What happens when someone asks ChatGPT to read my page?

ChatGPT-User is the agent OpenAI uses when a person in ChatGPT triggers an action that needs your page. Pasting a URL and asking for a summary is the common case. The fetch happens at answer time, for that one conversation.

OpenAI's crawler documentation says two things about it, as of October 2026:

• It is "not used to determine whether content may appear in Search".
• "Because these actions are initiated by a user, robots.txt rules may not apply."

So a ChatGPT-User request is a read, not an indexing decision. It doesn't add you to ChatGPT search, and it doesn't remove you. For a fuller definition, see what the ChatGPT-User agent is.

What it means for you: treat ChatGPT-User like a visitor's browser acting on someone's behalf. Your job is to make sure that visitor gets the real page.

OpenAI runs three agents with three separate jobs. OpenAI's bots page says "Each setting is independent of the others."

The practical split is index-time versus answer-time. GPTBot and OAI-SearchBot decide what OpenAI stores and may surface later. ChatGPT-User reads on demand. The comparison of index-time crawlers and answer-time fetchers goes deeper.

The common mistake is a blanket AI block. Blocking GPTBot does not hide you from ChatGPT search. Blocking OAI-SearchBot does. Neither setting is the reliable switch for user-triggered reads.

robots.txt may not stop the fetch, but two mechanical things still shape the result.

JavaScript-only content

Vercel's crawler study (December 2024, updated February 2025) found that ChatGPT and Claude crawlers fetch JavaScript files but "don't execute them. They can't read client-side rendered content." Independent re-testing in June 2026 found ChatGPT-User also does not render JavaScript. If your pricing, product names or key claims load only after hydration, the fetch can return an empty shell. ChatGPT then answers the user from whatever text was in the raw HTML, or from what the model already knew. You can test this with the free JavaScript rendering check.

Firewalls and CDNs

A bot-management rule can turn the request away before your server answers. OpenAI's own guidance notes that these systems "can mistakenly block legitimate crawlers, often returning 403 Forbidden errors," and that challenge pages may block automated access. A JS challenge or CAPTCHA is a failed read, even if a human would pass it. OpenAI's published IP ranges are IPv4 only, so an origin reachable only over IPv6 is also out of reach. How often these blocks happen across the web has not been measured.

• Other engines behave similarly. Perplexity documents that Perplexity-User "generally ignores robots.txt rules." Anthropic runs Claude-User for user-triggered fetches, separate from ClaudeBot and Claude-SearchBot. See how Perplexity-User treats robots.txt.
• A successful fetch is not a citation. ChatGPT reading your page for one user says nothing about whether it will name you to the next user. Search eligibility runs through OAI-SearchBot and the index, and OpenAI discloses only that it ranks results using "multiple factors."
• IP blocks can backfire. Blocking an agent's IP ranges can also stop crawlers reading your robots.txt at all. The page on whether a blocked IP can still read robots.txt covers this.
• Hidden text aimed at AI fetchers is a risk, not a tactic. Research on LLM search agents shows manipulated page content can sway answers. Cloaking content for AI agents is illegitimate and can backfire on your reputation.

Say you run a hypothetical project-management SaaS and a prospect pastes your pricing page into ChatGPT. Here's how to make sure that read works.

1. Request the page as ChatGPT-User. Check the status code. A 200 with your real content is a pass. A 403, 429 or challenge page is a fail. The free ChatGPT page fetch test shows what comes back.
2. Compare raw HTML with the rendered page. Anything that appears only after JavaScript runs is invisible to the fetch. Server-render the facts that matter: prices, plan names, the one-line description of what you do.
3. Check your bot-management rules. Make sure your WAF or CDN isn't challenging OpenAI's agents. Validate on user agent plus verified-bot programmes rather than IP lists alone.
4. Set robots.txt per bot for the decisions it does control. For example, to stay in ChatGPT search while opting out of training:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

To see every AI crawler's result side by side, the AI Crawler Checker requests a page as each one and shows which your firewall turns away. For the broader picture, read the guide to user-triggered AI fetchers.

Can I block ChatGPT-User with robots.txt? You can add a rule, but OpenAI says robots.txt rules may not apply to ChatGPT-User because a person initiated the action. Don't rely on it. If you truly need to stop these reads, that's a server or firewall decision, and it also stops prospects who paste your URL into ChatGPT from getting an accurate summary.

Does a ChatGPT-User visit in my logs mean I'm in ChatGPT search? No. OpenAI states ChatGPT-User is not used to determine whether content may appear in search. A visit means someone asked ChatGPT to read that page. Search eligibility depends on OAI-SearchBot being allowed in robots.txt and on your host or CDN letting OpenAI's searchbot traffic through.

Will ChatGPT see content inside tabs or accordions? Only if that content is in the HTML the server sends. Testing reported as of June 2026 found ChatGPT-User does not execute JavaScript. Content injected by a script after a click or on load is missing from the fetch. Content present in the initial HTML but visually collapsed is readable.

Should I block ChatGPT-User to protect my content? It depends on what you're protecting. Blocking at the firewall stops answer-time reads for users, but it doesn't affect training, which GPTBot governs, or search, which OAI-SearchBot governs. For most commercial sites, the bigger risk is ChatGPT summarising your pricing from an empty page, so test before you block.