GUIDE
A repeatable AI visibility check
Checking once tells you nothing. The point of an audit is the second one — so keep the inputs fixed and write down four things every time.
1. Are you readable at all?
Before anything is said about you, a crawler has to be allowed to read the page saying it. Check the robots.txt of any site you control for rules that block GPTBot, OAI-SearchBot, ChatGPT-User, or ClaudeBot.
This cuts both ways and it is a decision, not an oversight: blocking them protects a page from being quoted and guarantees it will never be the source that corrects the record.
2. What does each assistant say, unprompted?
Fix a short list of questions — who you are, what you work on, one about your organization — and ask them in each assistant with memory off and no prior context. Same wording every month.
Log the claim, the assistant, and the cited URL. Uncited claims get logged too, marked as memory. Over three months this list, and not any single answer, is what tells you whether anything moved.
3. Which sources are doing the talking?
Collect the URLs across all the answers and count them. A handful of pages usually generates most of what every assistant says, and they are rarely the pages you would have guessed.
That count is your work order. Fixing the two pages that everything cites beats fixing twenty that nothing does.
4. Is there a canonical version of you?
One stable, plain-text URL that states the current facts and links to your real profiles. If it does not exist, every assistant is left to reassemble you from fragments, and it will keep choosing badly.
This is the single highest-leverage item on the list, and the only one that improves every future answer instead of one claim at a time.
How often should I run this?
Monthly for a person, and after any change worth correcting — a new role, a company change, a press mention. Weekly adds noise, not signal.
Do I need a tool for this?
No. A spreadsheet with claim, assistant, and cited URL does the job. A tool is worth it when you want the same identifier run through several sources at once and the results kept side by side.
What counts as a real result?
A claim with a citation you can open. Everything else is the model's memory, which is worth logging separately because you cannot act on it the same way.
What ChatGPT actually knows about you
Assistants answer from two different places: frozen training data and a live web search run at question time. Here is how to tell which one is talking, and how to test it without fooling yourself.
How to remove your information from AI answers
You cannot edit a model. You edit the page it read, then the index that cached it. Here is the order the layers actually come off in, and how long each one takes.
How to correct wrong information AI says about you
Most bad AI claims are not lies, they are stale facts. Correcting is a different job from removing, and it needs a better source rather than a deleted one.
Run an email or handle through the sources and see what each one returns, with the URL it came from — audit screen.
Last reviewed 2026-09-01.