Skip to content
All AI readability rules
Caps your score at 40CriticalAI Crawler Access8 of 30 ptsCrawlers

How to fix: OAI-SearchBot is blocked from crawling

OAI-SearchBot is OpenAI's retrieval crawler for ChatGPT search, and it is a separate system from GPTBot, the training crawler: if your robots.txt disallows OAI-SearchBot, ChatGPT search cannot fetch the page for an answer, no matter how much of your content GPTBot already ingested for training.

The check

What is this?

This check fails when your site's robots.txt disallows OAI-SearchBot, OpenAI's crawler for ChatGPT search. A blocked retrieval crawler cannot fetch the page, so the page cannot be used as a source for a ChatGPT search answer. AccessKnight scores it at 8 points, the heaviest single check in the AI Crawler Access pillar.

OpenAI runs two independent crawlers with two different jobs. GPTBot collects content for model training. OAI-SearchBot fetches pages so ChatGPT search can quote them in a live answer. They read separate robots.txt groups and neither decision affects the other, which is why blocking GPTBot does not remove you from ChatGPT search, and why allowing GPTBot does not put you back into it while OAI-SearchBot is disallowed.

Most blocks are not deliberate. Three patterns account for nearly all of them: a blanket User-agent: * group with Disallow: / left over from a staging environment; an SEO plugin whose block-AI-crawlers toggle writes the rule for you without naming which crawlers it covers; and a CDN or WAF bot rule that rejects the request before robots.txt is ever read. The third is invisible in robots.txt, so a file that looks correct can still leave the crawler shut out.

The impact

Who does this affect, and why does it matter?

ChatGPT search is affected directly and immediately, because it is the surface OAI-SearchBot feeds. Anything built on OpenAI's search stack inherits the same block. Google AI Overviews, Perplexity, Copilot, and Claude are unaffected by this specific directive, because each uses its own crawler and its own index — this rule closes one door, not the corridor.

The person who absorbs the loss is the site owner, usually without knowing it. A disallowed retrieval crawler produces no error, no email, and no report anywhere in your own tooling. The page keeps ranking in Google, keeps earning traffic, and simply never turns up as a source when someone asks ChatGPT a question the page answers well.

AccessKnight scores this check at 8 of 8 points, the heaviest single check in the AI Crawler Access pillar. It is also one of only two gating checks in the whole scan: when it fails, the headline Machine-Readability Score is capped at 40 regardless of how well the other seventeen checks score, and the report states both the cap and the reason for it.

The cap is deliberate rather than punitive. A perfect score on structure, semantics, and extractability describes a page that is beautifully built for a retrieval crawler that has been told not to fetch it. The asymmetry is worth stating plainly: blocking OAI-SearchBot means the page cannot be retrieved for ChatGPT search, while allowing it only makes the page eligible to be retrieved. Being fetchable is a precondition, not an outcome.

The practical consequence is one of influence rather than traffic. When someone asks an assistant a question your page answers, the assistant assembles its answer from whatever it can reach. A disallowed retrieval crawler removes your page from that pool entirely, so the answer gets built from pages that did let the crawler in — and because there is no click, there is no line in your analytics telling you it happened.

The fix

How do I fix it?

Add an explicit OAI-SearchBot group to robots.txt with Allow: /, because a group naming the crawler overrides the wildcard rules that cause most of these blocks. Then rule out the two causes robots.txt cannot show you: a plugin that wrote the rule for you, and a CDN rejecting the crawler before the file is ever read.

  1. Open your robots.txt at https://yourdomain.com/robots.txt and look for any group that disallows OAI-SearchBot, including a blanket User-agent: * group with Disallow: / that would catch it by default.
  2. Add an explicit User-agent: OAI-SearchBot group containing Allow: /. A group that names the crawler wins over the wildcard * group, so the specific rule is the one that decides access.
  3. Write the token exactly as OAI-SearchBot, with no version suffix. The parser matches user-agent tokens, so a group written as User-agent: OAI-SearchBot/1.0 is not recognised as that crawler's group and the wildcard rules apply instead.
  4. Check your CMS or SEO plugin for an AI-crawler blocking toggle. Several write the disallow into robots.txt for you, and most do not distinguish retrieval crawlers from training crawlers.
  5. Test the file as the crawler itself, which catches an edge block that robots.txt cannot describe: curl -sA "OAI-SearchBot" https://example.com/robots.txt. Getting the file back means the path is open; a 403, a challenge page, or a timeout means your CDN or WAF is rejecting the request before robots.txt is ever consulted, and the fix belongs in the bot rules there rather than in the file.
  6. Re-scan with AccessKnight and confirm the check reports 8 of 8 and the headline score is no longer capped at 40.
robots.txt — Before
User-agent: *
Disallow: /

User-agent: OAI-SearchBot
Disallow: /
robots.txt — After
User-agent: *
Disallow: /admin/

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /
Detection

How does AccessKnight detect this?

AccessKnight fetches /robots.txt from the origin of the final URL after redirects, parses it into per-user-agent groups, and resolves the rules for OAI-SearchBot against the path / — the site root, not the scanned page's path. A group naming the crawler wins over the wildcard * group, which is consulted only when no specific group matches; inside the winning group the longest matching rule decides and Allow wins a tie. An empty Disallow: line counts as allow-all at the lowest specificity, so any more specific rule overrides it. If robots.txt is missing, unreachable, or served as an HTML soft-404, everything is treated as allowed and the check passes. It is worth 8 points and is one of two checks that cap the headline score at 40. One limitation worth knowing, and how to cover it yourself: the fetch is made with AccessKnight's own user agent, so the check reports the policy your robots.txt declares rather than whether OpenAI's crawler is actually served when it arrives. To test that second question, request the file as the crawler — curl -sA "OAI-SearchBot" https://example.com/robots.txt — and treat a 403, a challenge page, or a timeout as an edge block rather than a robots.txt problem.

On screen right now

What strings mean this?

If one of these strings is in front of you right now — in a config file, an HTTP header, an AccessKnight report, or another checker’s output — it is describing the failure explained on this page.

robots.txt
User-agent: OAI-SearchBot
robots.txt
Disallow: /
User-agent
OAI-SearchBot
AccessKnight report
OAI-SearchBot is blocked — you will not appear in ChatGPT search, even if GPTBot already crawled you for training. The two systems are independent.
AccessKnight report
Capped at 40: OAI-SearchBot (ChatGPT search) is blocked in robots.txt. The site is effectively invisible to major AI surfaces until this is fixed.
FAQ

Frequently asked questions

What is the difference between OAI-SearchBot and GPTBot?

They are independent systems with different jobs. GPTBot crawls content for model training. OAI-SearchBot fetches pages so ChatGPT search can quote them in a live answer. They read separate robots.txt groups, so a decision about one has no effect on the other. AccessKnight scores only OAI-SearchBot, because retrieval is what determines whether a page can be used as a source today.

Does blocking GPTBot remove me from ChatGPT search?

No. Blocking GPTBot opts your content out of model training and leaves ChatGPT search retrieval untouched, provided OAI-SearchBot is still allowed. AccessKnight does not score training-crawler access at all — no check in the AI Crawler Access pillar looks at GPTBot, ClaudeBot, Google-Extended, or CCBot, because whether to allow training is a business judgment rather than a technical defect.

I allowed OAI-SearchBot but the check still says it is blocked. Why?

Three causes explain almost all of them. The token may be misspelled or carry a version suffix, in which case the group is never matched. The scan may be reading a different host: robots.txt is fetched from the origin of the final URL after redirects, so editing the apex file will not help a site that redirects to www. Or a CDN may still be serving a cached copy of the old file.

Why is my AccessKnight score capped at 40?

A blocked OAI-SearchBot is one of two gating failures, and the report says so directly: "Capped at 40: OAI-SearchBot (ChatGPT search) is blocked in robots.txt." The cap exists because the remaining ninety-two points describe how well the page would be understood by a crawler that has been told not to fetch it. Allow the crawler and the cap lifts on the next scan.

Do I still need Bingbot allowed if OAI-SearchBot is allowed?

Yes. ChatGPT search selects candidate pages from a search index rather than crawling the open web itself, then fetches the ones it wants to quote, so the two crawlers guard different stages of the same path. Allowing OAI-SearchBot without allowing bingbot leaves the page fetchable but absent from the index it would be chosen out of. The bingbot rule in this guide covers that half, including what is and is not documented about the index behind it.
Sources

Where does this come from?

Every external claim on this page traces to one of the documents below — the vendors’ and standards bodies’ own documentation. What AccessKnight adds is the scan, not the standard.

Is this check
failing on your site?

AccessKnight scans any page against all 18 AI-readability checks — including this one — and shows every issue with a fix. Free, no credit card.

✓ Free — no credit card✓ Nothing installed on your site✓ WCAG + AI readability