Write down a fixed list of questions about yourself, ask each one in ChatGPT, Claude, Gemini and Perplexity in a clean session three times, and record whether you were named, what was said and which sources were cited. Then repeat the same list on the same terms every month or so, because a single answer is a sample, and only the same questions asked the same way show whether anything has changed.
Typing your own name into ChatGPT once tells you little: the answer depends on the assistant, on whether it searched the web, on what it remembers about you and on chance. The method below removes what noise it can and records the rest.
Which questions should you ask about yourself?
Ask three kinds of questions, because each one tests something different.
- Identity questions contain your name and one detail that sets you apart: "Who is Jane Doe, the pricing consultant in Hamburg?" They show whether the assistant can tell you from namesakes and what it believes about your role, your company and your history.
- Expertise questions connect your name to your field: "What is Jane Doe known for in SaaS pricing?" They show whether the assistant links you to the topics you want to be known for.
- Discovery questions do not contain your name at all: "Who are good pricing consultants for B2B SaaS companies in Germany?" or "Which experts write about usage-based pricing?" These are the questions your future clients actually type, and they are the hard test. Being described correctly when someone asks for you by name is not the same as being named when they don't.
Write 10 to 20 questions, mostly discovery questions, phrased the way a client would ask rather than the way you describe yourself. Then freeze the wording: a changed word can change the answer, and you are no longer comparing like with like.
For what makes an assistant pick one expert over another, read How to Get ChatGPT to Recommend You as an Expert.
How do you set up a clean test in each assistant?
The assistant you use every day knows things about you that a stranger's assistant does not. Claude, for example, saves memory as you chat, with memory on by default on the Free, Pro and Max plans; its incognito chats are not saved to memory or chat history, and memory can be switched off for a single chat only before you send the first message. A test inside your own account can make you look better known than you are, because the assistant may draw on your earlier conversations rather than the public web. Look for the equivalent settings in every assistant you test.
For each assistant, then:
- Use a private browser window, signed out where the assistant allows it, or its temporary or incognito chat when you are signed in.
- Open a new chat for every question. A follow-up in the same chat inherits everything said before it.
- Keep the same settings for every run: account state, mode, web search on or off.
- Note whether the answer came with sources. In Gemini, a Sources button appears at the bottom of a response or inline when sources are available, and a response without it had no links provided. An answer with sources gives you pages to check; an answer without them gives you nothing to check, and it may rest on what the model learned in training.
Why do answers change between asks?
Three things move the answer even when you change nothing.
Sampling. A language model writes by choosing among likely next words, and the choice has randomness in it. Anthropic's glossary says that even with temperature set to 0, results are not fully deterministic, and identical inputs might produce different outputs across API calls. You cannot switch that off in a chat app, so you measure around it.
What was retrieved. When an assistant searches, the answer is built from whatever pages came back for that query at that moment. The same glossary describes retrieval-augmented generation as data retrieved at runtime and passed to the model along with the question. A different set of pages means a different answer.
Context. Memory, being signed in or out and the model version can all change the answer, and models are updated over time, so keep dated records.
So one ask proves nothing. In our own measurements, asked the same question three times, an assistant often names different businesses. We count a business as named for a question if any of three asks names it, and compare checks only on the same questions and the same number of asks. Use the same rule for yourself: ask each question three times, count yourself as named if any of the three asks names you, and never compare a month with three asks against a month with one.
What should you record each time?
Use a spreadsheet with one row per ask. These are the columns that turn a pile of screenshots into something you can compare:
- Date of the check, for example 2 November 2026, and the assistant with the model or mode it showed.
- Settings: signed in or out, memory off or incognito, web search on or off.
- The exact question, copied from your frozen list, and the ask number (1, 2 or 3).
- Named: yes or no. If yes, your position in any list and the words used to describe you.
- What it said about you, pasted verbatim, with every error marked: a wrong employer, an outdated title, a mix-up with a namesake, an invented credential.
- Who else was named for the same question.
- Every cited URL, and which of them are yours: your site, your LinkedIn profile, an interview, a directory, a conference page.
At the end of each round, summarise per assistant: the share of questions where any of the three asks named you, the errors that keep coming back, and the domains cited most often. The cited domains are the most useful column: they show which pages the assistants read about you and your field, and that is where any change has to start.
Where do the assistants get what they say about you?
When they search, they read public pages. OpenAI says its OAI-SearchBot is used to surface websites in ChatGPT's search features, and sites opted out of it will not be shown in ChatGPT search answers, apart from navigational links; a separate agent, ChatGPT-User, handles actions a person starts in ChatGPT, such as visiting a page while answering their question. If your personal site turns up in no ChatGPT answer at all, ask whoever runs it whether it blocks these agents, and whether that is deliberate.
For most experts, LinkedIn is the other page that matters. LinkedIn says some sections of your profile may appear off LinkedIn when people search for you in tools like Google or Bing, and you decide which sections through your public profile setting. Whether a particular assistant reads your profile is not something to assume: the cited URLs in your spreadsheet will show you, assistant by assistant.
Interviews, podcast pages, guest articles, speaker bios and old employer pages turn up too. A stale conference bio can be the page behind a wrong title.
What should you do when an answer is wrong or you are missing?
Fix the sources, not the chat. Correcting an assistant inside your own conversation does not rewrite the pages it reads when someone else asks about you.
- Wrong facts: find the cited page that carries them. Update what you control, such as your site's about page and your LinkedIn profile, and ask the owners of the other pages to correct theirs.
- Confusion with a namesake: make your own pages unambiguous. Use the same full name, role, city and field on your site and every profile, so each page clearly points to the same person.
- Missing from discovery questions: publish on the exact topics in your list, regularly and under your own name, and check which domains the assistants cite for those topics. A body of work on LinkedIn built around a few topics gives a search more to find on those topics than scattered posts do. If the hard part is keeping that up, our separate LinkedIn and X product drafts posts in your own voice on topics you choose, and you approve or edit every one before it goes out.
Then run the same list again after a few weeks, with the same settings. Nobody can make an assistant name a particular person, so judge the work by the trend in your records, not by any single answer.
How often should you re-check?
Once a month is a workable rhythm, more often while you are changing pages. Keep every round in one sheet, never edit old rows, and compare only rounds with the same questions, assistants, settings and number of asks.
If you run a company, the same logic applies to your business, with buyer questions in place of questions about you. ItGrows has a free check that generates five buyer questions from your homepage and gets answers from ChatGPT and Perplexity in 25 to 40 seconds; one answer shows right away, and all five open once you create an account, which costs nothing. The business side of the question is covered in How to Check If ChatGPT, Claude, Gemini and Perplexity Name You.
Frequently Asked Questions
Can I just ask ChatGPT what it knows about me?
You can, but treat the answer as one data point, since a single ask inside your own account mixes in memory and chance. Ask in a clean session, three times, and add discovery questions that do not contain your name.
Why does Claude say something different about me than ChatGPT?
They are different models that search and cite in different ways, so they can read different pages about you. That is why the method records each assistant separately and lists the sources each one cites.
How many times should I ask the same question?
Three times per question and per assistant is a workable minimum. Count yourself as named if any of the three asks names you, and compare only rounds with the same questions and the same number of asks.
Can I get an assistant to correct a wrong statement about me?
Not by telling it in your own chat. Find the pages it cites, correct the ones you control and ask the owners of the others, then run the same questions again after a few weeks.