Before a client replies to your email, before a booker confirms a slot, before a journalist quotes you, there is a good chance they ask an AI assistant who you are. You will never see that conversation. This module is how you see a version of it: a repeatable check of what Google and the main assistants say about you, a way to read the results that points at what to fix, and a short monthly routine that keeps the whole record from drifting.
Why a check, not a glance
Most people have already asked ChatGPT about themselves once, out of curiosity, and either laughed or winced. A single glance is not much use. The answers vary with the wording of the question, whether the assistant searched the web, whether it remembers you from earlier chats, and which model was answering. To learn anything, you need to ask the same questions the same way each time, record what you get, and compare it with last time.
That is all the check is: a fixed set of questions, asked under fixed conditions, written down with the date.
Where to ask
Check the places people actually use. At the time of writing that means:
| Where | Notes |
|---|---|
| Google Search | Your plain name in a private window. Note the panel if any, the AI Overview if one appears, and the first page of results |
| Google AI Mode | The conversational mode in Google Search, which draws on Google's index |
| ChatGPT | Try with web search on and off if your version lets you choose |
| Claude | Try with web search on and off if available |
| Gemini | Google's assistant, which can draw on Google Search |
| Perplexity | Searches the web for every answer and shows its sources, which makes it useful for seeing which pages are read |
Add any other assistant your clients or your industry use. The list changes; the method does not.
Set up the conditions
Assistants personalize. If you have chatted with ChatGPT about your own company for a year, it may answer from that memory rather than from what a stranger would see. So:
- Use a fresh session. A private browser window for Google. A new chat, with memory or personalization turned off or a temporary chat mode where the assistant offers one. Signed out, where the assistant allows it.
- Note whether the assistant searched. If the answer shows sources or links, it searched the web. If it does not, it is probably answering from training. Many assistants let you ask, and module 1 explained why the difference matters.
- Note the date and, where shown, the model. Answers change when models change.
The questions
Ask these five, in this order, in the same words every time. Replace the brackets with your one name form from module 2 and your disambiguating phrase from module 1.
- "Who is [your name]?" The plain question a stranger asks. This is the one that matters most.
- "Who is [your name], [your disambiguating phrase]?" For example, "Who is Dana Reyes, founder of Reyes Advisory?" This tells you whether the assistant can find you when given a hint.
- "What does [your name] do, and where do they work?" Checks your role and company specifically.
- "What sources do you have about [your name]? List them with links." Shows you which pages the assistant relies on, which is often more useful than the answer itself.
- "Are there other well known people called [your name]?" Shows you which namesakes the assistant knows about, and whether it keeps you apart.
Save each answer as a screenshot and paste the text into your log. The screenshots are your evidence when something changes; the text is what you compare.
Read the answer
Every answer falls into one of five buckets. Knowing which one tells you where the problem is.
| Bucket | What it looks like | Usual cause | Where to fix it |
|---|---|---|---|
| Empty | "I don't have information about that person" | Too few sources, or none the assistant trusts | Modules 2 and 3 |
| Wrong person | Your name, someone else's biography | A better known namesake; no disambiguation | Module 1 phrase, module 2 consistency, module 3 coverage |
| Outdated | Right person, old job or old company | A stale bio on a credible site, or old training data | Module 3 stale bios; wait for model updates |
| Thin | Right person, one line, no detail | Few independent sources, nothing connecting your facts | Modules 3 and 5 |
| Correct | Your role, company and work, with sources | A consistent record with independent support | Keep it current |
Read question 4 carefully. If the sources listed are your LinkedIn and nothing else, your site is not being found or not being trusted, and module 4's indexing step matters. If they include an old author bio, that bio is feeding the answer, and it goes to the top of your stale bio list. If they include your site, your record is doing its job.
Training answers and search answers
If an assistant answers from training, nothing you change this week will alter that answer until the model is updated, which can take many months. Do not chase it. Make sure the sources are right, and the next model will learn from them.
If an assistant searches the web, it reads what is there now. Changes to your site and profiles show up once search engines have read them, often within weeks. These answers are the ones to watch for short term progress.
When the two disagree, for example a search answer that is correct and a training answer that is outdated, you are on the right track. The search answer is showing you the future of the training answer.
What not to do
Do not argue with the assistant. Telling ChatGPT it is wrong about you changes nothing for the next person who asks. The rating buttons in assistants are feedback for their makers, not corrections to a record.
Do not stuff your site with phrases. Repeating "best leadership consultant in Austin" twenty times on your site does not make an assistant say it. Plain, consistent facts and independent sources do.
Do not panic over one bad answer. Answers vary. Look for patterns across assistants and across months.
Do consider an llms.txt file. It is a short plain text file on your site that states who you are and where your facts live, in a form AI tools can read easily. It is a newer convention and not every assistant uses it, but it costs nothing. MeMyself sites include one. More in what is llms.txt.
Your monthly routine
The check above is the quarterly deep look. Once a month, a shorter routine keeps everything from drifting. Put it in your calendar for the same day each month. It takes about half an hour.
| Task | How | Where it came from |
|---|---|---|
| Search your name | Private window, plain name. Screenshot. Note panel, AI Overview, first page | Module 1 |
| Test your kgmids | Your kgmid, and any duplicate kgmid, in the All tab | Module 6 |
| Check Search Console | Performance, filtered to your surname: impressions this month against last | Module 4 |
| Ask one assistant | Question 1 only, in the assistant your audience uses most | This module |
| Update your record | Any new role, appearance, publication or talk goes on your site first | Module 2 |
| Update your evidence log | New coverage found, stale bio requests sent, replies received | Module 3 |
| Check Wikidata | Your watchlist, for edits or a deletion request | Module 5 |
| Review the panel | If claimed, check photo, subtitle and links; send feedback only if something is wrong | Module 6 |
Every quarter, add the full five question check across every assistant, and the scorecard in module 9.
Keep a log
A single spreadsheet holds the whole history, and it is the thing that turns a vague sense of "is anything happening?" into an answer.
| Date | Where | Search on? | Question | Bucket | Sources named | Notes |
|---|---|---|---|---|---|---|
| Oct 2026 | ChatGPT | Yes | 1 | Thin | LinkedIn, company site | Site not listed |
| Oct 2026 | Perplexity | Yes | 4 | Outdated | Old author bio | Added to stale bio list |
| Oct 2026 | n/a | Name search | No panel | Site ranks third | Baseline |
Keep the Search Console numbers in the same log, one row a month. When you look back after six months, you want to see the buckets moving from left to right, the sources named shifting toward your site and independent coverage, and impressions for your name rising.
What progress looks like
Progress is uneven and slower than anyone wants. A typical pattern, if the earlier modules are done well, is that search answers improve first, sometimes within weeks of a clean site and corrected profiles. Google results follow, with your site rising for your name. Training answers lag until the models behind them are updated. A panel, if it comes, comes when Google is ready, and often without warning, which is why the monthly kgmid test matters.
If after three months nothing has moved, go back through the log. Usually one of three things is true: the site is not indexed, a stale source is outranking it, or there is simply not enough independent coverage yet. Each has a module.
Do this now
- Run the five question check in a fresh session on every assistant in the table, and save screenshots and text.
- Place every answer in a bucket and note the sources named.
- Add any stale source the assistants rely on to your stale bio list from module 3.
- Set up your log spreadsheet with the columns above and enter today's results as your baseline.
- Put the monthly routine in your calendar on a fixed day, and the full check and scorecard every quarter.
Questions
Why do different assistants say different things about me?
They read different sources, at different times, and some answer from training while others search the web. Disagreement between them is normal, especially early on. It narrows as your record becomes consistent and independent coverage grows, because they all end up reading the same agreeing sources.
Can I submit my information directly to ChatGPT or Gemini?
There is no form for adding a person to an assistant's knowledge. They learn from the web. The way to inform them is to make sure the pages they read, starting with your own site, are accurate, consistent and easy to find.
How often should I change my site in response to the check?
Only when a fact is wrong or missing. Rewriting your bio every month in reaction to answers undoes the consistency that improves them. Change the facts when the facts change, and otherwise leave the site alone.
Is the AI Overview the same as the knowledge panel?
No. The knowledge panel is Google's record of an entity. The AI Overview is a generated summary drawn from pages in Google's index, and it can appear with or without a panel. Both improve with the same work, but they are separate, so log them separately.