A shop owner asked a reasonable question last month: if the AI-generated background in her new ad looked a little off, would people notice, and would it cost her anything if they did? The honest answer splits into two parts, and the split is the whole story. Whether people can tell is a weaker predictor of trouble than whether they suspect.
Start with detection. Researchers at Berkeley's Haas School of Business, Huang and Katona, marketing professors there, built a pipeline of AI-generated ad images and asked people to sort them from human-made ones. Overall accuracy landed at 64%, only 14 percentage points above a coin flip, and confidence in the guess had no relationship to whether the guess was right. People who felt sure were about as often wrong as people who weren't.
That would be reassuring news for anyone worried about getting caught, except for what the same study found next. Ads that people merely believed were human-made scored significantly higher on every effectiveness measure the researchers tracked, regardless of whether an AI actually made them. Perceived origin had a much larger effect on those scores than actual origin did.
The halo forms even when the detection fails.
This is the part that should change how a small business owner thinks about the question. It isn't about getting away with it. It's that perception carries its own weight, separate from fact, and that weight lands on the ad whether or not anyone can prove where it came from.
NIQ's study measured this two ways at once: what people said on a survey, and what their brains did, using EEG, a brain-activity scanner, on a subset of roughly 150 participants out of more than 2,000 total. People consistently picked up on AI-generated ads at a nonconscious level and rated them more annoying, more boring, and more confusing than traditional ads, a pattern NIQ calls a negative halo. It can bleed past the ad and onto the brand.
The memory data is the sharper finding. Even AI ads that people judged high quality still showed weaker memory encoding in the brain than traditional ads. Advertising only works if it gets remembered; an ad someone liked in the moment but forgot by dinner did its job halfway. NIQ's read is that AI visuals, even good ones, don't always fit the mental shortcuts people's brains already use to file familiar ad patterns, so the ad slides off rather than sticking. A pretty ad that never lodges in memory is a risk you can't spot just by looking at it, which is why Knowing judges an ad against the goal you set for it.
A dentist's office runs an AI-generated image of a smiling family for a check-up promotion. Nobody in the waiting room can prove it's AI. But if the lighting is slightly too even, the hands slightly too smooth, viewers don't consciously flag "AI," they just feel a flicker of unease, rate the ad as less trustworthy, and remember the practice's name less well a week later. The damage happens below the threshold of detection.
Executives think this is a smaller problem than it is.
The gap between what marketers believe and what customers feel is widening, not closing. IAB's 2024 tracking study with Sonata Insights found that 82% of ad executives believe Gen Z and Millennial consumers feel positive about AI-generated ads, while only 45% of those consumers actually do. That 37-point gap grew from 32 points the year before.
It's tempting to assume this softens as people get used to seeing AI ads everywhere. The data says the opposite. Exposure is rising fast, 71% of Gen Z and Millennial consumers now believe they've seen an AI-generated ad, up from 54% a year earlier, and negativity is rising right along with it. The neutral middle, people who hadn't formed a strong opinion, shrank from 34% to 25% as more of them moved into the negative camp. Familiarity is hardening opinions, not softening them, and the split is sharpest among the youngest consumers: 39% of Gen Z report negative feelings toward AI ads versus 20% of Millennials.
Executives keep assuming AI signals innovation. Customers increasingly read it as a shortcut.
The same IAB report found that the share of consumers who'd call a brand "innovative" for using AI dropped from 30% to 23% year over year, even as the share of advertisers who believed AI use signals innovation rose from 40% to 49%. Two groups moving in opposite directions on the same question is not a gap that closes on its own.
Disclosure isn't one decision, it's several.
Most owners think of "should I disclose the AI" as a single yes-or-no question. The research says it splits by what, exactly, the AI touched. Placing an ad means getting it in front of the right people; making it means creating the actual image and words, a distinction most commentary collapses into one. A 2024 study by Wu, Dodoo, and Wen, published in the International Journal of Advertising, ran three experiments testing exactly this split. People were consistently more willing to share an ad they knew was AI-placed than one they knew was AI-made, though only when they already believed AI was capable of handling complicated tasks well. The trust runs through a mental shortcut researchers call the machine heuristic, the idea that people trust machines more for routine, rule-based tasks than for creative ones: people extend more credit to AI doing a mechanical job like ad targeting than to AI doing a creative one like writing the words or making the image.
A second study, by Hien Thu Bui (2025, Journal of Research in Interactive Marketing), used real output from ChatGPT and Stable Diffusion, unlike the hypothetical ad descriptions in the Wu, Dodoo, and Wen experiments, and found that disclosing AI involvement in creation was associated with lower perceived ad value and purchase intent. The mechanism wasn't general dislike of AI, it was a specific drop in how credible the ad seemed once people knew. A person's existing attitude toward AI cushioned the blow somewhat, but didn't erase it.
What actually moves the needle.
Put the two halves of the research together and a practical rule falls out. Detection is close to useless as a defense; a sharp-eyed customer has barely better odds than a coin flip of correctly identifying AI work. But suspicion, even a wrong suspicion, is doing real damage below the surface, on trust, on memory, on whether someone shares the ad with a friend.
NIQ's finding on execution quality points at the lever that's actually in an owner's hands: low-quality AI visuals raise the mental effort it takes to process the ad, which pulls attention away from the message itself. That's not an argument against using AI tools. It's an argument against shipping the rough draft. The fix isn't hiding that a tool was involved, it's making sure whatever the tool produced clears the same bar a human-made ad would have to clear before it goes out.
The shop owner's original question, whether people would notice, was the wrong question all along. The right one is whether the ad gives anyone a reason to feel that flicker of unease in the first place, because that flicker costs money whether or not anyone can name its source.
Frequently asked
Common questions.
Do people actually notice when an ad was made with AI?
Not reliably. A Berkeley study found people correctly identified AI-made ads only 64% of the time, close to a coin flip, and their confidence in the guess didn't predict accuracy. What people believe about an ad's origin, right or wrong, affects their rating far more than the true origin does.
Does disclosing that an ad was made with AI hurt performance?
Research shows disclosure of AI in ad creation lowers perceived credibility and purchase intent, more so than disclosing AI was used for targeting or placement. Per Hien Thu Bui's 2025 research, the negative effect runs through reduced trust in the ad, not general dislike of AI, and a positive existing attitude toward AI only partly offsets it.
Why do AI-generated ads sometimes feel off even when they look fine?
NIQ's EEG research found AI ads produce weaker memory encoding in the brain than traditional ads, even when rated high quality on a survey. The visuals may not fit the mental shortcuts people's brains use to process familiar ad patterns, so the message doesn't stick as well.