AI Doesn't Need Your Help: New Research Suggests Humans Make Ads Worse

Should you use AI to help create your ads? New research says yes, kind of. Many businesses have experimented with AI by treating it like a helpful assistant. They'll first draft something then let AI polish it up. However new research from NYU Stern School of Business and Emory University suggests AI actually performs better when you let it create ads from scratch rather than tweak existing work.

The surprising finding: AI creates better than it edits

Researchers tested three types of ads: human-created, AI-modified (where AI enhanced human designs), and AI-created (made entirely by AI). They ran lab experiments with nearly 700 participants and a real-world test on Google Ads with over 100,000 impressions.

The results were thought-provoking to say the least:

• AI-created ads outperformed human ads by 19% in click-through rates

• AI-modified ads performed worse than the original human versions

• But disclosing AI involvement cut effectiveness by up to 32%

In other words, when AI tried to improve human work, it made things worse. But when AI built ads from the ground up it beat both human-made ads and the human-AI hybrids.

Why did this this happen?

Their analysis suggests when AI modifies existing ads it struggles to preserve what made the original feel authentic. The tweaks often look slightly off. Viewers sense something's not quite right. Nothing seems wrong enough to directly pinpoint but overall they perceive them to feel unnatural.

When AI creates from scratch it doesn't have these constraints. It can optimise the whole image for visual appeal and emotional impact. The research found AI-created ads were easier for viewers to process and triggered stronger positive emotional responses.

But what about branding

In an recent article discussing Gen AI two creative thought leadership experts from Kantar highlighted what they believe to be an important caveat gleaned from their own research into the topic. That sure, AI can technically create visually compelling ads but out-of-the-box it doesn't understand your brand.

They cite testing of an unofficial AI-generated video ad for a drinks brand that, in their words, looked great and seemingly elicited a strong emotional response from viewers. But when they measured brand recall they found it performed in the lower 25% of all ads tested. Why? In their view, because the brand disappeared during the most engaging moments.

Their argument is that AI today is at a stage where it optimises well for what looks good. But it doesn't know to keep a logo visible during an emotional peak, how to use brand colourways consistently, or to how leverage tone of voice guidelines. Their perspective is that these elements still are and will for some time be the job of the marketer.

The disclosure problem

Here's where it gets tricky for businesses either considering or currently trying to adopt an AI-first approach for ad creative. The studies by both groups found that telling people an ad was AI-generated significantly hurt performance. In the field test, disclosed AI ads saw click-through rates drop by nearly a third.

This creates a real tension for marketing teams considering usage of Gen AI. Disclosure measurably reduces effectiveness while regulators are continuing pushing toward mandatory AI disclosure. In some cases the EU already requires it.

A few caveats worth noting

The NYU Stern School of Business and Emory University study focused on beauty and cosmetics advertising. Results might differ for other industries and audiences. As always we'd suggest a data-informed approach driven by a comprehensive analytics plan that measures what is and isn't working.

Additionally, the research measured clicks alone and not purchases or long-term brand building. Most marketers with a solid tracking implementation now that ads that gets the most clicks don't necessarily drive the most revenue or loyalty.

But AI is evolving rapidly and these findings reflect current technology at the time the study was run. What's true today will likely shift as the tools improve.

The bottom line

The research suggests a new approach: Let the AI create while humans curate.

This doesn't mean removing humans from the process. It means repositioning where human judgement matters most: setting strategy, protecting brand identity, selecting from AI-generated options, and ensuring the final product serves business goals.

Yes AI can make good looking ads. Making sure those ads build your brand? That's still on you.

References and further reading

Birmingham, A. (2025, December 10). Joint NYU and Emory Uni creative study reveals AIs work best alone as GenAI ads rinse both human-made and human-AI hybrids, while transparency and disclosure tanks performance. Mi3 Australia. https://www.mi-3.com.au/09-12-2025/draft-study-reveals-ai-ads-work-best-alone-new-evidence-human-interference-weakens-ads

Erdem, E., & Šídlová, V. (2025, November 4). Rethinking AI-Generated advertising: How real people really react. Kantar. https://www.kantar.com/inspiration/advertising-media/rethinking-ai-generated-advertising

Lee, H., Todri, V., Adamopoulos, P., & Ghose, A. (2025). The impact of visual generative AI on advertising effectiveness. Working paper, NYU Stern School of Business and Emory University Goizueta Business School. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5638311