How ChatGPT decides what to say about you
AI assistants answer questions about you from a small set of sources. Here is how those sources get chosen, and why one negative page can define the whole answer.
Someone is asking an AI about you
The background check has changed. Instead of typing your name into Google and reading ten links, people now ask ChatGPT, Perplexity, Gemini, or Copilot a single question: "Who is [your name]?" or "Is [your name] trustworthy?"
The answer comes back as a confident paragraph. If a lawsuit, a complaint page, a suspension notice, or an unflattering news story is sitting near the top of your name results, that paragraph will repeat it — usually without context and often without a date.
Where the answer actually comes from
Assistants blend two inputs:
- What the model absorbed during training, which is a stale snapshot of the public web.
- What it retrieves live at the moment of the question, which is heavily weighted toward the top-ranking pages for your name.
The second input is the one that moves. When the first page of results for your name is dominated by a single negative item, that item becomes the source the AI leans on. When the first page is full of substantive, current, accurate pages you control or are credited in, the negative page loses its share of the answer.
Why the answer can be wrong
Two failure modes show up constantly:
- **Stale facts.** A complaint that was withdrawn, a case that was dismissed, or a role you left years ago still reads as present tense.
- **Confusion with someone else.** Common names get merged. An AI will happily attach another person's regulatory history to you if the surrounding text is ambiguous.
Both are fixable, and neither gets fixed by ignoring it.
What to do about it
Start by seeing the actual answer, in writing, across engines. You cannot argue with a summary you have not read.
Then work the sources rather than the summary. Suppression is the mechanism: build and strengthen accurate pages about you until they outrank the negative item, so both search engines and AI assistants have better material to pull from.
Mirror.fyi runs the scan, scores what it finds, and prepares the work plan. You publish the assets and submit the links back so progress can be verified.