How to Fix What ChatGPT Says About You
ChatGPT answers questions about you two different ways — from frozen training data, or by browsing the live web ("ChatGPT Search") — and fixing the pages it cites in that second mode is the lever we actually control.
OpenAI also offers an official personal-data removal request, useful but not guaranteed. Here's what really changes what ChatGPT says, and what doesn't.
Quick overview
- ChatGPT answers from frozen training data or live web browsing — which one matters a lot — here's the difference.
- Fixing the sources ChatGPT Search cites is the lever we actually control, plus OpenAI's own official request as backup — here's the exact process.
- OpenAI's own policy says removal rights "are not absolute" — it may decline a request outright — here's what that means.
- OpenAI doesn't publish a processing-time SLA for its privacy request — and approval itself isn't guaranteed either. More on what to expect.
How ChatGPT actually generates answers about you
ChatGPT answers in two different modes, and OpenAI's own documentation confirms both.
In one, it draws purely on training data — a frozen snapshot of information up to a knowledge cutoff date, "memorized" into the model itself. In the other ("ChatGPT Search"), it automatically searches the live web when a question might benefit from current information, then summarizes what it finds with citations.
This matters directly for fixing a bad answer: correcting a website can plausibly change a browsing-mode answer, since it's pulling from current pages — but it does nothing for a training-mode answer, since that's baked into the model's frozen weights until OpenAI releases a new version on its own schedule.
How to fix what it says about you
The most reliable lever is fixing what ChatGPT Search can actually cite — OpenAI's own privacy request is worth filing too, but it's discretionary and OpenAI can decline it.
- 1
Audit what ChatGPT currently says
Ask ChatGPT directly about your name or business and note whether it cites sources or links (browsing mode) or gives a confident answer with no citations (training-data memory).
- 2
Fix the underlying web sources
For browsing-mode answers, correcting or suppressing the source content ChatGPT is likely pulling from is the lever we actually control — the same source work behind our other suppression services.
- 3
File the official OpenAI privacy request
Submit a "remove my personal data from ChatGPT responses" request at privacy.openai.com, or email [email protected], as a secondary layer, for information that's inaccurate, excessive, irrelevant, or no longer appropriate.
- 4
Report it in-product as a backup
Use the thumbs-down button, select "Safety or Legal concern," and describe the issue. OpenAI states reported domains "may be reviewed" and filtered from future responses — discretionary, with no confirmation of action.
If it doesn't stick
OpenAI's own privacy policy is direct about the limits here: erasure and objection rights "are not absolute and we may decline a request," and even a granted removal "does not remove it from external websites or search engines" — it only affects what ChatGPT itself outputs.
In practice, that means a single privacy request or a single source correction isn't something to treat as a permanent fix — especially for training-memory answers, which can only really change when OpenAI releases a new model version, on a schedule it doesn't publish.
Common questions
Does OpenAI have a way to just delete what ChatGPT says about me?
There's a real request category at privacy.openai.com — "remove my personal data from ChatGPT responses" — for information that's inaccurate, excessive, irrelevant, or no longer appropriate. But OpenAI's own policy states these rights "are not absolute and we may decline a request," and a granted request only affects ChatGPT's responses, not the open web or search engines.
Can I stop ChatGPT from ever mentioning me?
No — there's no blanket opt-out from being discussed. The available official request, filed at privacy.openai.com or by emailing [email protected], is narrower: asking OpenAI to stop specific inaccurate or inappropriate information from appearing, not to exclude a name from the model entirely. OpenAI's own policy notes it may decline the request outright.
Does correcting a website actually change what ChatGPT says?
For browsing-mode answers, yes — that's the more reliable lever, since it doesn't depend on OpenAI's discretion the way its official privacy request does. ChatGPT sometimes browses the live web ("ChatGPT Search") and sometimes answers purely from frozen training data; source corrections can affect browsing-mode answers, but don't retroactively edit what a released model "remembers" from training.
Could someone else's ChatGPT "remember" things about me from conversations where they mentioned me?
Potentially, yes — ChatGPT's memory feature lets it retain context across a specific user's own conversations, separate entirely from what a released model learned during training. OpenAI's own guidance on fully clearing something from a person's ChatGPT experience says every place it could be stored has to be addressed individually — past chats, archived chats, uploaded files, the memory summary, and any connected apps — deleting just one doesn't clear the rest.
Related
Also want to know what Gemini or Perplexity say about you, or what shows in Google's AI Overviews, Microsoft's Copilot, or Grok? See our Gemini, Perplexity, AI Overviews, Copilot, or Grok correction guide, or the full AI Reputation Repair service.
We fix the source pages ChatGPT Search can actually cite — the lever we control — and file the official OpenAI privacy request as a second layer, not the whole plan, then keep checking as the model updates.
Start today
See what ChatGPT says about you.
We'll audit it, fix what's fixable at the source, and file the official OpenAI request as backup.
Get my confidential assessment