They Studied the Audience Through the Wrong End of the Glass.

A thirty-eight point gap between fear and finding

FreeWheel published research on 5 September 2026, conducted with AdExchanger and Dynata, reporting that 48 per cent of media buyers and 39 per cent of sellers assume viewers find AI-generated creative off-putting, while only 10 per cent of viewers said they disliked the AI advertising they had seen. The publisher does its own arithmetic on the gap: 38 percentage points on the buyer side, 29 on the seller side.

Before trusting any of it, follow the trail back to who paid for it. The samples are stated: 226 media buyers and 50 media sellers surveyed by AdExchanger, 2,496 US adults by Dynata, and an earlier survey of 216 marketers and agencies. Viewers also reported broad openness to specific AI uses, with 89 per cent open to AI reducing ad repetition, 89 per cent to AI choosing when ads appear so as to minimise disruption, and 76 per cent to AI creating personalised advertising. Only 22 per cent were open to fully autonomous AI campaign management.

This gap already has a name in the peer-reviewed record

The most useful thing about this ai ad creative finding is that it is not new in kind. Marketing science has a well-established name for practitioners misjudging audiences by projecting themselves onto them. Research published in the Journal of Marketing Research in 2021, on marketers projecting personal preferences onto consumers, examines the false consensus effect among marketing practitioners across two pilot studies and four experiments, and identifies preference certainty as the key moderator.

That matters practically, and it is where the scent leads somewhere useful. If the gap is a known cognitive bias rather than a one-off survey artefact, then it is predictable, it recurs, and it can be designed against. A team that knows it is prone to projecting its own discomfort onto an audience can build a cheap correction into its process, which is simply asking the audience rather than asking the room. Most ai creative advertising decisions this year are being made in rooms full of people who are professionally fluent in AI and therefore unusually sensitive to its fingerprints.

That voice is worth keeping in any ai ad creative review too. Ten per cent is a minority, but a vocal, articulate and motivated one, and aggregate indifference is not the same as universal indifference. A brand with an audience that skews toward that view carries a different risk profile to one whose audience does not, and no cross-sectional survey of 2,496 US adults will tell you which you have.

The complication: disclosure carries a real cost

Here is where a comfortable story stops being comfortable, and where most ai ad creative debates quietly stop reading. Viewers may not mind AI creative they were never told about. The peer-reviewed evidence on what happens when you do tell them is considerably less encouraging, and it is substantially stronger evidence than a gated vendor survey.

The most robust single body of work is The Transparency Dilemma, published in a leading peer-reviewed journal of organisational behaviour and decision-making, which reports thirteen pre-registered experiments finding consistently that actors who disclose their AI usage are trusted less than those who do not, supported by a within-paper meta-analysis across 4,093 observations. Thirteen experiments is not a fluke, and the effect runs in the opposite direction to the reassurance the FreeWheel headline offers.

Advertising-specific work points the same way. A study in the Journal of Research in Interactive Marketing with 358 participants found AI disclosure diminishes advertising value and purchase intentions, mediated by reduced credibility. Work from Tilburg University found disclosure produced a direct decrease in advertising attitude, with brand-level damage concentrated among consumers already averse to AI. And a 2026 paper in the Journal of Marketing Analytics tested whether explaining the reason for AI use softens the blow: it does not, and in some cases makes evaluations worse.

The counterweight is real and should be stated, because it is the part that decides whether ai ad creative is usable at all. A Columbia working paper on AI-generated advertising imagery in live display campaigns found the average effect of an AI-generated image on click-through rate was statistically null, and that what actually depresses performance is perceived artificiality rather than AI origin as such. In other words, well-made ai ad creative that does not look synthetic performs normally. Badly made creative that looks synthetic does not, whether or not a machine made it.

What this means for AI ad creative decisions

Put the two literatures together and the practical position on ai ad creative is narrower than either headline suggests. Audiences are broadly indifferent to AI involvement they cannot detect. Disclosure reduces trust. Quality, not provenance, drives measured performance. And practitioners systematically overestimate audience hostility.

The obvious commercial reading of that, which is to use ai ad creative freely and say nothing, is also the one that ages worst. Disclosure obligations are arriving regardless of preference. The Advertising Standards Authority's own guidance on artificial intelligence in advertising already addresses how AI-generated material is treated under existing rules on misleadingness, and disclosure regimes generally tighten rather than loosen. A creative strategy that only works while nobody has to mention it is a strategy with an expiry date somebody else controls.

Where the disclosure penalty is smallest

The literature does offer a usable steer on where ai ad creative is safest. The Tilburg finding that brand damage concentrates among AI-averse consumers implies the penalty is audience-specific rather than universal. Work in the Journal of Advertising on AI-generated charitable giving appeals found the negative effect ran through empathy and anticipatory guilt, which suggests the penalty is largest exactly where the creative asks for an emotional or moral response and smallest where it does not. A synthetic product shot in a specification-led category is a very different proposition to a synthetic human face in a fundraising appeal. The undergrowth is thicker in some categories than others, and ai ad creative should be routed accordingly.

A study in the Journal of Interactive Advertising with 304 participants found disclosure ran along two paths at once, increasing conceptual AI knowledge while also shifting attitude, which is a reminder that the effect is mediated rather than mechanical. Consumer trust in ai is not a single dial that disclosure turns down; it is several, and they do not all move together.

The brief a brand strategy team should actually write

A fox that judges the whole wood by the one bird that shrieked at it will go hungry. The point is not that the alarm is wrong, it is that one alarm call is not a census. Reading a room of nervous colleagues as though it were the audience is the same error in a nicer building. A quarry that never existed still costs a whole night of prowling the hedgerow.

So the brief writes itself in four lines. Stop using internal squeamishness as an audience insight, and test the creative with the actual audience instead. Judge ai creative production on whether it looks synthetic, not on whether it is, because that is what the performance data actually responds to. Assume disclosure is coming and pressure-test the work against a world where it is mandatory rather than optional. And treat the categories that trade on empathy, of which charity, health and anything involving a human face are the obvious ones, as the places where the penalty lands hardest and the caution is genuinely warranted.

The uncomfortable version of this article's finding is the honest one. The permissive result and the restrictive result are both true, they apply at different moments, and the gap between them is not a loophole to exploit but a genuine tension to manage. Anyone selling ai tv advertising strategy on the strength of the 10 per cent figure alone is quoting half a literature, and the other half is the half with thirteen experiments behind it. Good ai ad creative decisions live in that tension, not on either side of it.

Frequently asked questions