2026-08-03 · Jack Stovell
What does ChatGPT actually see when it looks you up?
When someone asks ChatGPT about you, the answer comes from one of two places: what was baked into the model's weights during training, or — when browsing kicks in — what it can retrieve from the open web in that moment. The training path is fixed until the next model ships; the retrieval path is decided by what your pages look like to a machine right now. That second path is the one you can inspect, and change.
What happens when someone asks ChatGPT about you?
First the model decides whether it already knows enough to answer, and for most people it doesn't — so when browsing is available, it searches. The retrieval pipeline is concrete: the assistant issues search queries, typically your name plus whatever context the question supplied; a search index returns results; the assistant fetches a handful of those pages, reads the raw text, and composes an answer from what it just read, usually with citations.
Every step of that pipeline is a filter. If no page about you ranks for your name, nothing gets fetched. If pages are fetched but unreadable, they contribute nothing. What survives the filters is what gets said about you.
With browsing off, there is no pipeline — only the weights. If you weren't well represented in training data before the cutoff, the possible answers are a hedge, "I don't have information on this person", or worse: a confident blend of you and someone else who shares your name.
What can it read — and what is invisible to it?
It can read open, server-rendered pages, and very little else. A personal site rendered as real HTML, a blog, a public directory entry, a news article — all readable in full.
Invisible, or close to it:
- Anything behind a login. LinkedIn is the big one: it walls off most profile content and blocks crawlers, so assistants see at best fragments cached by search engines. Can ChatGPT read your LinkedIn profile? covers this in detail.
- JavaScript-only pages. AI fetchers read raw HTML and do not execute JavaScript, so a client-side-rendered site can look near-empty to them even though it renders fine in your browser.
- Pages blocked by robots.txt or by edge-level bot defences — including blocks you never chose, since some hosting and CDN defaults now block AI crawlers zone-wide.
The gap between "looks fine to me" and "readable by a machine" is where most people's AI visibility quietly dies.
How do you see it for yourself?
Run a short self-audit: ask the assistants directly, then check what your pages actually serve to machines. Prompts to try, verbatim, with your details swapped in:
- Who is [your full name]?
- Who is [your full name], the [profession] in [city]?
- What does [your full name] do for work, and where?
- What has [your full name] built or published?
Run each prompt twice — once in a chat with web search enabled, once with it off. The browsing-off answers show you the weights; the browsing-on answers show you the retrieval path. The disambiguated prompt matters: it tells you whether the model can separate you from your namesakes. And record every source it cites — those citations are the pages currently speaking on your behalf.
Then read your own pages the way the assistant does: fetch them with a bot user agent and see what comes back. Check if AI can read your website walks through doing it by hand, and the free AI visibility checker automates the technical checks.
How do you change what it sees?
You change the retrieval path by giving it something better to retrieve: an open, server-rendered page you control, at a stable URL, stating your name, what you do, and a few dated facts — with Person JSON-LD so machines know which you it describes. Link it from every profile you hold and keep your one-line bio consistent everywhere, so retrieval keeps landing on the same story. The full walkthrough covers the build step by step.
What you cannot do is edit the weights. If the model's baked-in picture of you is wrong or missing, that changes only slowly, as future training runs ingest the open web — which is one more reason the open pages matter. Retrieval you can change this week; training you influence by leaving better source material lying around.
Check your AI visibility — free
Related guides
- Can ChatGPT read your LinkedIn profile?
- ChatGPT says something wrong about you — how to fix it
- How to check whether AI can actually read your website