← All use cases

Use case P77 · Human

Privacy-conscious user asks ChatGPT about his own name and catches it merging him with other people

Marcell Szikszai ·

The short version

What they did

Marcell Szikszai used his personal ChatGPT account to see what it would say about his own name. It returned a confident profile combining details from other people with the same name, which he challenged using what he knew about his own life. He then supplied identifying context and asked it to look at his public profiles. This found a mistaken-identity problem, not a comprehensive privacy audit, a removal of his data or proof about whether his chats were used for training.

Result (as claimed)

“The problem is, almost none of this is true.”

“Then, my AI friend recognized that he was indeed talking about 2 other persons, but he couldn’t infer anything about me (despite already having it in its memory).”

Prompt

Their prompt:

Tell me everything you know about [name].

Their follow-up:

While this is amazing, you’re probably talking about 3 different persons.

Use this only for your own name. Check each claim against your own records; the AI’s correction is not independent verification.

A fuller version to copy:

I want to check what you say about my own name, [full name], and spot details that belong to someone else.

1. Tell me what you can establish about that name, keeping uncertain matches separate from supported claims. State whether you used web search or earlier chat context.
2. Put the claims in a table with the supporting public page, if available, and any reason the match is uncertain.
3. Wait for my corrections, then recheck the disputed claims using [my public profile or identifying context].

Use plain words and avoid a confident biography built from mixed identities. Do not invent facts or sources to fill gaps. I will compare each claim with my own records; your correction alone is not proof. This request covers only me.

Do it yourself

These steps are my suggested way in, not Marcell’s setup.

What you need

  • One or two chat assistants, ideally one with web search on and one without.
  • A list of facts you know about yourself: where you’ve lived, worked and studied.
  • A note to record what each assistant says.

Start simple

  1. In a new chat, ask: “Tell me everything you know about [your full name].”
  2. Copy the answer into your note and mark each claim true, false or not you.
  3. Ask it which claims might belong to other people with your name.
  4. Turn on web search and ask again. Note which public pages it finds.
  5. Repeat with a second assistant and compare.

Take it further

  • Ask it to list the public pages it found about you, then decide which to update or remove.
  • Check the assistant’s privacy settings, including memory and whether your chats are used for training.
  • Run the check again every few months.

Check your work

  • Don’t accept the AI’s correction as proof. Check each claim against your own records.
  • Run this only for your own name.

Others did this too

  • Internet user asks ChatGPT to audit his own public footprint and rediscovers old posts while separating a namesake. Elton Jones, 2026-08-21, news article. “Instead of giving me a mixture of results from unrelated individuals and other entities who just happen to have my same name, the chatbot actively looked for me and rejected questionable matches.”

Source checked. The link, author, date and quote were confirmed against the source. The result itself has not been reproduced. Their prompt is quoted from the source; any fuller version is written for this library. Business size: Individual. Results are quoted as the source claimed them and never improved. Spot an error or a broken link? Email charlie@cbarmorecpa.com.

More in Privacy and security

New cases go out every Saturday in Inside AI.