Why do people recognize AI ads at first glance?
Because our brains are trained on people, not pixels. NielsenIQ (December 2024) showed more than 2,000 people AI-generated ads, from weaker ones to technically top-notch. For about 150 of them, it measured brain activity with EEG. The result: people intuitively identified most of the AI ads and described them as "annoying," "boring" and "confusing." More importantly, even the high-quality AI ads triggered weaker memory activation than traditional ads.
Now think back to the doorman from the article about the gorilla half of people miss. An ad has to climb four floors before anyone remembers it. An AI ad runs out of breath on the second.
NielsenIQ also warns about a negative halo effect: a bad impression of the ad can spill over onto the whole brand. So your ChatGPT poster doesn't just say "this event." It says "this company."
What does it cost you when customers merely suspect AI?
This is where it starts to hurt. Raptive (2025) had 3,000 Americans read similar content, some written by people, some by AI. When readers thought a piece was AI-generated, they rated it:
| Metric | Drop when readers suspect AI |
| Trustworthiness | −48% |
| Authenticity | −57% |
| Emotional connection | −60% |
| Purchase consideration for advertised products | −14% |
| Willingness to pay a premium | −14% |
Notice the word "thought." It doesn't matter whether AI actually wrote it. It only has to feel that way. To be fair: Raptive is an ad company that works with content creators, so it has an interest in showing that human content earns more. But the data lines up with independent research.
Washington State University (Cicek, Gursoy and Lu, 2024) showed more than 1,000 Americans identical product descriptions. The only difference: one mentioned "artificial intelligence." People who saw the term were less likely to buy. Across eight product and service categories. The effect was strongest for expensive, high-risk purchases like electronics, medical devices and financial services. The reason? Emotional trust dropped.
So yes, the aversion I see around me has numbers behind it.
What is insight slop, and why is it worse than an AI poster?
You can spot a ChatGPT poster. You can't spot a ChatGPT insight.
Lore Oxford of Reddit used the term "insight slop" on the WARC Podcast (October 2026). She means research outputs and reports stitched together from buzzwords, trends and data points with no deeper context. They look smart, sound confident and come nicely formatted. And there's nothing sensible you can do with them.
Picture it in practice. You ask AI: "Write me an insight about millennials who buy coffee." Twenty seconds later, you have three paragraphs about authenticity, sustainability and the search for meaning. It sounds like something an agency would bill $10,000 for. But it's the average of everything anyone has ever written about millennials. Not the truth about the people who walk into your coffee shop.
And here's the paradox. The public doesn't trust AI and spots it instantly. Marketers trust it far too much. With a poster, customers at least warn you by staying away. With an insight, nobody warns you. You just spend your budget on a campaign built on an average.
Why do synthetic personas talk like an average?
Synthetic personas are AI "respondents": you tell AI to act like a 35-year-old mom from Ohio, then ask for her opinion. It's fast and cheap, which is exactly why it's tempting.
Researchers at the survey platform Prolific (August 2026) put this to an honest test. They asked 996 real Americans political and consumer questions, then tried to reproduce their answers with AI. What they found:
- Personas built on demographics (age, gender, location) roughly tripled the error compared with simply asking the AI for the overall distribution of answers.
- Personas collapsed into the most common, stereotyped answers. The mom from Ohio answered like a "typical mom," not like a specific person.
- Even richer personas built on interview transcripts didn't beat the simplest approach.
- No method matched the accuracy of a second sample of real people.
The authors recommend AI simulation for directional, exploratory research. Not as a substitute for measuring real people.
And here's what matters for marketing: your business isn't built on the average. It's built on how you differ from it. Ask AI, and you get the average. Ask your customers, and you get the difference.
5 rules: AI as a tool, not the referee
I'm not saying don't use AI. I use it every day, including on this website. I just don't let it make the call. On the WARC Podcast (October 2026), the point came up that strategists get the most out of AI for prototyping and stress-testing ideas. I agree. My rules:
- AI writes the first draft, not the final one. Draft copy, a presentation structure, a source search. The final version always goes through a person who knows who it's for.
- Every insight needs a name. For every claim about your customers, ask: "Which real customer said or did this?" If nobody did, it's a hypothesis, not an insight.
- Ask AI for counterarguments, not confirmation. "Give me three reasons this campaign won't work" is a far more useful prompt than "Tell me why this campaign is great."
- AI visuals are sketches only. For a moodboard or a concept, sure. For a poster that has to sell, use a designer, or at least put in enough work that it doesn't look like ChatGPT's default setting.
- A human signs off on the decision. When a campaign flops, "the AI recommended it" isn't an explanation your CFO will accept. It's an admission that nobody was thinking.
Expert perspective: "AI is the best intern I've ever had. It works day and night, never complains and writes anything in a minute. But you don't let an intern sign off on the budget. And when it tells you it knows your customers, ask how many it has actually met," says Michal Krčmář, founder of Don Marketer and a CMO with nearly two decades in global marketing.
For a hair salon: let AI draft your Instagram post or a text message. But for the poster for your "Ladies' Night" event, hire a designer or photograph your real salon and real clients (with their permission). And when you're wondering whether to add a new treatment, don't ask ChatGPT. Ask ten regulars while they're in the chair. That's your best market research, and it's free.
For a corporation in Miami: use synthetic personas and AI research summaries for quick hypothesis tests and for preparing real research. But only conclusions validated with real people belong in board materials, with the source clearly labeled. Introduce a simple rule: every insight in a presentation gets a tag, either "validated with people" or "AI hypothesis."
The takeaway
People spot an AI poster at first glance, and they'll make you pay for it. Nobody spots an AI insight until the campaign flops. That's why the invisible part of AI slop is the more expensive one. AI is a great tool. Just don't hand it the referee's whistle.
A test for tonight: open your latest presentation or strategy and, next to every claim about your customers, write where it came from. A real conversation, sales data, a survey of real people, or AI? If more than half of the claims have neither a name nor a number, you've got insight slop at home.
Want know-how that came from two decades of practice, not from a prompt? It's in the Don Marketer guides: four PDF guides covering strategy, budget and a secure website. The full bundle is $70.
Sources: NielsenIQ: NIQ Research Uncovers Hidden Consumer Attitudes Toward AI-Generated Ads (December 2024; 2,000+ participants, EEG for ~150); Raptive: The "AI stink" is real, and it's costing brands (2025; 3,000 US adults); Washington State University: Using the term "artificial intelligence" in product descriptions reduces purchase intentions (Cicek, Gursoy, Lu, Journal of Hospitality Marketing & Management, 2024); Research Live: Synthetic respondents can't replace human measurement, finds Prolific study (Gordon et al., "Do Richer Personas Improve LLM Survey Simulation? A Fidelity Paradox," August 2026; 996 respondents); WARC Podcast: "Marketing success rests on better insights" (Oct 22, 2026, Lore Oxford, Reddit) and "Is AI a strategist's friend or foe?" (Oct 6, 2026); the author's own experience.