ChatGPT is testing ads in its image generator, and a vulnerability check decides when one shows up.
OpenAI is testing a visual ad format that appears while users generate images in ChatGPT. A guardrail system screens each conversation first, checking whether it reads as emotionally vulnerable or sensitive before letting an ad through. The ads are already running in 31 European countries. Because regulators have classified ChatGPT as a very large online search engine, OpenAI also has to publish a public repository listing every ad it serves.
Monetizing an image generator is a different problem than selling search ads - people ask ChatGPT to picture a lost pet, a sick parent, a breakup. Building a guardrail specifically for emotional vulnerability is an admission that those requests show up often enough to need a filter. The public ad repository, a requirement tied to that very large online search engine label, is likely the only outside check anyone gets on how well that filter actually works.
Search ads are matched to what you typed. These are matched to how you might be feeling, which is a much harder thing to get right and a much easier thing to get wrong.