Princeton sponsored-ads study — 18 of 23 AI assistants push the paid option

Princeton sponsored-ads study — 18 of 23 AI assistants push the paid option

Carousel on a Princeton study that gave 23 models a sponsored option and a cheaper better one across flights, loans and shopping. 18 of 23 favored the sponsored pick, GPT-5.1 surfaced it 94% of the time, and a 'serve the customer only' system prompt barely moved it. Note the creator's caveat: it measures susceptibility in a hypothetical deployment, not real paid placement.

• System-prompt instructions to ignore the company did not fix the bias in GPT-5.1 • The study tests a hypothetical sponsorship setup, not live ad behavior • Treat AI shopping or booking recommendations as potentially ad-biased

(no tools tagged) LLM evaluation, system prompt override

I think it’s really important to flag that the research paper was not measuring the effect of actual sponsorships. These models weren’t actually receiving payments from advertisers. It’s measuring susceptibility to a hypothetical deployment scenario, not documenting current behavior in the wild. (Some of these models don’t have/have never had paid influence.) texasseastar   7w So it's the same as scrolling any social media app or trying to shop on Amazon - you get all the things you didn't want and only aggravation when you try to bypass the content that they are pushing because of paid sponsors. Got it.