CartographContact

Guides · updated 2026-08-24

How do I remove my personal information from ChatGPT and other AI models?

You cannot erase what a model learned in training, and no provider offers that. What you can do is two things that work: file the personal-data removal request each provider publishes, and remove the web pages the model reads when it answers, because these models search the web at the moment they are asked.

The part that is impossible, and why

Personal information absorbed during training is spread across the model's weights. Removing it means retraining, and no provider offers that to an individual. This is known as the machine unlearning problem and it is unsolved. Any service telling you it deletes you from an AI model is overstating what it can do.

OpenAI is explicit about the limit of its own tool: removing personal data from ChatGPT responses does not remove that material from external websites or search engines.

What actually works, one: file the provider requests

  • OpenAI takes a request to remove your personal data from ChatGPT responses, through its privacy portal.
  • Google runs Results About You, which finds your contact details in Search and lets you request removal. It was extended in February 2026 to cover government ID numbers.
  • Perplexity has a GDPR data privacy form, and it explicitly accepts a request made by a legal agent on your behalf.

What actually works, two: remove the pages it reads

This is the part most people miss. When you ask a current model about a person, it searches the web and answers from the pages it finds. It is not reciting a memory. It is reading, now.

So the pages are the lever. Remove or de-index the page, and the claim that came from it stops appearing. In our own testing the pages the models cited about a real person were mostly professional profile aggregators that scrape career information, not the people-search sites that removal services usually work through.

When the model is wrong rather than intrusive

A common case is not exposure but error: the model states an old job, an old company, an old city. Removal does not fix that. Deleting the stale page makes the model say nothing, or find the next stale page saying the same thing.

The fix for a wrong claim is to publish a correct one that a crawler will read: clean markup, structured data, and a plain text version. That is the opposite of removal, and it is the only thing that works on a false statement.

How to check whether it worked

Ask the same question again. But be careful: these models reword themselves on every run. In a controlled test, two identical scans of the same person minutes apart, with nothing removed in between, produced 13 apparently new claims and 11 apparently deleted ones. A single before and after proves nothing. You need several runs on each side, and a check that the page itself is actually gone.

Cartograph does exactly that. See how.

Common questions

Can ChatGPT actually forget me?
Not in the sense people mean. Information absorbed during training is distributed across the model's weights and cannot be removed without retraining, which no provider offers to an individual. OpenAI does accept a request to stop your personal data appearing in ChatGPT responses, and states plainly that this does not remove the material from external websites or search engines.
Then why does removing a web page change the answer?
Because current models search the web at the moment they are asked and build the answer from what they find. They are reading, not reciting. Take the page away and the claim that came from it stops appearing.
How do I submit a request to each provider?
OpenAI takes personal data removal requests through its privacy portal. Google runs Results About You, which finds your contact details in Search and lets you request removal, extended in February 2026 to cover government ID numbers. Perplexity has a GDPR data privacy form, and it explicitly accepts a request made by a legal agent on your behalf.
What if the AI says something about me that is simply wrong?
Removal is the wrong tool. Deleting the stale page makes the model say nothing, or move to the next stale page saying the same thing. The fix for a wrong claim is publishing a correct one that a crawler will read: clean markup, structured data and a plain text version.
How do I check whether anything changed?
Ask again, several times, on each side. These models reword themselves constantly. In a controlled test, two identical scans of the same person minutes apart with nothing removed in between produced 13 apparently new claims and 11 apparently deleted ones. A single before-and-after proves nothing at all.
Does deleting my ChatGPT account remove me from its answers?
No. Deleting your account removes your account and your chat history. It has no effect on what the model says about you when someone else asks, because that comes from public web pages and training data, not from your account.
Which pages do the models actually cite about a person?
In our own testing, mostly professional profile aggregators that scrape career information, plus the person's own sites. Notably not the people-search sites that removal services usually work through.

Sources

Every figure on this page is either taken from a source above on the date shown, or measured by Cartograph on 2026-08-24. If something here is out of date, write to support@usecartograph.com.