OpenAI Contractors Read ChatGPT Chats to Improve Responses

An investigation by 404 Media has revealed that OpenAI employs hundreds of contract workers to read and evaluate real ChatGPT user conversations. These reviewers rate the chatbot's responses on a scale of one to seven, with the specific goal of reducing what internal documents describe as "excessive flattery" and human-like behavior in the model's output. The contractors are recruited through a firm called Crossing Hurdles and paid through the AI training company Mercor, with at least one North America-based reviewer earning more than $50 an hour.

Although OpenAI anonymizes these prompts and utilizes a privacy filter, the system is not foolproof, and sensitive personal data can still slip through. Internal documents show that many users are entirely unaware of this human oversight, with some explicitly asking the chatbot to keep their highly sensitive conversations private. OpenAI currently discloses this practice in a buried FAQ page, noting that authorized personnel and service providers may view data to improve performance, but critics argue this disclosure is too obscure to constitute informed consent.

To prevent their conversations from being read by humans, users must manually disable the "Improve the model for everyone" setting, which is turned on by default and only applies to future chats. Alternatively, they can use ChatGPT's temporary chat mode. This reliance on human feedback is not unique to OpenAI; Anthropic similarly employs human reviewers to analyze Claude conversations when users opt in, and Google utilizes human reviewers to evaluate saved Gemini chats.

For AI practitioners and enterprise developers, this revelation underscores the critical importance of data governance when interacting with commercial LLMs. Relying on default privacy settings can expose proprietary code, corporate strategies, or customer data to third-party human contractors. Teams must actively manage their API and UI settings, opting out of model training programs or utilizing zero-data retention deployments to ensure strict compliance and protect intellectual property.