OpenAI Pays People $50 an Hour to Read Your ChatGPT Chats

OpenAI is paying outside contractors more than $50 an hour to read selected ChatGPT conversations, according to internal documents reported by 404 Media. The project, reportedly called "Project Lily," has people scoring ChatGPT's answers from 1 to 7, flagging robotic phrasing and even too many green checkmark emojis. It's part of how OpenAI tries to make the chatbot sound less like a machine and more like a helpful person.

Here's the catch for you. Before chats reach reviewers, OpenAI runs them through an automated "Privacy Filter" that removes names and identifying details. But the company admits the filter can miss unusual names or vague clues. Reviewers may also see a "user memories summary" showing your past interests and approximate location. One contractor told 404 Media: "No, I don't think they would imagine some contractor somewhere is analyzing the conversations."

The defaults matter most. On Free, Plus and Pro accounts, the "improve the model for everyone" setting is switched on automatically. Business, Enterprise and Edu accounts are opted out by default — a big reason companies worry about employees using personal ChatGPT for confidential work. Turning the setting off only affects future chats; it doesn't pull back what's already been reviewed.

OpenAI isn't alone — Anthropic and Google Gemini also use human reviewers for chat improvement. The practical takeaway: think of ChatGPT as a helpful stranger, not a locked diary. If you wouldn't say it to a coworker you barely know, don't type it there.