ChatGPT-User
| Vendor | OpenAI |
| Type | Live retrieval / on-demand fetcher |
| robots.txt token | ChatGPT-User |
| JavaScript rendering | Sometimes — uses a headless browser for "browse" flows |
| Honors robots.txt | Yes |
| Vendor docs | platform.openai.com/docs/bots |
User-Agent strings
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; ChatGPT-User/1.0; +https://openai.com/botPurpose
ChatGPT-User is the bot that fetches a URL right now because a ChatGPT user just asked a question that requires it — clicking a link in a conversation, asking ChatGPT to summarize a specific page, or letting ChatGPT's "browse" feature follow a search result. Every request from this bot corresponds to an active human-driven query.
This is the bot that directly affects revenue: if it cannot reach a retailer's product page, the user gets no answer from that retailer when asking ChatGPT about a purchase.
Network identity
- Hostname pattern:
*.openai.com - IP list: shared with OpenAI's other crawlers; see
https://openai.com/chatgpt-user.json if available, or the combined bot IP ranges at the OpenAI docs.
In our audit
Blocked at 62/100 e-commerce sites — bundled with the training bots in the Cloudflare "Block AI Bots" cluster. This is the most consequential block of the four AI UAs we tested: it removes the retailer from ChatGPT shopping answers in real time.
In our v3 rendered probe ChatGPT-User received some content from walmart.com, samsung.com (post-fingerprint check), and amazon.com (small 202 stub). At amazon.co.uk, where Googlebot received the full 28 KB pre-rendered page, ChatGPT-User received a 200-byte stub — the dynamic-rendering allowlist excludes it.
How to allow / block
To allow ChatGPT live retrieval (recommended for ecommerce):
User-agent: ChatGPT-User
Allow: /To block:
User-agent: ChatGPT-User
Disallow: /The vendor documentation makes this distinct from GPTBot. Site owners who want to opt out of training but participate in live retrieval should block GPTBot and allow ChatGPT-User.
Quirks
- Per-query usage: bursty, low total volume, but each request maps
to a user expressing intent.
- Uses a headless browser tier for some flows — particularly the
"browse" feature where ChatGPT follows search results. In other flows it's a simple HTTP fetch.
- Documentation since 2024 explicitly clarifies that ChatGPT-User
requests "are made for users" and not for training. Sites that block it on AI-training grounds are misreading the bot.