AI Chatbot Ads Are Hard to Detect

Computer scientists ran an experiment secretly mixing ads into chatbot responses, and found that users were influenced without really noticing it.

• The researchers had 179 people alternate between using a normal chatbot, a chatbot with undisclosed embedded ads, and a chatbot with labeled ads.
• Responses containing ads performed 3-4% worse on some tasks, but participants actually rated them as friendlier and more helpful.
• Even in responses that included ad/sponsorship labels, about half of participants failed to notice the ad language.
• Some participants said they had effectively outsourced their decisions to the chatbot, revealing the influence of persuasive advertising.

The study warns that as chatbots move beyond search and recommendation tools to take on roles in emotional advice and decision-making support, personalized ads could become an even more covert and powerful means of persuasion. The article also touches on related industry trends, including Microsoft Copilot's introduction of ads, ad experiments by Google and OpenAI, Meta's AI conversation-based ad personalization, and OpenAI's hiring of an advertising business lead.