Skip to content
Free website build, and your domain name free for a yearCode COASTGUARD26 through Aug 31Start free
HarborlightWeb Co.

AI search

Has ChatGPT actually read your website?

It is a knowable thing, and almost nobody is told the answer. Your server sees every one of these visits.

When an assistant reads a web page it identifies itself, the same way Google’s crawler has for twenty years. It says which company it belongs to and, usefully, why it has come.

That second part is the one worth understanding, because the two reasons are worth very different things to you.

Indexing, and answering

An indexing visit is a crawler building its picture of the web. It is reading you so that months from now, when somebody asks a question you can answer, you are a candidate. This is the groundwork, and it is what makes you eligible to be quoted at all.

A live visit is different, and it is the one that should make you sit up. Somebody is in a conversation with an assistant right now, they asked something, and the assistant has come to your page to answer them. It is the closest thing that exists to watching yourself be looked up.

The names to look for

If you can reach your server logs, these are what to search for.

It calls itselfWho it isWhat it means
GPTBotOpenAIIndexing. Reading your pages so ChatGPT can draw on them later.
OAI-SearchBotOpenAIIndexing, for the search side of ChatGPT.
ChatGPT-UserOpenAILive. Somebody asked ChatGPT a question and it came to read your page to answer them.
ClaudeBotAnthropicIndexing.
Claude-UserAnthropicLive. A person is in a conversation with Claude right now.
PerplexityBotPerplexityIndexing.
Perplexity-UserPerplexityLive, on behalf of somebody asking.
Google-ExtendedGoogleIndexing, for Gemini and AI answers.
Applebot-ExtendedAppleIndexing, for Apple Intelligence.

Why your analytics shows none of this

Nearly every website statistics tool works by running a small piece of JavaScript in your visitor’s browser. That is how it knows somebody arrived, how long they stayed and what they clicked.

An assistant runs no JavaScript. It asks the server for the page, reads the text, and goes. There is no browser, so the counter never fires, and your dashboard says nothing happened.

Counting these means counting on the server, where the request actually arrives. That is what we do on every site we run, and it is why we can tell you which assistant, which page, and whether it was indexing or answering somebody live.

A user agent is a claim, not proof

Anyone can send a request calling itself GPTBot. It takes one line on a laptop. A number built from those claims without checking them is a number you should not act on.

The check is the network address. Each of these companies publishes the ranges its crawlers come from, and a read from outside them is a claim rather than a fact. We show those as unconfirmed instead of throwing them away, because a genuine crawler on a newly published address is not a liar either.

What nobody can tell you

Whether the assistant then recommended you. Those conversations are private and no one outside those companies can see inside them.

Reading is not citing. If somebody offers to report how many times ChatGPT recommended your business, they are estimating, and you should ask them how before you pay for it.

Questions people ask

How do I know if ChatGPT has read my website?
Your server sees it happen. When an assistant reads a page it identifies itself in the request, usually as GPTBot, or as ChatGPT-User when somebody is asking a question at that moment. If you can get at your server logs, search them for those names. If you are on a hosted builder, you probably cannot, and that is the honest answer.
Why does my analytics not show this?
Because almost every analytics tool counts visitors by running JavaScript in a browser, and an assistant runs none. It fetches the page and leaves. As far as a browser-based tool is concerned, nothing happened at all.
Does an assistant reading my site mean it is recommending me?
No, and this is the line worth holding on to. Reading is not citing. Nobody outside those companies can see what an assistant said about you in a private conversation. Any tool claiming to report how often you were recommended is estimating, and you should treat the number accordingly.
Can somebody fake being ChatGPT?
Easily. A user agent is a claim, not proof; anyone can send that name from a laptop. The check is the network address the request came from, against the ranges the companies publish. We show unverified reads as unverified rather than dropping them or dressing them up.
Should I block AI crawlers?
If your business is being found by customers, blocking them is closing your own door. There is a real argument for blocking on the other side of the fence, where a publisher's articles are the product being consumed. A restaurant is not in that position: you want to be quoted.

Related: why an assistant might not be mentioning you and what llms.txt is.

We show you every one of these.

Which assistant, which page, and whether somebody was asking. On every site we run.

Start your free website

Would rather talk to someone first? Ask us to come out