What Does ChatGPT Say About You?

A few years ago, the first thing a prospective client, investor, reporter, or date did was Google you. Today, a growing share of them ask an AI assistant instead. They type your name into ChatGPT and read whatever comes back as if it were settled fact. Most of the time you will never know it happened, and you will certainly never see the answer.

That is the quiet problem. By late 2025 ChatGPT had crossed 800 million weekly active users, and OpenAI’s own study of how people use the product found that “seeking information” had grown to roughly a quarter of all use: quick facts, research, comparisons. When the subject of that research is a person or a company, the assistant produces a confident summary on demand. Whether that summary is accurate is another matter.

How ChatGPT decides what to say about you

It helps to understand that a language model is not looking you up in a database. At its core it is a prediction engine that generates the most statistically likely next words based on patterns in the enormous body of text it was trained on. When that text contains a lot about you, the model has more to draw on. When it contains little, or contains contradictory information, the model fills the gap with its best guess. That guess can be wrong while sounding completely authoritative.

There are broadly two ways an answer about you gets assembled.

From training data (the model’s memory)

The base model learned from a massive snapshot of the public web, books, and other text gathered up to a cutoff date. Anything written about you before that cutoff may be baked into the model’s parameters. This is why answers can be stale, blending an old job title, an outdated company, or a years-old news story into a description of who you are today. It is also why the model can invent details. A 2025 analysis cited by Axios found that leading models still hallucinate in a meaningful share of responses when they lack solid grounding, producing fluent text that is simply not true.

From live web retrieval

Since ChatGPT search launched broadly in late 2024, the assistant can also pull current information from the web and cite its sources, with clickable links and a “Sources” panel below the answer. This grounds the response in documents the system fetches in the moment, and it helps a great deal. But it does not fix everything. The model can still misread a source, merge two different people who share a name, or lean on a low-quality page that happens to rank well. Pulling from live pages makes answers less likely to be invented. It does not make them true.

The practical takeaway: what ChatGPT says about you is a reflection of the public record it can see, filtered through a system that prizes a confident, fluent answer over an honest “I don’t know.” If the record is thin, dated, or polluted, the answer will be too.

This is not theoretical. It has already gone badly wrong.

Three documented cases show how far an AI answer can drift from reality, and how high the stakes get when it does.

In Australia, regional mayor Brian Hood discovered that ChatGPT was telling users he had been imprisoned for bribery. The truth was the opposite: he was the whistleblower who exposed the scandal and was never charged with anything. His lawyers prepared what would have been one of the first defamation actions against an AI company.

That same week, George Washington University law professor Jonathan Turley learned that ChatGPT had named him on a list of legal scholars accused of sexual harassment, citing a Washington Post article as evidence. The article did not exist. Neither did the trip the alleged incident supposedly happened on. The model had fabricated the accusation and the source to support it.

In the United States, radio host Mark Walters sued OpenAI after ChatGPT falsely described him as having been accused of embezzling money from an organization he had no involvement with. In May 2025 a Georgia court granted OpenAI summary judgment. Part of the reasoning: a reasonable reader, warned by the tool’s own disclaimers, would not take the output as a literal statement of fact. Sit with that. One of the first legal answers to “the AI defamed me” rests on the idea that you should not have trusted the AI in the first place. That is thin protection when the person reading your bio is a board member deciding whether to back you.

Why this matters more than a bad Google result

A misleading search result sits on page one where you can see it, dispute it, and watch it. A misleading AI answer is private, frictionless, and trusted. The person asking gets a clean paragraph with no competing viewpoints, often forms a judgment in seconds, and moves on. You get no notification and no right of reply.

Public concern tracks this reality. Pew Research found that 76% of Americans worry about AI producing false or misleading information, and that more adults now feel concerned than excited about AI in daily life. Yet they keep using it, and they keep trusting the answers. That combination, high reliance paired with low accuracy, is exactly where reputational damage hides.

How to check what ChatGPT says about you

You can do a basic self-audit in an afternoon. Treat it as reconnaissance, not a fix.

• Ask directly, and ask sideways. Try “Who is [your name]?” but also “Is [your name] trustworthy?”, “What controversies is [your name] associated with?”, and “Should I do business with [your company]?” The framing changes the answer.
• Use more than one assistant. Run the same prompts through ChatGPT, Gemini, Perplexity, and Claude. They draw on different sources and will not agree. The disagreements are where the risk lives.
• Test with and without web access. A model answering from memory versus one retrieving live sources can describe you very differently. Both are seen by real people.
• Check the citations. When sources are shown, click them. Note which pages the models treat as authoritative about you. Those pages are now load-bearing for your reputation.
• Look for confident fiction. The dangerous errors are not the obvious ones. They are the plausible, specific, false details delivered without hesitation.

Document everything. Screenshots with dates matter, because these answers change and you will want a record of what was said and when.

What you can actually do if the answer is wrong

Here is the hard truth: you cannot log in and edit the model. There is no settings page for your reputation. The instinct to demand a correction from the AI company rarely produces a fast or durable result, and as the Walters case shows, the legal path is uncertain and slow.

What does work is changing the inputs the models read. AI answers are downstream of the public record. When the authoritative, accurate, well-structured information about you is abundant and the inaccurate material is corrected, addressed, or outweighed, the answers shift in your favor over time. This is patient, technical work that sits at the intersection of reputation management and AI visibility, and it is closely related to the discipline we cover in GEO vs SEO and the broader practice of online reputation management. It is also continuous. Models update, the web changes, and an answer that is clean today can degrade next quarter, which is why monitoring is part of the job rather than a one-time check.

More and more, the first impression of you is written by a machine, in a conversation you never see. The question worth asking is simple. Have you given that machine accurate, authoritative material to work from, or are you leaving a summary that strangers act on to chance?

That question sits at the heart of our Search & AI Visibility practice. We study what the major assistants currently say about a client, identify the gaps and inaccuracies driving it, and work to strengthen the record those systems rely on, so the answer people get lines up with the reality of who you are. If you have never checked what the machines say about you, start there. Then come talk to us about what to do with what you find.

Sources

• Maxwell Zeff, “Sam Altman says ChatGPT has hit 800M weekly active users,” TechCrunch. https://techcrunch.com/2025/10/06/sam-altman-says-chatgpt-has-hit-800m-weekly-active-users/
• OpenAI, “How people are using ChatGPT.” https://openai.com/index/how-people-are-using-chatgpt/
• Megan Morrone and Scott Rosenberg, “Why hallucinations in ChatGPT, Claude, Gemini still plague AI,” Axios. https://www.axios.com/2025/06/04/fixing-ai-hallucinations
• OpenAI, “Introducing ChatGPT search.” https://openai.com/index/introducing-chatgpt-search/
• Pranshu Verma and Will Oremus, “ChatGPT invented a sexual harassment scandal and named a real law prof as the accused,” The Washington Post. https://www.washingtonpost.com/technology/2023/04/05/chatgpt-lies/
• Pranshu Verma, “An Australian mayor was accused of bribery. He blames ChatGPT,” The Washington Post. https://www.washingtonpost.com/technology/2023/04/06/chatgpt-australia-mayor-lawsuit-lies/
• Loeb & Loeb LLP, “Walters v. OpenAI, L.L.C.” https://www.loeb.com/en/insights/publications/2025/05/walters-v-openai-llc
• Pew Research Center, “Key findings about how Americans view artificial intelligence.” https://www.pewresearch.org/short-reads/2026/03/12/key-findings-about-how-americans-view-artificial-intelligence/