Why So Many AI Ads Look Fake, and What Actually Fixes It
In a 2026 test of 20 ads shown to 3,000 U.S. consumers, only 25% of the people who watched an AI-made ad were even somewhat confident it was AI, and 40% of all viewers were unsure either way. Yet the human-made originals were 14% stronger on short-term effectiveness and 17% stronger on long-term brand effectiveness (Ipsos, 2026).
That's the puzzle behind “AI ads look fake.” Most viewers can't reliably spot the AI, yet the work still lands weaker. Here's my reading of the research: AI ads usually feel fake not because viewers spot glitches, but because the work feels flat, and because people discount it once they suspect AI was involved.
A fairness note on that test: each AI ad was built entirely by AI. Google's Gemini reverse-engineered a brief from the original ad and wrote a shot list, and OpenAI's Sora 2 turned that into video, with no human creative intervention (Ipsos, 2026). So it measured what AI produces on its own, which is a different question from what it produces with a person steering. One more limit: these were 30-second video ads from well-known brands, not still images, so the image advice later in this piece is my extension of the research, not something the test measured.
The Glitches Are Fading. The Gap Isn't
In late 2024, NielsenIQ found viewers spotted most AI-generated video ads and rated them more annoying, boring, and confusing than traditional ads (NIQ, 2024). By 2026, most Ipsos viewers weren't confident they could tell. The studies were built differently, so don't read them as a head-to-head comparison, but the direction is clear: the visible glitches are fading as the tools improve. Ipsos itself says it would be naive to assume the gap won't keep narrowing.
For now, feelings about AI in ads are hardening. In a 2026 survey by The Harris Poll with the 4As and Infillion, 78% of respondents said AI makes ads feel less authentic, 73% said they'd be less likely to trust an ad they suspected was AI-made, and 63% said they'd be less likely to buy from brands that lean on AI (The Harris Poll, 2026). Those are stated opinions, not sales data, and the published summary doesn't give the sample size.
Two Different Problems
First, even when viewers don't know an ad is AI, AI work lands weaker. When Ipsos asked people to compare the ads directly, they found the AI versions less creative, less emotionally engaging, and less informative. Even without spotting the AI, viewers rated the human-made ads as more eye-catching and imaginative (Ipsos, 2026). Credible, in Ipsos's words, is not the same as compelling.
Second, once people know or suspect AI was involved, they discount the work. A 2026 study in the Journal of Consumer Research analyzed more than 1 million TikTok posts and ran eight preregistered experiments. Posts carrying an AI disclosure got roughly 7 to 8% fewer likes and about 7% less engagement overall. The researchers found the drop was driven in part by perceived effort: people assume the creator cared less, and they feel less connected. It did not come from concerns about quality, wariness of artificial content, or a general dislike of AI (Carney, Riveros, and Tully, 2026).
A separate 2026 study in the Journal of Business Research found the same pattern in ads: AI disclosure made the ad look less effortful and the product look lower in quality (Patil and Rice, 2026).
Together, that's a useful way to read “looks fake.” Sometimes the work really is flatter. Sometimes it's the suspicion that nobody spent real time on it.
Where AI Does Fine, and Where It Struggles
Ipsos found a pattern worth knowing. AI performed best when the brief was straightforward and product-driven. It struggled when the creative job called for storytelling, emotion, or a genuine point of view. AI chose a storytelling approach in only 30% of its ads, compared with 49% of ads in Ipsos's wider research. The most effective ad in the whole test was human-made and built on storytelling (Ipsos, 2026).
A strong brief narrowed the gap. The Cheerios pair came from the most human brief in the test, built on a clear, empathetic idea about being there for the people who need you. Both the human and AI versions posted the highest combined effectiveness of the 10 brands tested. Ipsos's takeaway: there's no substitute for a strong brief, whoever or whatever makes the ad (Ipsos, 2026).
For a local business, here's my read: that sorts your images into two piles. Simple, product-style jobs, like a clean promo graphic or tidying a background, are fair territory for AI. Anything that carries trust, like your team, your finished work, and your customers, is where AI struggles and where your real photos already do the job. One caution: none of this research tested a local business's own posts, so treat it as a guide, not a guarantee.
A Five-Step Image Check
Before any AI-made image or ad goes out, run it through these steps. They're my own recommendations, built from the research above, not a tested formula.
Start with one true thing. Before you write a prompt, write down the single human truth the image should carry, like why a customer calls you or what a finished job feels like. A strong starting idea is what narrowed the gap in Ipsos's test.
Keep real people and real work real. Use real photos for your team, your finished jobs, and your customers. That's where AI struggled most and where your own photos carry the effort for you.
Give AI the simple jobs. Backgrounds, clean-ups, promo graphics, and variations of a photo you already have. Ipsos's own advice is to lean on AI for product work, variations, and scale, and keep people at the center of storytelling.
Look at it like a customer. Does it look like someone took care, or like it took thirty seconds? Fix or drop anything that reads as effortless.
Follow each platform's labeling rules, and watch your audience. Early research suggests wording may matter, since disclosures that signal more effort softened the penalty in one study, but the evidence is thin, so there's no safe wording to copy. See how your own customers respond.
How I Make Our Blog Images
I use AI for the images on our own blog, and I hold them to a short set of rules. Each image gets its own composition built around the article's central idea, not a screenshot or a literal interface. Brand colors show up as small accents, not a wash over the whole picture. I don't ask the generator to render text or numbers. And every image comes with an explicit list of things to avoid, like stock-photo clichés and handshakes. That isn't proof my images perform better. It's the discipline I use so they don't read as an afterthought.
One Thing to Check Right Now
Open the last AI-made image you posted. Ask one question: if a customer saw this next to a real photo of your work, which one looks like you took care? If it's the AI image, replace it with the real photo this week.
Frequently Asked Questions
Can people actually tell when an ad was made with AI?
Often not with confidence. In Ipsos's 2026 test of 20 ads, only 25% of viewers who watched an AI-made ad were even somewhat confident it was AI, and 40% of all viewers were unsure. A Syracuse University summary of the same study reports a lower figure, 13%, so the two reports disagree on the exact number. Syracuse notes that its 13% matched the share of viewers who thought human-made ads were AI-made, so viewers weren't reliably telling the two apart. In a late-2024 study, viewers spotted most AI ads. The tools have improved and the studies differ, so treat all of these as snapshots.
Do AI ads work as well as human-made ads?
In one 2026 test, human-made ads were 14% stronger on short-term effectiveness and 17% stronger on long-term brand effectiveness. The AI ads were built entirely by AI from a brief reverse-engineered from each original, with no human creative intervention. AI did best on simple, product-driven briefs and struggled with storytelling and emotion. It's one study, so treat it as a strong signal, not a law.
Should I tell people when an image is AI-made?
Follow each platform's rules on labeling. The research on how a label affects people is still thin. One study found AI disclosure lowers engagement because people assume less effort, and the same study found that disclosures signaling more effort can soften the drop. That is not enough to hand you safe wording. The more dependable move is keeping AI away from the parts of your business where trust is the whole point, like your team and your finished work.
Is it okay for a small business to use AI images at all?
Yes, for the right jobs. The research points to where AI struggles, not to a ban. Use it for simple product-style visuals and keep your real photos for the people and work that earn trust.
Here's the Plan
You don't need to avoid AI images. You need your real work to show up where trust is on the line.
If you'd rather have someone set up your social posting so AI handles the simple jobs and your real work stays front and center: the research keeps pointing to the starting idea and the visible effort, not just the tool. Book a free, zero-pressure strategy session with me today. If you want us to build it for you, great. If not, you walk away with a free roadmap.
This piece is part of our AI series, “AI Isn't the Problem. How You Use It Is.” Start with the overview: The Marble and the Chisel: Why Bad AI Content Usually Isn't AI's Fault.
