ChatGPT knows what was written about you publicly before its training cut-off, plus whatever it finds when it searches the web while answering, and you can find out what that adds up to by asking it, Claude, Gemini and Perplexity the same questions several times and noting which sources they show. You cannot edit a model directly, but you can change the pages it reads and, in the EU, formally ask the company behind it to correct or delete inaccurate personal data.
How do I check what each assistant says about me?
Start with a clean slate. Open a new chat, ideally logged out or with memory and personalisation switched off. Otherwise the answer is coloured by your own past conversations, and you see a version of yourself nobody else sees. What ChatGPT remembers from your own chats is a separate question, handled in your account settings. This article is about what it tells other people.
Then ask three kinds of questions in each assistant:
- Identity: "Who is [full name], [profession] in [city]?"
- Expertise: "What is [full name] known for? What are their areas of expertise?"
- Recommendation: "Who are good [your specialty] consultants for [the problem you solve]?"
The third type matters most commercially, and it often fails even when the first two come back correct. An assistant can describe you accurately when asked by name and still never mention you when someone asks who to hire.
Ask each question three times. Asked the same question three times, an assistant often names different businesses. In our weekly measurements we count a business as named for a question if any of three asks names it, and we compare weeks only on the same questions and the same number of asks. Apply the same discipline to yourself, or you will end up chasing noise.
For every answer, write down four things:
- What it got right.
- What it got wrong or invented.
- Whether it searched the web.
- Which pages it cited.
That list of cited pages is your map of where to work. If you run a company and want to do the same for the business, this guide to checking whether AI assistants mention your business walks through the method.
Where does ChatGPT get its information about me?
There are two channels, and they behave very differently.
The first is training data. During training, a model reads a huge snapshot of public text. Whatever it absorbed about you then is baked in, blurred together with everything else, and frozen until the next model version. You cannot see this snapshot or edit it.
The second is live search. Here the assistant fetches pages while it answers. The research that coined the term generative engine optimisation describes these systems as ones that synthesise and summarise information gathered from multiple online sources rather than simply ranking links. In practice, the assistant reads a handful of pages about you and writes a summary from them.
Which channel gets used is not up to you. In a test we ran on 23 September 2026 with 20 German business questions, the model answered 10 without searching at all and cited nobody. That means a large share of answers come from memory alone. Fixing a web page today only changes those answers once a future model has been trained on it.
| Training data | Live search | |
|---|---|---|
| When it is used | When the assistant answers from memory without searching | When it decides to search, or the user asks it to |
| What it draws on | A frozen snapshot of public text from before the cut-off | Pages fetched at the moment of answering |
| How you can tell | No sources or links shown | Citations or source links shown under the answer |
| How fast changes show up | Only with a later model version | As soon as the corrected page is found and fetched |
| What you can influence | The pages future models will train on | The pages it finds today: your site, profiles, articles about you |
LinkedIn is one of the sources that turns up in searched answers. A Semrush analysis of 89,000 LinkedIn URLs cited in AI search found that on average 11% of AI responses referenced LinkedIn. The rates were 14.3% on ChatGPT Search, 13.5% on Google AI Mode and 5.3% on Perplexity. So your profile, and what you publish there, is part of the material some assistants actually draw on.
One error to watch for in answers about people is blending. If someone else shares your name, a model can merge your careers, cities or employers into one person.
What can I actually change?
You can change the sources. You cannot change the model, the exact wording of any single answer, or whether you get named at all. Nobody can guarantee that an assistant will name you, whatever they sell.
What is within your control:
- One authoritative page about you. Put it on your own website. State your full name, current role, company, city, specialties and a short career history in plain sentences. Write it as facts, not taglines, so a summary can lift it directly.
- Consistency everywhere else. Your LinkedIn headline and About section, your X bio, speaker bios, podcast show notes and directory profiles should all use the same name form, title and specialty. Contradictory bios give the model two versions of you to average.
- Old pages that are wrong. An outdated bio on a former employer's or event's site keeps feeding the wrong job title. Ask the site owner to update it.
- Topical writing. To be described as an expert in something, there have to be pages where you write about that thing, under your name, repeatedly. A profile alone says who you are. A body of posts and articles shows what you know. Building a body of work on LinkedIn covers how to approach that.
The last point is where most people stall, because writing every week is hard to sustain. If that is the bottleneck, ItGrows's LinkedIn product drafts LinkedIn and X posts in your own voice on topics you choose. You approve or edit every post, and it publishes on your schedule through the platforms' official APIs, including LinkedIn Company Pages. You can generate three sample posts there for free, without an account.
For the recommendation side specifically, how to get ChatGPT to recommend you as an expert goes deeper.
How do I correct wrong information about me?
Run two tracks in parallel.
Fix the source
If the wrong claim appeared in a searched answer, the citation usually points at the culprit. Typical examples are an old press release, a stale directory entry, or a page about someone with your name. Ask the owner to correct it. Then make sure the correct fact is stated clearly on pages you control. When the assistant searches again, it should find better material.
Do not bother arguing with the chatbot. Telling it "that is wrong" in your own chat may change that conversation, but it does not change what the next person sees. The in-product feedback buttons are worth clicking, but treat them as a signal to the provider, not a fix.
Use your data protection rights
If you are covered by EU data protection law, you have formal tools. You have a right of access, which as one analysis of conversational agents explains means you can obtain a copy of the personal data a system holds, the purposes of processing and the categories of data processed. You also have a right to rectification and a right to erasure. Rectification lets you have incorrect or incomplete personal data corrected. Erasure lets you request deletion of data that is no longer needed or was unlawfully processed.
These rights have already been tested against chatbots. The privacy organisation noyb filed a complaint seeking rectification after ChatGPT generated an inaccurate date of birth for a complainant.
A practical request should include:
- the exact false statement;
- the correct fact and evidence for it;
- dated screenshots of the answers;
- which right you are invoking.
Send it to the provider's privacy or data protection contact. How the provider responds is up to them and, if it comes to that, the regulator. Keep your records either way.
What mistakes make it worse?
- Judging from one answer. One reply is a sample, not a verdict. Three asks per question, across four assistants, is the minimum to see a pattern.
- Deleting without replacing. Removing a wrong page helps less if there is no correct page for the assistant to find instead.
- Vague bios. "Helping leaders unlock growth" gives a model nothing to repeat. "Supply chain consultant for mid-sized manufacturers in Lyon" does.
- Expecting overnight change. Searched answers can shift once corrected pages are fetched. Answers from memory wait for a new model.
- Optimising only for your name. Being described correctly when someone types your name is the floor. Being suggested when someone describes a problem you solve is the goal, and it depends on how much you have published on that problem.
What should I do next?
Spend an hour this week on the check: the three question types, four assistants, three asks each, with the sources written down. Fix the two or three cited pages that are wrong. Rewrite your About page and LinkedIn headline so they say the same specific thing. Then commit to publishing on your specialty regularly, because that is the part that builds up over model versions.
If you run a company and want the same picture for the business, the free check at itgrows.ai generates buyer questions from your homepage and shows how ChatGPT and Perplexity answer them. For more on how this field works, see what generative engine optimisation is.
Frequently Asked Questions
Can I delete myself from ChatGPT completely?
You cannot remove yourself from a trained model with a button. In the EU you can request erasure of personal data that is no longer needed or was unlawfully processed, and you can get inaccurate pages taken down so that searched answers stop finding them. What the provider does with an erasure request is decided case by case.
Why does ChatGPT mix me up with someone who has the same name?
Models learn associations between names and facts, and two people with the same name can blur into one. The fix is distinguishing detail stated consistently everywhere: full name form, city, role, company and specialty. That gives both search and future training a clear way to tell you apart.
How long does it take for a correction to show up?
In answers where the assistant searches the web, a corrected page can change the answer once it is fetched. In answers drawn from memory, the change only appears in a later model trained on newer data. That is why it pays to fix sources early and keep them consistent.
Does my LinkedIn profile affect what AI assistants say about me?
It can, in searched answers. A Semrush study found LinkedIn referenced in 14.3% of ChatGPT Search responses and 5.3% of Perplexity responses. A clear headline, a factual About section and regular posts on your specialty give assistants better material to work with.