ChatGPT still answering from an old preference? Open Sources under the reply to find and fix the stale memory

#What changed

ChatGPT now shows memory sources under personalized replies on the web. When a reply draws on personal context, a Sources (book) icon appears below it and lists what fed the answer: saved memories, past chats and custom instructions, and on Plus and Pro also files in your Library and emails from a connected Gmail. From each source you can correct it, delete it, or mark it as not relevant. MarMar Labs pointed out a practical use: if you moved a project from Python to TypeScript and ChatGPT keeps giving Python examples, Sources usually shows the old chat or memory behind it.

#Who is affected

Anyone who uses ChatGPT with Memory turned on, on any consumer plan. It is on the web first; OpenAI says mobile is coming later.

#What to do now

1. When a reply assumes something that is no longer true, open Sources below it before re-prompting.
2. If a saved memory is the culprit, open its three-dot menu to correct or delete it.
3. Or just state the change in chat, for example: "This project is TypeScript now, not Python. Remember that."
4. Deleting the old chat does not remove a separately saved memory, so clean up both if needed.
5. For a one-off question that should ignore your history, use a Temporary Chat.

#What is not confirmed

OpenAI says Sources may not show every factor behind a response, and the available controls vary by account and plan. There is no date yet for mobile. Sources are not shown in chats you share.