at Level 2:

• You’ve built at least one reusable tool that solves a specific, repeatable marketing problem

• Other people on your team use it — not just you

• “How we do X with AI” is something your team can actually articulate

The self-check: If your team has built custom tools that others actually use, and patterns are starting to emerge around specific workflows, you’re here.

⚙️ Level 3: Vibe-Coded Interfaces and Automated Workflows

What it looks like: Someone on your team — not necessarily an engineer — has built an actual tool. A dashboard. An internal app. An automated workflow that handles a task end-to-end without anyone babysitting it.

What it actually is: Systems thinking with an interface on top.

This is the level that makes people think they need a computer science degree. They don’t.

“Vibe coding” is what happens when you describe what you want and let AI write the code — using tools like Replit, Lovable, Claude Code, or Cursor. The skill isn’t writing code line by line. It’s knowing how to decompose the problem, describe it precisely, and iterate with AI as your debugging partner when things inevitably break.

For marketers, this opens up genuinely exciting territory: a competitor monitoring dashboard that pulls from multiple sources, a campaign brief generator with your full brand playbook baked in, an automation that turns long-form content into a full distribution plan, a lead scoring tool built specifically around your ICP. Not off-the-shelf. Yours.

What good looks like at Level 3:

• Someone on your team (not necessarily a developer) has actually shipped a working tool

• You’re starting to work with APIs, documentation, and developer tools — as a non-engineer

• Building and using software feel like different points on the same spectrum, not different worlds

The self-check: If someone on your team has built something that works — a dashboard, an automation, an internal app — even without an engineering background, you’re here.

🤖 Level 4: Agentic Systems

What it looks like: AI taking actions across multiple tools without a human signing off on every step. Data goes in. Something useful comes out. Nobody had to sit in the middle.

What it actually is: Where everything is heading. Where very few marketing teams actually are yet.

An agentic system is when AI stops being a tool you talk to and starts being a system that works on your behalf. It reads inputs, makes decisions, and pushes outputs — across platforms, across tools, across your stack.

For marketing, the early versions look like: a system that monitors brand mentions and flags anomalies before you’ve had your morning coffee. One that pulls campaign data from five different platforms and drops a clean performance summary into Slack every Friday. One that qualifies inbound leads and routes them to the right sequence without anyone touching them.

The unlock here isn’t programming. It’s decomposition — the ability to break any workflow into pieces small enough that each one can be handled by a single, well-scoped prompt. An agent is really just a series of Level 1 and Level 2 problems linked together. The agent handles the handoffs. You define the logic.

What good looks like at Level 4:

• You have at least one end-to-end workflow running autonomously — data goes in, something useful comes out, no manual steps in between

• Your team’s time is spent defining and refining the logic, not executing it

• You’ve learned to pick problems carefully — high-volume, repeatable, low-stakes-if-wrong — because you know where AI still needs a human backstop

The self-check: If you have automated workflows where AI acts across multiple tools without a human sign-off on every step, welcome to Level 4. There aren’t many of you yet.

Where’s your team?

Here’s a quick self-assessment:

Level 1: Your team uses AI for one-off tasks. Results are inconsistent. No shared practices or standards.

Level 2: At least a few people have built reusable tools that the team actually uses. Patterns are starting to emerge around specific use cases.

Level 3: Someone has built a working tool — a dashboard, an automation, an internal app — even without an engineering background.

Level 4: You have automated workflows where AI acts across multiple tools without a human in the loop for every step.

Most teams are somewhere between Level 1 and Level 2. That is completely normal. The issue isn’t being at Level 1 — it’s thinking you’re at Level 3.

Teams that skip Level 2 miss the systems thinking that makes everything else work. They end up with a pile of disconnected tools, no shared practice, and a growing suspicion that AI hasn’t really delivered on its promise.

So how should you proceed? Pick one workflow. Get specific about what you need it to do. Save those instructions somewhere reusable. Then ask yourself: what’s the next small problem I can hand off to AI? What can I build next?

The AI;DR

Elsewhere in the AIverse

• Both Google and Microsoft both dropped new models this week. Google released Gemma 4 — its most capable open model family, built for reasoning and agentic workflows, in four sizes. Microsoft countered with the MAI models, focused on transcription, voice, and image generation.

• Brett Adcock came out of stealth with Hark. The founder behind Figure and Archer Aviation is building “personal intelligence” — advanced AI paired with next-gen hardware. His verdict on existing chatbots: “incredibly dumb.” Watch the launch video.

• Claude now connects to work tools on mobile. Figma, Canva, Amplitude and more are now accessible inside Claude on your phone — the third Claude release this week, after Computer Use and Auto Mode. Busy week at Anthropic HQ.

Meme of the week

Let me describe a person I know well.

They’re quite good at their job. They have real opinions about their industry. They’ve seen things work and seen things fail. If you got them in a room and asked them about their niche, you might leave with three things to think about.

But online? They’re hard to find on Google. Sporadic LinkedIn posts. Nothing that would tell a recruiter (or an AI) what they stand for.

That person is me, and probably some of you reading this right now.

I could go on in a whole article about how professionals get discovered is shifting (actually yes, a piece on that is coming up) but first, I’ve decided to build a system that gets me visibility, inbounds, and treated as a credible voice in the field — for me, that’s being an AI-charged marketing leader for the tech space.

A personal branding agent — running on Claude Code — that takes raw reflection and routes it through a structured identity layer before generating anything.

💡 The idea

The problem isn’t that marketers like us don’t have things to say. It’s that there’s a gap between the thinking and the publishing — and most people run out of energy in that gap.

You have a sharp observation on Tuesday. By Friday, when you actually have time to write something, it’s gone cold, stuck inside a note app no one else gets to read. So you post something generic instead, or nothing at all.

I wanted to close that gap with a workflow. Not a content scheduler, not an AI ghostwriter — a system that captures raw thinking and routes it through a structured identity before generating anything. Stories and tasks in the same place. Capture once, activate everywhere.

Here’s what a typical session looks like in practice: I open the agent at the end of the week and answer one question: what happened, what did I notice, what am I thinking about that I haven’t seen anyone else write about?

No structure required. I might type three sentences or three paragraphs. The agent reads what I give it, classifies it, checks it against my identity file, and generates a LinkedIn draft, a newsletter angle, a content hook, and a follow-up question.

The whole thing takes about ten minutes. The editing takes longer. But the hardest part — getting from blank page to something real — is already done.

One design principle runs through the whole thing: the agent never invents your identity. It only activates it. Everything it generates flows through positions you’ve already stated you believe. That’s what makes it different from an AI writing tool.

⚙️ Why Claude Code?

Quick context if you haven’t used it: Claude Code is Anthropic’s agentic tool that runs in your terminal. You talk to it in plain English, it writes and executes code, reads and creates files, and chains multi-step tasks together. Think less chatbot, more dev collaborator who lives in your command line. You don’t need to be a developer to use it, but a comfort with tinkering helps.

I chose it over no-code tools for two reasons.

One: I needed control over the identity layer. A persistent context that encodes my specific positions, not just generic brand guidelines. Template-based tools don’t handle opinionated, specific context well and are built for output, not identity.

Two: this workflow isn’t a single output. It’s a chain: capture, classify, generate, score. Claude Code lets me build the chain, not just individual steps.

🪞 Start here: the visibility test

Here’s a two-minute exercise that will either reassure you or ruin your afternoon.

Open Claude, ChatGPT, Perplexity, or whichever tool you like, and run these four prompts:

"Who are the best [your specialty] creators or practitioners?" 

"Can you recommend someone who [does what you do]?" 

"I'm looking for a [your title/role] who specialises in [your niche]. Who should I follow?" 

"What experts should I learn from if I want to understand [your topic area]?"

If your name comes back, great, drop me a message — I’d love to interview you.

If it doesn’t, welcome to the club. Membership is free, the coffee is mediocre, and there are a lot of us.

What’s coming in Part 2?

The full build. How I structured the agent across four layers, the prompts I used in Claude Code, what’s working, what’s half-finished, and what the workflow looks like end to end.

If you want to run this yourself — or just understand how an agentic workflow actually gets built by a non-developer — Part 2 is for you.

In the meantime: run the visibility test. Hit reply and tell me what you find.

The AI;DR

Elsewhere in the AIverse

• OpenArt launches Worlds. Asingle prompt or photo goes in, explorable 3D environment comes out. Drop characters in, move through the space freely, capture production-ready shots without losing scene consistency.

• Google upgrades AI Studio into a full-stack vibe coding platform — generates production-ready apps with multiplayer capabilities, real-world service integrations, and an agent that picks up where you left off. The barrier to building your own AI workflows just got lower again.

• Gamma just overhauled its core product. The headline feature is Gamma Imagine, which generates on-brand posters, logos, and visuals from a prompt. As they approach 100M users, they’ve also rebuilt the template experience so you can update entire decks without breaking the design.

Meme of the week

A few months ago, I wrote a piece for e27 about something that’s been bugging me as an APAC marketer: how Western marketing playbooks, particularly for startups, get copy-pasted into APAC markets without much interrogation — growth loops built for Palo Alto, messaging frameworks designed for English-first audiences, and distribution strategies that assume your customer has a credit card and 5G connection.

That piece wasn’t about AI, but lately I’ve been sitting with an extension of the same thought: what happens when those assumptions get baked into the model itself?

AI marketing tools learn from historical data: who clicked, who converted, who stayed. That data skews Western, urban, and English-literate, because that’s who’s been transacting online the longest. The underserved Asian audiences that APAC marketers are actually trying to reach are largely absent from that training signal.

I think marketers frame it wrong when they call it a bias problem. It’s actually a customer acquisition cost (CAC) problem.

The segments that AI tools tend to deprioritise are often less saturated, faster-growing, and — once you acquire them — stickier. When you’re the first brand to reach someone in their language, on their terms, with a product that actually fits their life, you don’t lose them easily to a competitor whose algorithm never found them either.

Look at something like GCash in the Philippines or GoPay in Indonesia. The dominant narrative around their growth tends to focus on product. But a significant part of what made them work was a willingness to reach customers that existing financial infrastructure and existing marketing logic had written off. The algorithm didn't find those customers. Someone made a deliberate choice to go looking.

Three things worth doing before you trust your tool's outputs

1. Ask where your tool was trained and pay attention to what it builds automatically.

• Which audience segments does it construct automatically? Which ones do you have to build manually? The gaps between those two lists are the assumptions baked into the model.

• Run a simple audit: pull your tool’s recommended audience and map it against your actual customer. If they don’t match, you need to decide how much of your budget you’re willing to spend proving the algorithm wrong before you take back control.

2. Seed your own first-party data early and deliberately.

• The fastest way to correct for training bias is to give your algorithm better signal. That means running intentional campaigns into underserved segments. Not to convert immediately, but to generate behavioural data your tool can actually learn from.

• This might look like: a small always-on budget for Bahasa or Filipino content, even when your primary market is English-speaking. Or a landing page variant built specifically for first-time digital finance users. Or a retargeting pool you build manually from offline events and community touchpoints, rather than letting the algorithm decide who’s worth following up with.

• The startup that builds a proprietary data advantage in a segment their competitors have algorithmically abandoned is the one that wins the next three years.

3. Treat “low-converting segments” as hypotheses, not verdicts.

• When your tool tells you a segment underperforms, the question isn’t “should I cut it?” The question is: why is it underperforming?

• Work through it systematically. Is the creative wrong for that audience, built around visual cues, humour, or social proof that doesn’t translate? Is the landing page in a language they’re not comfortable transacting in? Is the call-to-action built around a behaviour your customer doesn’t have yet, like entering card details when your audience is used to paying via e-wallet or cash on delivery?

• The algorithm can’t distinguish between “this segment won’t convert” and “this segment hasn’t been spoken to correctly.” It just sees low signal and pulls back. Only you can make that call. And the first time you isolate the real reason and fix it, you’ll often find a segment that was never actually underperforming.

In short, the algorithm will always find you a customer. The question is whether it's finding yours.

The AI;DR

Elsewhere in the AIverse

• Adobe launches Photoshop AI assistant. The new assistant, available on Photoshop Web and Mobile, lets you edit images through natural language prompts. Public beta is live now.

• Canva introduces Magic Layers. Flat JPGs and PNGs, including AI-generated images, can now be converted into fully editable, multi-layered designs. Move objects, edit text, change layouts.

• Amazon opens Health AI to everyone. Previously limited to Prime and One Medical members, Amazon’s healthcare assistant is now available to all users on its website and app. One to watch for what it signals about AI moving into high-stakes, high-trust verticals.

• Anthropic upgrades Claude for Excel and PowerPoint. The two tools now share full conversational context across open files — meaning Claude can update a spreadsheet, build a model, and drop the output into a pitch deck in one continuous workflow without losing context. The update also brings Skills to both apps.

Meme of the week

For a while now I’ve been meaning to sit down and write about how and why I use AI tools for marketing. This new series looks at a different AI tool each time and my experimentations, starting with Perplexity. Let’s dive in!

I don’t know about you, but I used to spend at least 10 hours a week doing research.

Competitive intelligence. Customer backgrounds. Market trends. Content topics. Fact-checking dubious LinkedIn stats. It’s a significant chunk of my week, and the workflow is... messy.

Sometimes it’s Google for 15 minutes, ChatGPT for context, back to Google to verify what ChatGPT just made up, LinkedIn stalking, company websites, back to Google. Repeat until I have what I need.

Enter Perplexity.

The promise: A research tool that combines search, AI, and citations in one place. Real-time information (unlike ChatGPT). Actual sources (unlike ChatGPT). No hallucinations (allegedly).

The question is, can it replace your current Google + ChatGPT + manual verification workflow?

So what is Perplexity?

Perplexity is an AI-powered research tool that searches the web and synthesises answers with citations. The free version gives you basic searches, while Pro ($17/month) adds things like unlimited “Pro” searches (more advanced AI model), file uploads and analysis, API access to multiple models (GPT-4, Claude, etc.), and better source diversity.

Basically, it’s Google + ChatGPT + Research Assistant, rolled into one. The hypothesis for my test was that Perplexity can reduce my research time by 40-50% while maintaining quality.

What I’ve been using it for:

• Guest research for this newsletter - Background on people I want to interview. Company info, recent work, LinkedIn activity, why they’d be interesting.

• Topic validation - Is this newsletter idea actually interesting or just something I think is clever?

• Competitive intelligence at my marketing job - What other competitors are doing. Positioning, recent campaigns, how they talk about themselves.

• Market trends - What’s happening in B2B marketing + AI in APAC. Not just “what everyone’s talking about” but “what’s actually shifting.”

• Fact-checking - Verifying the AI marketing stats that get 5k likes but feel made up.

Use case 1: Content ideation for the newsletter

The situation: I needed to figure out if “Why AI makes B2B marketing more human” was actually an interesting angle or just something I thought sounded clever. I also needed 3-4 backup topics in case my first choice fell flat.

What I asked Perplexity: “What are B2B marketers in APAC struggling with regarding AI in 2026? Recent pain points, trending discussions, emerging challenges”

What I got was a mixed bag. Perplexity surfaced several real pain points: ”proving ROI on AI tools,” “maintaining brand voice with AI-generated content,” “team adoption challenges” — all legitimate topics people are talking about. It pulled from recent marketing publications, LinkedIn trending topics, and industry reports.

Time: 16 minutes total, including follow-up queries to try to get APAC-specific insights rather than more Western-centric reports

Old method: 45 minutes of LinkedIn scrolling, reading comments on popular posts, checking what marketing communities in Singapore and Jakarta are discussing.

The reality is Perplexity gave me solid topic validation — ”AI and brand voice” is clearly something people care about — but the regional nuance I needed a bit more prompting. I ended up using Perplexity to identify broad themes, then validating APAC relevance.

Use case 2: Fact-checking stats

The situation: I’ve been collecting AI marketing statistics from LinkedIn – the kind that get 5,000 likes and feel suspiciously made up. “87% of marketers report AI increased productivity.” “AI reduces content creation time by 73%.” “Only 12% of B2B companies have an AI strategy.” I needed to know if these were real.

What I asked: For each stat, I’d query “Verify statistic: [exact stat and claim] - find original source and study methodology”

I tested 10 statistics. Perplexity verified 6 correctly, found original sources for 4, and flagged 2 as unsourced. The 4 it couldn’t verify? Those were the ones being shared the most on LinkedIn. Turns out people love quoting statistics that don’t exist. One widely-shared stat about “AI adoption in B2B” traced back to... a single tweet from a marketing influencer. No study or data, just a confident assertion that got screenshotted and spread.

Time: About 2 minutes per stat, 20 minutes total

Old method: 5-7 minutes per stat manually Googling, trying to trace back to original source, often hitting dead ends

The citations were legitimate when they existed. Perplexity would surface the original study, publication date, sample size, methodology. When sources didn’t exist, it would say so — unlike ChatGPT, which would confidently invent a plausible-sounding study. This is probably Perplexity’s strongest use case for me. It’s significantly faster than manual verification and hasn’t hallucinated sources yet.

Use case 3: Competitive intelligence

The situation: For my fintech event marketing role, I needed to research two competing events in the region. What’s their positioning? Recent campaigns? How do they talk about themselves on their websites and LinkedIn?

What I asked: Separate queries for each event — ”[event name] APAC fintech events - positioning statement, recent campaigns, how they differentiate”

Both events are well-established with plenty of web coverage — press releases, speaker announcements, LinkedIn posts from past attendees, tech media coverage. Perplexity aggregated all of it efficiently. For Event A, I found their focus on institutional finance (banks, wealth management), keynote speakers from DBS and UOB, estimated 800+ attendees based on LinkedIn posts from 2024, and a clear B2B networking angle. For Event B, the positioning skewed toward fintech startups and innovation (VCs, founders, early-stage), speakers from regional startups and accelerators, smaller but more intimate format (~300 attendees), and a pitch competition component.

Time: 12 minutes for both companies

Old method: 35-40 minutes (visiting each event website, reading past coverage, LinkedIn stalking previous attendees, checking speaker backgrounds)

What made this work was that both events are large enough to generate consistent web coverage. Every speaker announcement gets a press release. Past attendees post on LinkedIn. Tech publications cover the event. Perplexity had plenty of current, high-quality sources to pull from.

Use case 4: Topic validation

The situation: Before committing to “AI making B2B marketing more human” as my first newsletter topic, I wanted to validate that it was actually interesting—not just something I thought was clever.

What I asked: “Are B2B marketers discussing how AI creates more time for customer relationships rather than less? Recent trends 2025-2026”

Perplexity found several articles and discussions about AI freeing up time for strategic work, including a few that specifically mentioned increased customer interaction. The trend was real — people are talking about this.

Time: 8 minutes

Old method: 20-30 minutes scrolling LinkedIn, checking marketing Slack communities, reading recent articles

The value here was speed. Perplexity quickly confirmed the topic had traction.

What I’m learning a week in

Where it actually saves time:

Fact-checking is the killer use case. This is where Perplexity clearly beats my old workflow. Two minutes to verify a stat vs. 5-7 minutes of manual Google detective work. And it either finds the real source or tells me there isn’t one — unlike ChatGPT, which would invent a plausible study.

Competitive intelligence for well-covered companies. When researching the two fintech events, Perplexity was genuinely impressive. Twelve minutes got me everything I needed — positioning, speakers, audience, reviews — because both events had strong web presence. This is where it shines: entities with consistent online coverage.

Content ideation gets faster. Fifteen minutes to validate that “AI and brand voice” is a real concern vs. 45 minutes of LinkedIn scrolling. It’s good at surfacing what’s trending broadly, even if it can’t tell me if it matters in Singapore specifically.

No more tool-hopping. Not bouncing between Google, ChatGPT, LinkedIn, and verification tabs makes the whole process feel less chaotic. Everything’s in one place.

Where it falls short:

Struggling with local, specialised APAC research. For Singapore, Jakarta, Manila—anywhere outside the US and Europe—Perplexity defaults to Western sources talking about “global trends.” If my newsletter was focused on US B2B marketing, this wouldn’t matter. But it’s not.

The sources exist — regional marketing publications, local tech blogs, LinkedIn groups— but Perplexity doesn’t surface them. Google with geographic filters still wins here.

The pattern is clear. Large, well-covered entities (major events, big companies) = Perplexity works great. Smaller regional players with limited web presence = outdated or incomplete information. But this is not so much a tool problem as it is a data availability problem.

Topic validation only goes halfway. Perplexity confirmed that “AI making marketing more human” was trending. What it couldn’t tell me: Does this resonate in APAC? Is this a US observation or a universal shift? I still needed LinkedIn and local communities to validate regional relevance.

What I didn’t expect:

Follow-up queries work better than Google. Being able to refine conversationally ”Now show me Singapore-specific sources” or “Focus on 2025-2026 only” — is smoother than rebuilding a Google search query.

It’s honest about missing sources. When I fact-checked those LinkedIn stats, Perplexity said “no source found” for the ones that didn’t exist. ChatGPT would have invented a study. That honesty matters.

The web coverage threshold matters. I’m learning to predict when Perplexity will work well: Does this entity generate regular online content? Are there multiple recent sources? If yes, Perplexity excels. If no, expect gaps or stale data.

The AI;DR (elsewhere in the AIverse)

• Meta’s Smart Glasses privacy lawsuit — Meta got hit with a class action lawsuit over Ray-Ban AI glasses. Contractors in Kenya were viewing users’ intimate footage—bathrooms, bedrooms, the works. Marketing said “designed for privacy” while personal videos went to overseas workers.

• Luma launches creative AI agents — New “Unified Intelligence” models that handle entire creative briefs—text, images, video, audio. Give it a product photo and 200-word brief, get full ad campaign concepts. Interesting for multi-format content, but skeptical about brand voice and regional nuance.

• Raycast’s just announced Glaze — Build native Mac apps just by describing them. No coding. Runs locally, works offline, $20/month. Could be useful for custom marketing tools—event trackers, competitive dashboards. Question: Will AI-generated apps be reliable for daily use?

Meme of the week

P.S. If you're using Perplexity (or tried it and quit), I'd love to hear your experience. What worked? What didn't? Reply and let me know!

After all, the whole thing about AI is that it’s moving fast enough that something genuinely useful probably launched while I was writing this.

So, quick disclaimer: this isn’t the definitive list. It’s my list — the tools I’ve used, tested, or seen work for marketers in the real world.

What I’m sharing here are the tools I’ve actually used or watched other marketers use in their day-to-day work. Some are obvious. Some you might not have heard of. None of them are here because they have a good PR team.

With that said — here’s where I’d start:

For writing and brand voice → Claude: Best for newsletters, emails, and long-form copy that actually sounds like you. Feed it your brand guidelines and it holds them. Doesn’t produce that unmistakable “AI voice” most tools default to.

For strategy and thinking → ChatGPT: Less of a writing tool, more of a thinking partner. Use it to pressure-test ideas, build campaign frameworks, or work through a problem out loud. The o1 models are worth it for complex strategic thinking.

For research and documents → Gemini: Where it earns its place is with large inputs — massive spreadsheets, lengthy reports, multiple documents at once. If you’re doing research-heavy work or need to pull insights from a lot of data fast, this is the one.

For quick ad visuals → Nano Banana 2: Not glamorous, but fast. Good for spinning up test creatives before you invest in anything polished. The paid version closes the quality gap significantly.

For design without a designer → Canva AI: Magic Expand, background removal, AI image generation — all inside a tool most marketers already know. Won’t replace a designer, but it will stop you waiting on one for every small thing.

For ABM and demand gen prospecting → Clay: If you’re doing any kind of account-based marketing or outbound, Clay is hard to ignore. It pulls data from dozens of sources, enriches it with AI, and helps you build highly targeted lists without a full ops team. The learning curve is real but the output is worth it.

For understanding what’s working in paid social → Motion: Shows you why your creatives perform or don’t. Useful if you’re running volume on Meta or TikTok and need to stop guessing about what to kill or scale.

For meeting intelligence → Fireflies.ai: Records, transcribes, and summarises your meetings — but the useful part is what happens after. Search across past calls, pull action items automatically, and feed customer conversation insights back into your marketing. Stops useful information dying in someone’s notes.

For connecting your tools → Zapier: If you’re manually moving data between platforms, this is where to start. Straightforward automation for straightforward workflows.

For more complex automation → Make: Same idea as Zapier but built for multi-step, conditional workflows. The visual builder makes it easier to map out how everything connects. Worth it when Zapier starts feeling like a workaround.

For research without the noise → Perplexity: Marketing research that skips the SEO-optimized fluff and gives you actual sources. Faster than trawling Google for anything trend or industry related.

For turning long video into short clips → Opus Clip: Does the heavy lifting of identifying the best moments in a long video and formatting them for short-form. Useful if you’re trying to get more mileage out of webinars or interviews.

For editing video and audio → Descript: Edit video by editing a transcript. Filler word removal in one click. Voice cloning for corrections. Genuinely changes how long video editing takes.

For building landing pages and simple tools without a developer → Lovable: If you’ve ever had a campaign idea held hostage by a dev backlog, describe what you want in plain language and it builds functional web apps and landing pages. Not perfect, but fast enough to test an idea before investing properly.

For UI and front-end without code → v0: Similar to Lovable but skews more toward marketers who want polished, component-level design output. Describe a page or interface and it generates clean, usable code. Good if you’re working with a developer who can take the output further.

For stock content and voiceovers → Artlist: One subscription covers high-quality video, music, and AI voiceovers. Cleaner than cobbling together multiple stock sites and cheaper in the long run.

For ecommerce analytics → Triple Whale Moby: AI agents built on a serious base of ecommerce data. Less “chat with your data” and more “actually act on your data.” Relevant if performance marketing is part of your remit.

For real-time trends → Grok: Useful specifically for what’s happening on X right now. Not a general research tool, but good for staying current on fast-moving conversations.

This Week in AI

• Perplexity launches Computer — Perplexity’s new multi-model system routes complex projects across 19 different AI models to get each task done by the best tool for the job. Think competitor analysis reports or full app builds, running autonomously for hours or months. Currently available for Max subscribers on web.

• Google adds agentic features to Gemini and Opal — Gemini on Android can now complete multi-step tasks inside third-party apps like booking an Uber or ordering food, rolling out in beta on Pixel 10 and Samsung Galaxy S26. Google Labs also upgraded Opal, its no-code app builder, with an agent that automatically figures out the best approach to your goal and executes it.

• Anthropic expands Claude Cowork for enterprise — Anthropic added a wave of new Cowork plugins covering HR, investment banking, design, private equity, and more, alongside connectors to Google Workspace, DocuSign, and WordPress. Admins can now build private plugin marketplaces and deploy AI agents across entire departments.

Coming up, I’m going to start going deeper on individual tools — what they actually do well, where they fall short, and how marketers are using them day-to-day.

Which tool would you want me to start with? Let me know!

I’m Gary Lim, a senior marketing manager from Singapore.

Like many of you, I’ve spent the last year and a half watching AI go from “interesting demo” to “wait, this could genuinely change how I work” faster than I could keep up with.

Every week there’s a new model, a new tool, a new way of doing things. It’s exciting. It’s also exhausting, and if I’m being honest, sometimes anxiety-inducing. Am I adapting fast enough? Am I using this right? Is everyone else figuring this out faster than me?

These past months I’ve been watching other marketers in our region quietly figure out ways to use AI that actually work—and I realised nobody’s really documenting it.

Everyone on LinkedIn is posting about their amazing AI workflows, but when I talk to people one-on-one, we’re all just... figuring it out. Nobody really knows if what they’re doing is good or if we’re all just making it up as we go.

So that’s what How Marketers AI is: marketers sharing how they’re really using AI, what’s actually working, and what the hype is missing.

Not “10 prompts that will transform your workflow”, “How I 10x’d my content output in 30 days”, or anything so performative like that—just honest thoughts and conversations about what’s changing in our work.

Each edition, I’ll publishing real insights on what’s shifting, and ideas you can apply to your context, with the occasional conversations with a real marketer.

But first: if this resonates, hit reply and tell me. What questions are you wrestling with? What’s confusing you? What do you wish someone would write about?

Second: if you know a marketer, preferably in APAC, who’s doing interesting work with AI—someone with a perspective, someone who’s paying attention—introduce us. I’m always looking for people to talk to.

Third: If you know another marketer who's tired of the hype and wants real conversations, send them this link.

I'll be in your inbox next week with the first real issue. Let’s figure this out together.

Gary

Across Asia and its neighbours, impact-first startups are tackling some of the region’s most pressing challenges, like helping farmers increase their income, bringing healthcare to remote areas, and making education more equitable.

But in markets where trust in new technology can be low, digital literacy uneven, and access to infrastructure inconsistent, having a brilliant product is not enough. Without marketing that connects with people’s real needs, habits, and contexts, adoption stalls.

These three startups show why impact-savvy marketing isn’t about flashy ads or clever slogans, but about about empathy, trust-building, and translating value in ways that resonate with the communities you’re trying to serve.

iFarmer: Building trust through local partnerships

When iFarmer set out to connect smallholder farmers in Bangladesh with finance, buyers, and agronomy advice, it faced a barrier bigger than technology: trust. Many farmers had been burned before by outsiders promising quick profits or easy loans. Years of scams and failed projects left them wary of anything new, no matter how well-intentioned.

Rather than pushing its app through mass advertising, iFarmer went local, partnering with microfinance institutions, agricultural cooperatives, and NGOs that already had strong community relationships. These organisations became iFarmer’s bridge to credibility, introducing the platform in settings where farmers felt safe and respected. Village meetings and in-person onboarding sessions gave people the chance to ask questions face-to-face, see demonstrations, and hear from early adopters.

This approach worked because it placed human trust ahead of digital adoption. Once farmers felt confident that iFarmer was an ally rather than a risk, word-of-mouth referrals took over. Today, the platform supports over 100,000 farmers.

Halodoc: making the unfamiliar feel safe

Telemedicine offers enormous potential in Indonesia, where reaching a doctor in person can require hours of travel and high costs. Yet when Halodoc launched, many people were sceptical. Could a doctor on a screen really replace one in person? Concerns about misdiagnosis, data privacy, and affordability slowed adoption, particularly outside major cities.

Halodoc tackled this hesitancy by grounding its marketing in real human stories. Campaigns showed relatable situations like parents getting timely care for a feverish child and elderly patients avoiding long and exhausting trips, always with the reassurance of qualified doctors on the other side of the screen.

To reach rural communities, Halodoc partnered with local health workers who could explain the service in person, often in local dialects. Offline events gave people a chance to try the app on the spot, removing the mystery from the experience.

By combining high-tech healthcare with high-touch outreach, Halodoc transformed telemedicine from an abstract concept into a trusted household service.

Edmicro: making schools their strongest advocates

Vietnam’s education market is crowded with free online content and a deeply rooted tutoring culture. For Edmicro, which offers an adaptive learning platform aligned to the Vietnamese curriculum, the challenge wasn’t just proving that it worked—it was convincing teachers and parents that it was worth integrating into daily learning.

Rather than trying to sell directly to parents through ads, Edmicro embedded itself in schools. It ran free pilot programmes in low-income districts, allowing students and teachers to experience the platform without financial risk. Crucially, the company worked closely with educators to adapt the platform to their needs, providing training and support so it became a tool they enjoyed using rather than a burden. Teachers who saw results began sharing their experiences online, often through personal Facebook posts that reached far into their networks.

These endorsements carried far more weight than any marketing campaign could have achieved. By making teachers advocates rather than just users, Edmicro built a growth engine powered by trust and community credibility.

Marketing, the bridge between tech and trust?

If there’s anything that these companies prove, it’s that that startups often don’t scale on good intentions alone. In Southeast Asia and surrounding markets, the leap from “great idea” to “widely adopted solution” is built on trust—and trust is built through marketing that’s grounded in local reality.

For founders, this means marketing can’t be an afterthought or a final step before launch. For investors, it’s a reminder that funding the tech is only half the battle. Without a strategy for how that tech will be understood, trusted, and embraced by the people it’s meant for, the most promising solutions risk fading into obscurity.

As global marketers we’ve long been told: cram the logo in, shout through the ear-drums, and glitter-bomb the ROI data. Hyundai, in collaboration with Innocean Seoul, disregarded those rules with a 13-minute short film shot entirely from cameras mounted on its Ioniq 5 electric vehicle—and accepted a Grand Prix at Cannes Lions 2025 for doing so.

Titled Night Fishing, the production directed by Moon Byoung-Gon, starring actor Son Suk-Ku. The film unfolds as a paranormal thriller at a charging station, using only seven fixed in-car cameras. Crucially, there is no brand mention or product reveal until the end credits.

The approach, described by Hyundai as a “snack movie”, defied standard practice in automotive advertising, which typically showcases vehicle features, speed shots, and logo placement. Instead, it adopted a cinematic narrative designed to be chosen by audiences rather than pushed to them.

Audience first, brand second

Hyundai’s global chief marketing officer Sungwon Jee said the work reflects a changing reality for marketers. “In the past people chased brands. Now brands have to chase people… not just to be seen, but to be chosen”.

The film was distributed in cinemas and online, with some audiences paying a nominal ticket price to watch it—a deliberate move to frame the content as entertainment rather than advertising.

Results suggest the gamble worked. According to Hyundai, the campaign generated a 26% increase in dealership visits and a 55% surge in Ioniq test drives during its run.

Night Fishing took the Grand Prix in the Entertainment Lions category and a Silver Lion in Film Craft. Hyundai’s other campaign, Tree Correspondents, an AI-powered environmental initiative under its IONIQ Forest programme, also won two Gold Lions and another Silver in Digital Craft.

Could long-form branded storytelling prove commercially viable when the content competes with entertainment rather than other ads? As the Cannes jurors noted, Night Fishing was “advertising disguised as art”. In 2025’s cluttered media environment, disguise might just be the ultimate competitive advantage.

As Wimbledon season passes, we take a relook at an extraordinary campaign by Jaipur Rugs, the Indian heritage brand known for its artisan-led weaving model, quietly stole headlines last year in a collaboration that felt both unexpected and entirely natural.

Made to celebrate the opening of the company's new London store and featuring a series of tennis matches on a court covered with 93 exquisite rugs, the campaign blended Indian craft traditions with the refined aesthetics of the British sport typically synonymous with white linen, manicured lawns, and strawberries with cream.

The campaign centered on a simple tennis match dubbed ‘The Jaipur Rugs Tennis Championships’, between a village team of four master artisans from the Jaipur Rugs weaving village of Manpura — nurtured through months of tennis training – and one of India’s greatest tennis players, Rohan Bopanna, between his games at the French Open and Wimbledon. Also playing was Prachi Mantri, a bronze medalist in singles at the Asian Cup held in Nepal.

This special tennis match was brought to life with gloriously detailed scenography, including the net, which was hand knotted using the same yarn used in carpet weaving.

The playing field, or the ‘Court of Carpets’ consists of 93 antique and vintage rugs that were chosen for their sophisticated patina. Beginning with a glorious antique Persian carpet at its centre, which spans both sides of the net, our set designers rolled Bakhtiar runners down the sides to define the outer boxes. The alignment of white wool fringes sharply mark the serve lines.

The campaign was amplified through PR, social media, and editorial features across design, luxury, and sports media — picking up coverage in Architectural Digest India, Wallpaper, and UK-based design blogs.

It worked. On Instagram, the campaign earned engagement from Wimbledon fans, design aficionados, and Indian diaspora accounts alike. Wimbledon is peak British tradition. Jaipur Rugs represents Indian craft. A literal court classic pairing, if I may say so myself.

You’re probably hearing a lot of AI jargon out there. You probably sort of know what some of them mean… but not really. And that’s OK.

Whether you're a CMO, content strategist, or the poor soul managing eight product launches, this AI guide for marketers is designed to keep you sounding smart without spending your weekends binging AI research videos on YouTube. Make sense of the 21 most common AI terms floating around your WhatsApp chats, team meetings, and LinkedIn feed.

1. Model

An AI model is like a super-powered brain in a box. You give it input (like a question or a photo), it processes it, and spits out a result. Think of the model as the engine that makes AI run.

There are different kinds:

• LLMs (large language models) like ChatGPT, Gemini, and Claude: they process and generate text.

• Image models like Midjourney or DALL•E: they generate images.

• Video models like Sora or Veo: they create moving visuals.

• Voice models like ElevenLabs: they replicate speech.

2. LLM (large language model)

These are AI models trained to understand and generate human-like text. Most can now also handle images, voice, and even videos, making them multi-modal. The "GPT-4o" model in ChatGPT? That "o" stands for "omni", because it can accept text, image, or audio inputs.

3. Transformer

No, not the robot kind. Transformers are a type of AI architecture developed by Google in 2017 that changed everything.

They introduced "attention," which lets the model understand relationships between words all at once, rather than one by one. It's why AI can now write poems, code apps, or help you brainstorm campaign taglines.

4. Training / Pre-training

Training is how an AI model learns. Developers feed it billions of words from books, websites, Reddit threads, movie subtitles—you name it. The model tries to predict the next word in a sentence, gets feedback, and adjusts. Over time, it gets really, really good at language.

5. Supervised learning

When a model is trained with labelled examples (e.g. this email = spam, this one = not spam), it’s called supervised learning. The model learns from examples where the "right answer" is known.

6. Unsupervised learning

No labels here: the model has to figure things out itself. Unsupervised learning is great for grouping similar things or spotting anomalies (e.g. fraud detection, market segmentation).

7. Post-training

This is the polishing phase. Once a model finishes pre-training, developers fine-tune it for specific use cases or industries (e.g. medicine, law, marketing).

8. Fine-tuning

You take a general-purpose model and give it a crash course in your own data. Want it to write in your brand tone? Train it on your emails, blogs, and Slack messages.

9. RLHF (reinforcement learning from human feedback)

AI doesn't always do what we want. RLHF fixes that. Humans rank outputs ("this response is better than that one"), and the model learns what people prefer. It’s why ChatGPT feels more helpful and polite than a raw language model.

10. Prompt engineering

How you talk to an AI matters. Prompt engineering is the skill of crafting inputs that produce the best responses.

"Write me a tagline for an AI-powered instant noodle brand" might work. "Act as a creative director. Pitch 5 witty, Gen Z-style taglines for an AI-enabled ramen brand" will work better.

11. RAG (retrieval-augmented generation)

Models can sometimes forget things or make stuff up. RAG helps prevent that. It lets the model search your documents, databases, or websites for relevant info before answering you. Think of it as giving the model Google access in real time.

12. Inference

This is when the model is "running." You send it a prompt, it replies. That's inference.

13. MCP (model context protocol)

MCP is an emerging standard that lets AI models talk to your apps—calendar, Slack, HubSpot, even GitHub. Instead of you copying and pasting data, the model can fetch it and act on it. Magic.

14. Hallucination

When AI just makes things up. Common with LLMs. You ask for facts, it gives you fiction. RAG helps reduce hallucinations. Always double-check your AI-sourced insights.

15. Token

AI doesn’t read text like humans. It reads chunks called tokens. "AI is cool" = 3 words = 4 tokens. Most models have token limits, so keep that in mind for long documents.

16. Latency

The delay between your question and the AI’s answer. Faster models = lower latency. Important if you want real-time customer support bots or campaign testing assistants.

17. Fine-tuning vs RAG

Confused between the two? Think of it like this:

• Fine-tuning = teach the model new knowledge permanently

• RAG = give it temporary access to relevant info

Want your AI to always know your company values? Fine-tune it. Want it to pull the latest sales numbers? Use RAG.

18. Multimodal

This means the model can work with more than one type of input: text, image, voice, video. GPT-4o, Gemini 1.5, and Claude 3.5 all support this now.

19. API

Stands for Application Programming Interface. Think of it as a pipe that lets your website or app access the AI model programmatically. Marketers use APIs to build chatbots, automate email writing, or personalise landing pages.

20. Agent

An AI agent isn’t just answering one prompt—it’s working toward a goal. Agents can complete multi-step tasks like:

• Pulling data

• Making a deck

• Writing an email

• Sending it via Gmail

AutoGPT, OpenDevin, and ChatGPT agents are early examples.

21. Vector database

A special kind of database that stores data in the form AI models understand: numbers. Super useful for search and RAG systems, especially when trying to match questions with answers.

Final thoughts

Feeling overwhelmed by all of these? Don't stress. The goal isn’t to become an AI engineer overnight, but to build a working vocabulary so marketers like you can ask smart questions, scope the right projects, and avoid being bamboozled by AI buzzwords in the next strategy meeting.

Bookmark this glossary, share it with your team, and remember: you don’t need to know everything about AI to start using it effectively—you just need to know enough to be dangerous.