Consumers Could Not Distinguish AI Ads From Human Ones. The Sales Data Could. The Gap Is The Product.
Advertising professor Carrie Riby spent three years studying whether consumers can identify AI-generated advertisements. They cannot. The study effectively maps which AI-assisted creative tasks are commercially viable right now: product descriptions, feature-benefit copy, A/B test variations, localized campaign versions, and high-volume performance marketing creative.
This illustrates a principle I like to call perceptual equivalence. When output quality is indistinguishable to the end consumer, the only remaining question is production cost. AI does not need to be superior. It needs to be indistinguishable and cheaper. That gap between perceived quality and production efficiency is where margin lives. The strategic lesson: do not ask whether AI is good enough. Ask whether your customer can tell. If they cannot, you are overpaying for human labor.
Professor Carrie Riby conducted the three-year study. The findings outline specific categories where AI-assisted work is currently sellable with confidence, including direct-response variations and localized campaign versions for performance marketing.
- Pick one product you know well and write a 50-word feature-benefit ad description yourself. Note the time it took.
- Ask ChatGPT or Claude to write the same description with the prompt: "Write a 50-word product description highlighting key features and customer benefits for [your product]."
- Show both versions to three friends without labeling which is which. Ask them to guess which is AI-generated. If they struggle to tell, you have just demonstrated perceptual equivalence. That is your business case for using AI in that specific task category.