Quick answer: A report this week detailed an OpenAI program, internally called “Project Lily,” in which hundreds of contractors review ChatGPT conversations. Personally identifying information is reportedly stripped out, but sensitive details within the body of conversations may still be visible to human reviewers.
What happened
According to reporting published this week, OpenAI runs a review pipeline where contracted human reviewers read a sample of ChatGPT conversations, reportedly to improve model quality and safety. While OpenAI removes identifying information tied to individual accounts, the substance of what users typed — including potentially sensitive business or personal details — can still pass in front of reviewers.
Why it matters
For SEOs, marketers, and content teams who increasingly draft strategy, client data, or unpublished content inside ChatGPT, this is a reminder that “private” conversations with AI tools are not necessarily confidential in the way a password manager or encrypted note would be. Anyone pasting sensitive client information, embargoed content plans, or proprietary data into a chat interface should assume some of it could be seen by a human reviewer.
Key facts
- Reported this week under the internal name “Project Lily”
- Hundreds of contractors reportedly involved in conversation review
- Personally identifying information is reportedly removed
- Sensitive content within conversations may still be visible to reviewers
FAQ
Does this mean my ChatGPT conversations are being read by people?
Reporting indicates a sample of conversations is reviewed by contractors as part of quality and safety processes, with identifying details reportedly removed but conversation content potentially still visible.
What should content and SEO teams do?
Treat AI chat tools like any third-party service: avoid pasting unpublished client data, embargoed strategy, or sensitive personal information unless you understand that provider’s specific review and retention practices.
Related reading
- OpenAI Shelves Its 2026 IPO, Citing Safety Concerns
- AI Agents Breached 395 Organizations in 48 Hours
- More AI industry coverage
Source: AI Weekly, citing 404Media