The Trust Drawdown

Here's a number that should make every platform executive quietly close their laptop and stare out of a window: 63% of consumers say they'd be less likely to buy from a brand that uses AI-generated ads.

Not "mildly annoyed by." Not "vaguely suspicious of." Less likely to buy from. That's stated purchase intent — the one metric the entire optimisation machine exists to protect — moving in the wrong direction because of the optimisation itself.

New research shared at this year's Cannes Lions (by The Harris Poll, the 4As, and Infillion) has put hard numbers on something we've all been sensing: consumers are not just noticing the AI flood. They're actively recoiling from it.

Two-thirds of global consumers now regard AI messaging as a marketing ploy. Seventy-eight per cent said it makes advertising feel less authentic. The same proportion said brands come across as cringeworthy when they lean on it too heavily. And 73% said they'd be less likely to trust an ad they suspected was machine-made.

That's not soft brand-perception noise you can bury in a quarterly deck. That's the audience telling you, in terms that don't require a data scientist to interpret, that they can see what you're doing and they don't like it.

Everyone Hates It. Everyone's Doing It.

Here's where the comedy writes itself (and not the good kind).

At the exact same moment consumers are developing an allergic reaction to AI-stamped advertising, the industry is doubling down on it with the enthusiasm of someone who's just discovered a gym membership they forgot to cancel — except in this case, they're choosing to keep paying.

A majority of practitioners now use AI for idea generation and concepting. Around half use it for producing visual assets. Roughly four in ten are deploying it for copywriting, targeting, and media planning. Separate analysis from Sensor Tower shows a double-digit rise in ads that reference AI directly, with "AI-powered" now sitting among the most common phrases in app store descriptions (right up there with "revolutionary" and "game-changing," two words that have never once changed a game).

Supply of AI signalling is rising at precisely the rate consumer tolerance for it is falling. If you plotted both lines on the same chart, the intersection point would be labelled "we should probably talk about this" and the industry would still be three quarters behind it.

The Tragedy of the Trust Commons

So why are intelligent operators — people who can read a survey, interpret a trend line, and presumably tie their own shoes — collectively accelerating into a wall they can see?

Because the incentive structure is perfectly designed to make this the rational individual choice, even when it's collectively suicidal.

The efficiency gains from AI are immediate, measurable, and attributable. A team that swaps a studio shoot for a generated asset can point to the saving this quarter, in its own budget, on its own dashboard. You can practically hear the CFO purring.

Trust erosion works in the opposite way. It's slow, diffuse, and collective. No single AI-assembled ad measurably drains the reservoir. The decline shows up in aggregate, across a category, over months, and can't be traced back to any one actor's decision. Each firm captures the full benefit of its own optimisation while paying only a tiny, shared, heavily lagged fraction of the cost.

This is, for anyone who sat through an economics degree (or at least the first year before switching to something more employable), a textbook commons problem. Trust is the shared pasture. Every operator's AI-generated cow is grazing on it. And nobody's counting the grass.

The kicker is that trust is a stock, not a flow. It accumulates slowly through years of consistent, credible communication and depletes quickly once consumers decide they're being handled rather than addressed. A reservoir drawn down over eighteen months does not refill on the same timetable. The efficiency curve and the trust curve can move in opposite directions for a sustained period without anyone appearing to have made a mistake — because they're measured on different clocks, and only one of them reports quarterly.

Meta: The Most Impressive Machine Nobody Trusts

Nowhere is this tension more perfectly embodied than at Meta. The commerce package they unveiled around Cannes — real-time creative assembly from product-catalogue data, a brand memory that learns an advertiser's tone, business agents handling customer conversations, live shopping, expanded affiliate and checkout tooling — is the most fully realised version of the efficiency logic the industry has ever produced. It is, on a purely technical level, extraordinary.

It's also being built on a surface whose only real asset is the attention it has spent two decades learning to capture and hold. Meta is refining the machinery of attention capture at the exact point at which the trust that gives captured attention its commercial value is declining. That's a bit like upgrading your fishing equipment while the lake is being drained — technically impressive, strategically questionable.

There's a subtlety here worth flagging. The research suggests the backlash falls hardest on visible AI — the loud "AI-powered" claim, the ad that looks like it was assembled by an algorithm with a stock photo subscription. The rational platform response (and you can practically watch it happening in real time) is to make the AI invisible: assemble, target, and optimise beneath a surface that still shows a creator tagging a product or a clean checkout flow.

That manages the symptom. It does not address the cause. It reduces the immediate perception problem while leaving the underlying question — whether consumers are being addressed honestly — unresolved and arguably compounded. Hiding the machinery doesn't rebuild the trust. It just means the trust is being spent less noisily.

When Authenticity Becomes a Premium Feature

The clearest signal that trust has become genuinely scarce? Some brands have started charging a premium for it.

Apple has leaned into handcrafted, behind-the-scenes advertising that foregrounds human labour. Others — Almond Breeze and Equinox among them — have used deliberately artificial-looking output semi-ironically, poking at the technology to draw attention to the realness of what they actually sell. These are early moves, and some are more gesture than strategy, but the direction is consistent: authenticity is being repositioned from a default assumption into a deliberate point of difference.

Here's the rule: when a quality becomes worth advertising, it's because it's become scarce enough to price. We are now, officially, in the era where "a human being made this" is a selling point. Let that one sit for a moment.

What This Means for Us

The efficiency case for AI across creative production and media is real and it isn't reversing — competitive pressure alone guarantees continued adoption. What's changed is that the trust cost is now measurable enough to belong in the same conversation as the efficiency saving, rather than being treated as a free input that materialises from nowhere (like WiFi at a hotel that definitely, definitely works).

Three things follow:

Treat AI as infrastructure, not as a marketing message. The data is unambiguous: stamping "AI-powered" on consumer-facing material is a cost, not a benefit. The value is in what the technology does, not in announcing that it did it. Nobody cares that your recommendation engine runs on a transformer model. They care that it recommended the right product.

Protect the human-legible signals. The parts of your brand's communication that a consumer reads as evidence a person stood behind the decision — those now command a premium. AI is increasingly effective at driving discovery, but the trust that closes a conversion still tends to rest on signals a consumer can attribute to human judgement. Optimising the discovery layer while hollowing out that trust is a false economy if it depresses the conversion the whole funnel exists to produce.

Recognise the efficiency advantage is temporary. When every operator has access to the same generative tools (and they will — these things commoditise faster than a new restaurant concept in Shoreditch), the cost advantage equalises. What remains as a differentiator is precisely the trust asset that the rush to optimise is depleting.

The Bottom Line

The industry can keep optimising against a declining trust reservoir for a while yet, because the efficiency it gains reports immediately and the trust it spends reports late. The risk isn't a dramatic cliff edge. It's the slow, dawning realisation that the attention you've become exceptionally good at capturing is no longer worth what it was — because the people on the other side have quietly decided to believe less of what they're shown.

The operators best positioned for that moment will be the ones who treated trust as a balance-sheet asset rather than a free resource. Who spent the current window of cheap efficiency building the credibility that can't be generated on demand.

The technology that makes everything cheaper to produce does not make trust cheaper to earn. If anything, it's about to make trust the most expensive thing in the market.

And, ironically, the most valuable.

• Getty Images strikes a display deal with OpenAI, surfacing its licensed libraries across ChatGPT search and discovery. Licensed visuals signal AI platforms paying for imagery rather than scraping it, betting trust drives adoption.

• TikTok's Symphony Agent builds entire video ad campaigns from text prompts, analyzing trending clips and generating footage via ByteDance's Seedance 2.0. Marketers skip the camera and reverse-engineer what's already winning.

• McKinsey argues AI pays off in ecommerce when four levers (growth, productivity, efficiency, profitability) compound as a flywheel, not isolated pilots. Small shops start with customer-feedback loops, not enterprise data stacks.

• OpenArt launches Director, pitching "vibe directing" as video's answer to vibe coding: describe an idea, refine it through conversation, generate finished footage. Production collapses into a chat with no camera or crew.

• A marketer maps how non-engineers build AI workflows, from picking models to setting up code editors to controlling token spend. The pitch: automation needs job knowledge, not a CS degree, plus routing cheap tasks to cheap models.

• Gemini 3.5 Flash gains computer use as a built-in tool, letting agents see, reason, and act across browser, mobile, and desktop. Google folds standalone agentic control into its main model, targeting enterprise automation.

🔠 Copywriter - Helps write engaging ad copy to drive action.

💻 Brainner - Streamline talent acquisition by automating AI-driven resume analysis

🌐 Ayraa AI: Your personal search engine and knowledge assistant at work. Connects with Slack and your web browser to instantly capture and organizes everything you do at work.

🌊 Instant Summaries by Shortwave - Smart TL;DRs for every email.

💥 Zocket - AI social media ad platform

😍It was great, all I needed

🙂Good, not great

🙄It sucked