Does ChatGPT read your website?

robots.txt guidance

If you want ChatGPT-related systems to reach public content, name GPTBot, OAI-SearchBot and ChatGPT-User explicitly. Private paths stay blocked. You can allow search and live fetches and block only GPTBot. For ChatGPT-User, robots.txt is not a reliable blocking tool; hard blocks need server or firewall rules.

OpenAI starter rules

User-agent: GPTBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: OAI-SearchBot
Allow: /

Sitemap: https://www.example.com/sitemap.xml

The three routes into a ChatGPT answer

First through training: your page was in the data the model learned from. You have no influence over that today, and no way to check it. Second through search: OAI-SearchBot holds the page in an index that ChatGPT draws on when needed. Third through a live fetch: someone asks something and ChatGPT retrieves the page at that moment. Only the last two routes can be influenced — and that is exactly where it is decided whether your company is named as it is today or described from an outdated training snapshot.

What ChatGPT actually sees when it fetches

A fetch is not a browser visit. No JavaScript runs, no mouse moves, no accordion opens. What is in the source counts; what a script loads afterwards usually is not there. Text inside images is not there either — a price on a graphic simply does not exist for an AI system. You can check this without any tool: open the page, switch JavaScript off in the browser, reload. Whatever is still readable is roughly what arrives.

Six reasons a page does not show up

A Disallow in robots.txt, often left over from an old test phase. A firewall or bot protection that rejects unknown agents wholesale — then the fetch never even arrives. Content only visible after a login or a consent dialog. A page whose actual statement sits inside an image. A very long page where the important details are at the bottom and get cut when it is truncated. And the most common case of all: the question the customer asks is answered nowhere on the site as a question.

What you can check in half an hour

Open your robots.txt in the browser and read whether GPTBot, ChatGPT-User and OAI-SearchBot are allowed there. Switch JavaScript off and check whether your core statement is still on the page. On your most important page, look for the three questions customers ask on the phone — are they there as questions, with an answer underneath? Check your server logs for the three agents. And ask ChatGPT itself about your company: what comes back is the most honest baseline you can get.

What can be measured — and what cannot

Access can be measured: which agent fetched which page when, and with what result. What cannot be measured is whether that turned into a mention. OpenAI publishes no statistic on how many answers named your page — nobody can honestly promise that, and anyone selling a number for it has estimated it. What you can do: prove the access, ask ChatGPT about your own company regularly, and compare the answers over months. That is less convenient than a metric, but it is true.

How KI-Console proves visits

KI-Console recognises GPTBot, ChatGPT-User, and OAI-SearchBot in uploaded logs and through the tracker where installed. The Radar separates the groups so live access does not get hidden inside generic bot traffic.

What a visit does not prove

A crawler visit does not guarantee ranking, inclusion, citation or a positive answer. It proves technical access at a point in time. An llms.txt does not change that: OpenAI has not committed to using it.