Technical guide · 30 August 2026

Classify OAI-SearchBot, GPTBot and PerplexityBot correctly

Blanket rules for AI crawlers often mix up two different decisions: should a page be discoverable in current AI search, and may its content be used for future model training? Each user agent needs to be assessed according to its documented purpose.

The short answer

OAI-SearchBot is relevant to ChatGPT search; GPTBot is not. GPTBot concerns OpenAI's potential use of content for model training. PerplexityBot supports search and linking in Perplexity. Google's AI Overviews and AI Mode rely on normal Google Search and therefore Googlebot; Google-Extended is not a separate search crawler and, according to Google, has no effect on Google Search.

Crawler, purpose and consequence

User agentDocumented purposeControlWhat blocking means
GooglebotGoogle Search, including the search foundation for AI Overviews and AI Moderobots.txt and robots meta tagsThe URL can lose crawl and Search eligibility, removing the foundation for Google's generative search features.
BingbotBing's index, which also supports selected Copilot and grounding experiencesrobots.txt and Bing-compatible robots meta tagsContent may not be indexed reliably or selected for supported grounding results.
OAI-SearchBotDiscovery and linking in ChatGPT searchrobots.txtOpenAI cannot fetch page content through this search crawler; under some conditions the bare URL may still be known.
GPTBotPotential use of web content to improve future modelsrobots.txtTraining use is declined; this is a separate decision from ChatGPT search.
PerplexityBotDiscovery and linking in Perplexity searchrobots.txt and, where relevant, network allow-listsThe page is unavailable to this search crawler; that says nothing about other systems.
Google-ExtendedA control token for Gemini Apps and Vertex AI; not a separate HTTP user agentrobots.txt as a user-agent tokenAccording to Google, it has no effect on inclusion, ranking or presentation in Google Search.

A credible policy begins with the business decision

Before changing a file, establish which content should be publicly discoverable and which training uses the organisation accepts. The policy can then remain minimal and auditable.

  1. 1

    Identify public search content

    Only pages intended to be public and citeable for people belong in the search strategy.

  2. 2

    Separate search from training

    OAI-SearchBot and PerplexityBot are search crawlers; GPTBot represents a separate training decision.

  3. 3

    Keep specific groups complete

    Once a user agent has its own group, do not assume generic rules are automatically inherited.

  4. 4

    Validate configuration, not imagined behaviour

    A validator can test rules and reachability, but cannot claim that an external bot actually indexed or cited the page.

Example: keep a public website open to search crawlers

User-agent: *
Allow: /

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

Example: allow ChatGPT search while blocking GPTBot separately

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

User-agent: *
Allow: /

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

The second example is not a blanket recommendation. It only demonstrates that search access and training access can be expressed separately.

What llms.txt does not do

llms.txt is a community proposal for a compact content index. It replaces neither robots.txt nor sitemaps, indexing, internal links or accessible HTML. Google explicitly says that its AI features require no special AI files or markup. llms.txt should therefore be treated, at most, as supplementary documentation — not a ranking lever.

What can be validated technically

  • robots.txt is reachable and can be parsed
  • Googlebot, Bingbot, OAI-SearchBot and PerplexityBot are not accidentally blocked from target paths
  • the sitemap and canonical pages are reachable
  • robots meta tags do not contradict crawl policy
  • important content is present in delivered HTML
  • the server or WAF does not return unintended errors to legitimate crawlers

Primary sources

From policy to a concrete website review

The AI crawlability review examines more than the text file: status codes, redirects, renderability and the actual target pages are part of the assessment.