Human Bias Against AI Is Limiting Its Potential for Brands and Publishers

For the last few years, there has been a lot of debate about whether AI can actually produce advertising as good as humans can. New research suggests we may be focusing on the wrong part of the question.

A new working paper, AI in Disguise, from researchers at Columbia, Harvard, the Technical University of Munich and Carnegie Mellon studied AI-generated advertising running on Taboola. The researchers created a quasi-experimental comparison of 4,633 "sibling ads," matching AI and human-created images from the same advertisers running in identical campaign settings at the same time. Those ads accounted for more than 369 million impressions and 2.5 million clicks.

They found no detectable average click-through-rate disadvantage for AI-generated images compared with their human-made siblings. The much more interesting finding was what happened when the AI creative actually looked like AI.

AI-generated ads performed better than human-created ads when people did not perceive them as AI. As the researchers put it, the average performance of AI creative masks an important condition: whether people perceive the image as artificial.

That distinction matters. If the actual origin of the creative isn't driving the difference in performance, but the perceived origin is, AI may already be better at producing effective advertising than we are giving it credit for. Our reaction to AI could be limiting some of the benefit.

What does "AI creative" even mean anymore?

A large share of the AI-generated ads in the study were not perceived as AI, which makes the way we typically categorize AI creative increasingly problematic.

We tend to define creative based on how it was produced. Consumers obviously don't have that information when an ad appears on a screen. They are making their own judgment based on what it looks and feels like.

The research found some patterns in those judgments. Sharper and clearer images with larger faces were more likely to be perceived as human-made, even though AI itself tended to produce those characteristics. Intense color saturation, lower warmth and some of the aesthetic characteristics people have come to associate with AI made an image more likely to be identified as AI-generated.

That means two ads can both be generated by AI and get very different reactions. One can have the characteristics people associate with AI and perform worse, while another can be perceived as human and perform as well as or better than human-created advertising.

Putting both into a single "AI creative" bucket misses that difference.

There is also evidence that the reaction goes beyond the creative itself. New research from Rival Technologies, conducted among 901 Gen Z consumers in the U.S. and Canada in July, found that 74% reacted negatively when they realized a brand's marketing was made with AI. Only 8% reacted positively. More concerning for brands, 72% said they had taken some action against a brand over AI marketing, including 43% who said they had stopped buying from one.

This becomes more important as brands produce more advertising with generative AI. The economics are compelling. AI makes it possible to produce more creative, create more variations, personalize those variations for different audiences and do it all much faster.

There has been an assumption that creative quality will be one of the major limitations on that growth. These findings suggest consumer perception may be just as important. If AI can produce effective advertising but some of that effectiveness disappears when consumers recognize its origin, then the way people perceive AI becomes part of understanding its performance.

We should measure the perception, too

This is where I think the findings become particularly interesting from a measurement perspective.

A straightforward test of AI versus human creative wouldn't necessarily separate these effects. We could see that one execution performs better than another, but we wouldn't know how much of that difference came from the creative itself and how much came from people's perception of how it was made.

It isn't particularly difficult to add that dimension to creative testing. Show someone an ad without telling them how it was produced and ask what they think. Was it made by a person, by AI, by some combination of the two, or can they not tell?

You can then look at perceived origin alongside the effectiveness measures you already collect. That creates some interesting questions for brands. Does knowing AI was used change someone's response to the same creative? Does perceived AI affect clicks differently than brand favorability or trust? Are certain audiences or categories more sensitive to it? What happens when AI-generated advertising is explicitly labeled? And do those effects change as people become more accustomed to AI-generated content?

There is also a real opportunity here for publishers. Individual advertisers can test their own creative, but publishers see advertising across many brands, categories and audiences. They are in a position to understand these patterns at a much broader level and potentially help advertisers understand where AI creative is working, where it isn't, and why.

I don't think the takeaway from this research is that consumers simply don't like AI-generated advertising. The advertising study itself points to something more nuanced. AI-generated creative can perform extremely well when consumers don't perceive it as AI.

The more interesting implication is that the technology and our perception of the technology may be developing at different speeds. AI's ability to make effective advertising is improving quickly, while people are developing their own assumptions about what AI-generated content looks like and how they feel about it.

As brands and publishers put more AI-generated creative into the market, understanding that gap is going to matter. If perception is affecting performance, we need to treat it as part of what we measure.

Sources

• Exner, Hartmann, Ding, Zhang & Netzer, AI in Disguise: Quasi-Experimental Analysis of a Large-Scale Deployment of AI-Generated Ads, Columbia Business School Research Paper, revised August 12, 2026.
• Columbia Business School, AI in Disguise: How AI-generated ads' visual cues shape consumer perception and performance, August 26, 2026.
• Rival Technologies, New Research: Gen Z Attitudes on AI Use in Marketing and Advertising, August 13, 2026.