Can you use AI images in marketing?

Yes, but the register decides everything. Research shows consumers accept AI imagery most readily for inanimate subjects such as atmosphere, textures and environments, and punish it hardest where it replaces a credibility signal: people, products and proof. A workable rule for a small business: AI may color the frame, never be the painting. Product imagery is always the real screen.

Do AI images have to be labeled under the AI Act?

It is more nuanced than often claimed. Article 50, in force from 2 August 2026, requires providers to make AI output machine-readable and demands visible labels mainly for deepfakes and AI text on matters of public interest. An atmospheric image on a company website does not automatically fall under the strictest labeling duty. The direction of the law is clear though, and research shows that voluntary transparency also simply works better than hiding.

Why show dummy data instead of real customer data?

Because it is the only combination that solves everything at once. Buyers trust their own observation of a working product more than any claim, but publishing customer data is something you may not and should not do. A real screen with coherent, fictional data is costly to fake, respects your customers' confidentiality, and shows exactly what the software does. Say honestly that the names are fictional: that transparency demonstrably raises trust.

Do people actually recognize AI images?

Barely, and that is the heart of the problem. In a Dutch study, 81 percent of participants took an AI image for a real photo, and real photos, the other way around, were regularly labeled as AI. In an American experiment, only a quarter of viewers were certain that an AI ad was AI. So the eye cannot check origin, which makes every claim about realness unverifiable. What remains is proof that verifies itself: a product you can click into.

Why real product screens beat AI imagery in marketing

In December 2025, McDonald’s Netherlands pulled a fully AI-generated Christmas commercial within four days after a storm of criticism (BBC, 2025). The easy lesson is “keep AI out of your marketing”, but that is not what the data says: in nearly every study, the majority of consumers is neutral about the use of AI. The real lesson cuts deeper. Now that every polished image can be made for free, a polished image is worth nothing as proof. What remains is proof that is costly to fake: the real screen of a real product, with fictional data in it if need be. Whoever dares to show that no longer needs to claim anything.

This essay is the practical side of ai slop and trust, where I work out the disclosure question channel by channel. This one is about a single rule for anyone selling software or services: show the screen.

what happened to trust in images?

It is tipping, and faster than most marketing departments realize. In March 2026, Gartner measured that half of American consumers prefer brands that do not use generative AI in customer-facing content, and that 68 percent regularly doubt whether what they see is real (Gartner, 2026). Research agency Fractl asked the same questions in two consecutive years and saw distrust of heavy AI use double, from 20 to around 40 percent (Fractl via Search Engine Land, 2026). And the largest consumer study in the series, eight thousand respondents across eight countries, found that visible AI content makes people four times as likely to trust a brand less rather than more (Klaviyo via eMarketer, 2026).

Two nuances belong here, because without them this essay becomes an exaggeration itself. First: in all of these studies the middle group is neutral; the strong disapproval comes from a minority. That minority, however, is vocal, visible and reputation-defining, as McDonald’s discovered. Second, and against intuition: the strictest judges are not the older generations but precisely the one that works with AI the most itself. Of Gen Z, 54 percent say they trust a favorite brand less after heavy AI use, against just over 30 percent among older generations (Fractl, 2026). The Netherlands shows the same pattern: 45 percent of 18 to 29 year olds disapprove of AI in advertising, more than people in their thirties and forties (Norstat, 2025). Whoever assumes AI imagery works fine “for the young target group” has it exactly backwards. And this is no snapshot: the Dutch AI Barometer shows use and distrust rising together, use from 47 to 65 percent, negative attitudes from 25 to 31 percent (MSI-ACI via Nederland Digitaal, 2026).

can’t you just hide the ai imagery?

Hiding is the worst-performing strategy there is, for two reasons that reinforce each other. The first: it does not even work when it succeeds. In a controlled experiment where AI ads and human-made ads were produced to identical briefs, only a quarter of the three thousand viewers could point out the AI versions with any certainty. Yet the AI ads scored 14 percent weaker on sales effect and 17 percent weaker on brand value (Ipsos, 2026). People do not consciously see it, and apparently feel it anyway.

The second reason: openness actually pays. In the Yahoo/Publicis research, a visible AI disclosure raised the perceived trustworthiness and appreciation of ads considerably, by tens of percentage points depending on the category (Yahoo & Publicis Media, 2024). And the counterexample exists too: when Vogue ran ads with AI models for Guess, the disclosure was there, but so small that readers felt deceived anyway (CNN, 2025). Transparency only works when it is visible; fine print counts as hiding. The law is moving the same way: article 50 of the AI Act, in force from 2 August 2026, requires machine-readable marking of AI output and visible labels for deepfakes among other things. Its reach for ordinary marketing imagery is narrower than often claimed, but the direction is unmistakable (AI Act, article 50).

what does convince a business buyer, then?

Their own eyes, and that is not intuition but one of the most stable findings in b2b research. In the annual Buying Disconnect series from review platform TrustRadius, an interested party itself, but with a consistent method across five editions, product demos and free trials top the list of sources buyers trust year after year, above testimonials, above ROI claims, above the sales conversation (TrustRadius, 2023). At the same time, that same research reports that 73 percent of buyers regularly suspect fake reviews (TrustRadius, 2024). The proof system itself is under pressure, and that raises the bar: the more claims turn out to be fakeable, the more weight lands on what a buyer can see and click for themselves.

Why that works is by now academically documented as well. Peer-reviewed research into perceived AI authorship of marketing copy shows a chain: what is experienced as machine-made feels less authentic, that triggers a form of moral unease, and that translates into less loyalty and fewer recommendations (Kirk and Givi, 2025). That research was about text, not images, so I generalize the mechanism, not the numbers. But the mechanism is exactly what you see in all the data above: people do not punish technology, they punish the feeling that a credibility signal has been replaced by something that no longer costs anything.

why does a real screen also beat polished human work?

Because polish itself has become the warning sign. That is the uncomfortable version of this story: it is not AI against human, it is fakeable against unfakeable. The Dutch Norstat study makes that painfully concrete: 81 percent of participants took an AI image for a real photo, and real photos, the other way around, were regularly labeled as AI (Norstat, 2025). The eye cannot establish origin, in either direction. That makes the claim “we do not use AI” unverifiable too, exactly as unverifiable as the image itself.

What a visitor can verify for themselves: an interface that behaves like a working product. Browser chrome, loading states, a slightly clumsy company name in the sample data, buttons that lead somewhere. Screens like that are costly to fake, because real software has to sit behind them. It is the same logic by which a brand like Aerie positioned itself explicitly against AI imagery and grew double digits that quarter, though that is one case and a correlation, not proof (The Cut, 2026). The lesson is not that everyone should start advertising as anti-AI. The lesson is that provable realness is the only marketing material that gains value as faking gets easier.

how do you show proof without leaking customer data?

With one rule and four agreements. The rule: product imagery is always the real screen. The agreements:

The data on screen is always fictional, but coherent: an invented company with names, addresses and amounts that add up, no lorem ipsum. You leak nothing and the screen stays credible.
Say so, visibly: real screens, fictional data. All the research above shows that this openness raises trust rather than lowering it; fine print does the opposite.
Reserve AI for the one register where research tolerates it: atmosphere. Backgrounds, textures, abstraction. Never the product, never people, never proof.
Let people click for themselves wherever you can. A demo or trial environment beats any image, moving or not.

None of this is theory for me: every product screen on this site, from the homepage to the case pages, is the real software with a fictional customer in it, and it says so right there. Not because a law demands it, but because it is the only claim that proves itself.