Hundreds of OpenAI contract workers read and rate real ChatGPT conversations

OpenAI works with hundreds of contract workers who read real user conversations to improve ChatGPT's answers, according to a 404 Media investigation based on leaked internal documents and sources. The reviewers rate responses on a scale of one to seven, and part of their task is to reduce excessive flattery and human-like behavior. They are recruited through Crossing Hurdles and paid through Mercor. One reviewer in North America said they earn more than $50 an hour.

The conversations are anonymized but can still contain sensitive personal data, and OpenAI acknowledges that its privacy filter can make mistakes. Some reviewed chats included explicit requests to keep the content private. To opt out, users must turn off the “Improve the model for everyone” setting, which is on by default. Anthropic confirmed it also uses human reviewers, and Google notes in Gemini that humans may review saved chats.