Researchers at Berkeley built a pipeline to generate ad images with AI, mixed them in with human-made ads, and asked people to sort them. The guessers got it right 64% of the time. A coin flip gets you 50%. That's the entire edge human judgment had over chance, across a study designed specifically to test this.
The more interesting finding sits one layer down. The people who felt confident in their guesses were not more likely to be correct. And the cues people said they were using, like whether an ad seemed 'creative' or 'attractive,' actually made them worse at guessing when they leaned on them. The study, by Huang and Katona, found that confidence levels do not correlate with correctness, and that reported cues such as attractiveness and creativity are often misleading.
Picture the moment this actually happens to a small business owner. You're scrolling past competitor ads, trying to size up whether yours looks as sharp as the restaurant three blocks over, or the salon that always seems to have slicker photos. You're doing exactly what the study's participants did: eyeballing polish and guessing at origin. The Berkeley data says that guess, however confident it feels, is not a reliable signal of anything.
Why looking human mattered more than being human.
The study measured eleven different effectiveness metrics, built on a decades-old advertising framework called the hierarchy of effects — in plain terms, a way of tracking whether an ad gets noticed, gets remembered, and makes someone want the product. Across nine of the eleven, AI-made and human-made ads performed about the same. The two real exceptions were creativity and memorability, where genuinely human-made ads scored a bit higher.
But here's the twist that should reframe how you think about this: ads that were merely perceived as human-made, regardless of whether they actually were, got significantly higher ratings across all the effectiveness metrics. Not just the two where actual origin mattered. All of them. The study found that ads perceived as human-made received significantly higher ratings across all effectiveness metrics, even when the ad was actually AI-generated.
The belief that a person made the ad was doing more work than the actual fact of whether a person made it.
That's a strange result to sit with. It means the label in someone's head, 'a person made this,' was carrying real weight in how they judged quality, separate from anything measurably different in the ad itself. Which tells you something uncomfortable: a lot of what people call good or bad about an ad is a story they're telling themselves about its origin, not a judgment about its structure. So the real question isn't who made the ad, but whether it actually fits what you're selling, which is why Knowing learns your business first and judges the ad second.
The origin question is the wrong question.
If neither professional evaluators nor casual viewers can reliably sort AI-made from human-made, and if the belief about origin swings perception more than the reality does, then asking \"was this made by AI\" is not a useful test before you spend money running it. It tells you nothing about whether the headline is too vague, whether the image gives the eye somewhere to land, or whether you're asking the viewer to hold four ideas in their head when most people manage one or two.
Those are structural questions. They don't care who, or what, made the ad. A vague promise like \"better results\" is vague whether a copywriter typed it or a model generated it. An ad with no clear focal point has no clear focal point either way. The origin of the ad and the soundness of its construction are two separate facts, and the Berkeley data shows people conflate them constantly.
A shop owner reviewing two draft ads, one written by a freelancer and one drafted with an AI tool, might be tempted to trust the human one by default. The research suggests that instinct isn't grounded in anything reliable. What matters is whether either ad actually earns attention, states a specific and checkable claim, and matches what the brand has already told people to expect. Those are things you can check directly, instead of guessing at who, or what, wrote the draft.
The gap nobody has filled yet.
"Was this made by a human or a machine" is a question anyone can ask and feel like they've done due diligence. "Is this ad built to do the job I need it to do" requires an actual framework: a defined read on what grabs attention, what counts as an overpacked claim, what a specific promise looks like versus a vague one. That framework doesn't change based on who or what typed the first draft, and it's a more useful question to ask than who made the ad.
What this means before you hit publish.
None of this means AI tools can write or judge ads unsupervised, and none of it means human judgment is worthless. Genuinely human-made ads did score a bit higher on creativity and memorability in the Berkeley data, which matters if those are the qualities your campaign needs most. But it does mean the question worth asking before you spend money is not who made this ad. It's whether the ad holds up against the specific brand it's for, the specific audience it's trying to reach, and the specific job it's supposed to do. That's a structural check, not a gut check about polish or origin, and it's the one that actually predicts whether the ad works.
Frequently asked
Common questions.
Can people tell if an ad was made by AI?
Not reliably. A Berkeley study by Huang and Katona found people correctly identified an ad's origin only 64% of the time, barely above the 50% you'd get by guessing randomly. Confidence in the guess didn't make it more likely to be right.
Do AI-made ads perform worse than human-made ones?
Across most effectiveness measures tested, no. The Berkeley study found AI-made and human-made ads scored about the same on nine of eleven metrics, with human-made ads scoring somewhat higher only on creativity and memorability.
Does it matter if customers think an ad was made by AI?
It can shift perception even when the ad itself hasn't changed. The study found ads merely believed to be human-made, regardless of actual origin, got rated higher across every effectiveness measure tested, which suggests the belief matters more than the fact.