Autonomous agent
ChatGPT Agent is operated by OpenAI. Browses, compares and buys on a person’s behalf. Blocking it removes you from the shortlist before anyone sees it.
OpenAI runs it as an autonomous agent, which means it works through a task on somebody’s behalf — compare these suppliers, book the table, find the part that fits. It reads pages to decide, and what it cannot read it cannot put on the shortlist it hands back to the person.
What blocking it costs you: The agent that browses, compares and fills in forms on a customer’s behalf cannot get past your front door, so you are never on the shortlist it comes back with. You are removed from the comparison before the customer sees it. The agent shortlists what it can read, and a refused fetch is not a lower score, it is an empty row.
These are two different strings and confusing them is why a rule that looks right does nothing.
A robots.txt rule matches on the product token, case-insensitively, and
ignores the rest of the user-agent header entirely.
| robots.txt token | ChatGPT Agent |
|---|---|
| Operator | OpenAI |
| Category | Autonomous agent |
| Full user agent | Not published by OpenAI. Match on the token above. |
Where an operator has not published the exact string, this page says so rather than inventing a plausible one. A user agent you can grep your logs for is worth nothing if it is a guess.
Paste this into robots.txt at the root of your domain.
User-agent: ChatGPT Agent
Allow: /
Sitemap: https://axray.online/sitemap.xml
An empty Disallow: and Allow: / mean the same thing to a crawler. What does
not mean the same thing is having no rule at all: absence is permission by default, but it is
permission nobody wrote down, and it survives only until somebody adds a blanket
User-agent: * block without thinking about this crawler.
User-agent: ChatGPT Agent
Disallow: /Before you paste that: the agent that browses, compares and fills in forms on a customer’s behalf cannot get past your front door, so you are never on the shortlist it comes back with. That is the cost, and it is paid silently — no error appears in your logs, no report tells you, and the traffic you lose was traffic you never saw arrive.
Measured on 2026-09-06, by reading the robots.txt of
292 public websites and resolving each one against this crawler specifically.
| Verdict | Sites | Share |
|---|---|---|
| Blocked | 6 | 2% |
| Explicitly allowed | 232 | 79% |
| No rule either way | 54 | 18% |
No site is named, here or anywhere else on this domain. The count is the useful part; a league table of businesses that never asked to be measured is not.
The row worth reading twice is the last one. 54 of 292 sites have written no rule about
ChatGPT Agent at all, which means their position on it is an accident rather than a decision — whatever
their User-agent: * block happens to say.
OpenAI runs 4 other crawlers with different jobs, and this is where the expensive mistake happens. Somebody means “do not train on my content”, writes one rule against the operator’s name as they remember it, and blocks the crawler that would have sent them a customer instead.
| Crawler | What it does instead |
|---|---|
OAI-SearchBot |
Indexes you so an assistant can cite and link you. Blocking it removes you from AI answers. |
ChatGPT-User |
Fetches your page because a human asked an assistant to open it right now. Blocking it breaks a live request. |
Operator |
Browses, compares and buys on a person’s behalf. Blocking it removes you from the shortlist before anyone sees it. |
GPTBot |
Collects content for model training. Blocking it is a legitimate business choice with no traffic cost. |
Rules are matched per token. Blocking one of these says nothing about the others, which is the point: you can refuse training and keep every crawler that puts your link in front of a person.
Reading your own robots.txt is not the same as knowing what a crawler concludes from it.
Precedence between a User-agent: * group and a named one, the longest-match rule between
overlapping paths, and a token you spelled slightly wrong all produce a file that looks correct and
behaves otherwise.
npx axray-cli your-site.comThe scan resolves your file against all 54 crawlers on this list individually and reports which ones are allowed, blocked, or covered by nothing. It is free, needs no account, and installs nothing.
ChatGPT Agent is an autonomous agent operated by OpenAI: it works through a task on somebody’s behalf — compare these suppliers, book the table, find the part that fits. It reads pages to decide, and what it cannot read it cannot put on the shortlist it hands back to the person.
Only if you are willing to pay what it costs. The agent that browses, compares and fills in forms on a customer’s behalf cannot get past your front door, so you are never on the shortlist it comes back with. You are removed from the comparison before the customer sees it. The agent shortlists what it can read, and a refused fetch is not a lower score, it is an empty row.
OpenAI does not publish a full user-agent string for this crawler. The robots.txt product token is ChatGPT Agent, and that is what a rule matches on; matching is case-insensitive.
Scan your site with AXRAY. It resolves your robots.txt against 54 named AI crawlers individually rather than reporting one verdict for all of them, so it will tell you whether this specific crawler is allowed, blocked, or covered by no rule at all. It is free and needs no account.
Operator — OpenAIGoogleAgent-Mariner — GoogleGemini-Deep-Research — GoogleAmazonBuyForMe — AmazonNovaAct — AmazonManus-User — Butterfly EffectThe full reference: all 54 AI crawlers, and who blocks each one.
See which of these 54 crawlers your own site is blocking, and what each one costs you. One scan, no account.