Should You Tell Customers Your Content Was Made With AI?

The short answer

Disclose when the AI involvement would change how a reasonable person reads the content, and do not disclose for routine production help. The trust penalty for undisclosed AI that customers detect themselves is significantly higher than the modest engagement cost of a label, and detection is getting easier every quarter.

The longer answer is more interesting, because the 2026 data contains a genuine tension. Consumers say loudly that they want AI content labelled. They also do not particularly reward brands that label it. Understanding why is the difference between a disclosure policy that helps you and one that just adds a badge to mediocre work.

The trust penalty doubled in twelve months

The single most striking figure in the 2026 consumer research is the rate of change. In 2025, roughly 20% of consumers said heavy AI use would decrease their trust in a favourite brand. By 2026 that had risen to around 40%. Distrust did not creep up; it doubled inside a year.

Alongside it, a 2026 Gartner survey found that 50% of US consumers would prefer to give their business to brands that do not use generative AI in customer-facing messages, ads, or content. A Clutch survey put roughly one in three consumers saying AI use negatively affects their perception of a brand.

These are stated preferences, and stated preferences overstate behaviour — people report caring about many things they do not act on at checkout. But three independent surveys moving the same direction, quickly, is a signal worth planning around rather than dismissing. The prudent reading is not that AI use is fatal. It is that the cost of being sloppy about it rose steeply and is still rising.

Who penalises hardest

Averages hide the risk here. The 2026 breakdown by age is wide enough to change decisions: 54% of Gen Z said their trust would fall if a favourite brand used AI for most of its marketing, against 33% of Gen X and 32% of Baby Boomers. Women penalised more than men, 44% against 34%.

So the question is not "do consumers mind" but "do mine". A B2B manufacturer selling to procurement managers in their fifties is operating in a very different risk environment from a D2C brand selling to twenty-four-year-olds. The second brand is exposed at roughly one and a half times the rate of the first, and its audience is also far more fluent at recognising generated work.

That fluency is the part that compounds. The tells that gave away AI writing in 2024 are now common knowledge, and audiences that spend their days inside these tools recognise the cadence instantly. You are not being judged against the average consumer's detection ability. You are being judged against your own buyers', which is usually better.

The disclosure gap

Consumers want labels across every format: 84% want written AI content labelled, 91% for video, and 90% for images. Against that, only 20% of organisations always disclose AI use to their audiences, and 33% never disclose at all.

A third of brands publishing AI content with no disclosure at all, against an audience where more than eight in ten say they want to be told, is a wide gap. Gaps like that usually close through some external event rather than through voluntary change: a platform policy, a regulator, or one well-publicised incident that makes the practice look disreputable.

The strategic point is about timing. Brands that develop a disclosure position now do it on their own terms, as a deliberate choice. Brands that wait do it later under pressure, which reads as an admission rather than a standard.

Why disclosure alone does not earn trust

Here is the uncomfortable finding that most write-ups of this data skip. Labelling content as AI-generated does not, by itself, make audiences trust it. Transparency is necessary but it is not persuasive. A label on thin content tells the reader precisely why the content is thin.

This resolves the apparent contradiction in the survey data. Consumers are not asking to be told so they can approve. They are asking to be told so they can calibrate how much to believe. Disclosure changes the standard the work is judged against, and work that was already weak fares worse under the label, not better.

Which means the disclosure decision and the quality decision are not separable. A policy of "use AI heavily and label it" performs badly. A policy of "use AI where it genuinely helps, keep human judgment on everything customer-facing, and disclose the material cases" performs well, because the label is attached to work that survives the extra scrutiny.

A disclosure policy that survives contact with reality

The rule that works in practice: disclose when AI involvement would change how a reasonable person interprets the content. Applied concretely, that gives you a workable split.

Disclose: synthetic imagery that could be mistaken for photography of a real product, place, or person; AI-generated voice or likeness; generated case studies, reviews, or testimonials; AI-written commentary published under a named expert's byline; AI-generated data or research findings.

No disclosure needed: AI used to draft outlines, suggest headlines, tidy grammar, resize or retouch images within normal production limits, translate, transcribe, or summarise internal material. Nobody discloses spellcheck, and this is the same category of assistance.

The genuine grey area is a piece drafted by AI and then substantially edited by a person. Our view is that if a named human took editorial responsibility, checked the claims, and rewrote enough that the argument is theirs, the byline is honest and no label is required. If the edit was a light pass over a generated draft, the byline is doing work the human did not do, and that is the case customers punish when they notice.

Whichever line you draw, write it down and apply it consistently. Inconsistent disclosure is worse than either policy, because it implies the undisclosed pieces were a deliberate omission.

What actually protects you

Disclosure is a floor, not a strategy. The brands that come through this period well are the ones whose content could not have been generated in the first place, because it contains things a model does not have: proprietary data, named clients, specific numbers from real projects, opinions with something at stake, and the details that only come from having done the work.

That is also the durable answer to the wider sameness problem, which we cover in why every brand suddenly sounds the same and, on the visual side, in why your brand looks generic. Content that carries evidence of real experience does not need a disclaimer to be believed, and it is the only category the trust penalty does not touch.

If you are unsure where your own content sits, the test takes a minute: take your last five published pieces and ask whether a competitor could have published them by changing the logo. If the answer is yes, disclosure is not your problem. Tell us what you are publishing and we will tell you which of the two you are dealing with.