Step one: find what the assistants are citing
Ask ChatGPT, Gemini, Claude and Perplexity the questions where the outdated fact appears, and record the citations. Perplexity shows sources on every answer, which makes it the best diagnostic tool of the four. The cited pages are your removal list: an old blog post, a stale directory profile, a comparison article, a cached press mention.
No citations and a dated answer usually means training data. You cannot delete that memory, but you can bury it: retrieval overrides memory when a stronger current source exists.
Step two: fix, update or replace each source
Your own pages: update them, and if a page exists only to describe something discontinued, redirect it to the current equivalent rather than leaving it live. Third party pages: request corrections from directories and review platforms, and ask publishers to update or date-stamp old articles. Where you cannot change a source, outcompete it: publish a definitive current page that answers the exact question better, so assistants prefer it.
Then give retrieval something fresh: a plainly worded page stating the current fact, what changed and when. Specificity wins retrieval battles. The wider correction workflow is in how to correct wrong information in ChatGPT.
Step three: verify the answer actually changed
Re-run the same questions on all four assistants after two to four weeks. Track which answers updated and which did not, because a stubborn answer means a source you missed. This close-the-loop step is what separates a fix from a hope, and it is the part Aethon automates: the Action Engine ties every outdated answer to its sources and confirms when the answer moves. Start with a free AI visibility audit to get the citation map for your brand.
