Skip to content
AEO Canon · the reference for answer-engine optimization

AI Crawler User Agents — The 2026 Cheat Sheet

One scannable reference for the major AI bots — GPTBot and OAI-SearchBot (OpenAI), ClaudeBot and Claude-SearchBot (Anthropic), Google-Extended, Applebot-Extended, PerplexityBot, CCBot, and Bytespider — with each bot's owner, purpose, and whether it actually obeys robots.txt.

BBurke Atkerson2 min read

The major AI bots split into three jobs — training, search-and-cite, and on-demand fetch — and each needs its own robots.txt rule. OpenAI alone runs three separate crawlers; Anthropic runs three more. Below is one scannable reference so you can decide who to allow.

Quick answer

There is no single AI bot. GPTBot and Google-Extended train models; OAI-SearchBot and Claude-SearchBot retrieve and cite; ChatGPT-User and Claude-User fetch on demand. Each is a distinct user-agent — address them individually, and know that a few (Bytespider, some stealth crawlers) ignore robots.txt.

Which bots power training vs citation?

They are different bots with different goals. Training crawlers like GPTBot, Google-Extended, and Applebot-Extended collect content to teach future models — blocking them keeps you out of the training set but not out of live answers. Search-and-cite bots like OAI-SearchBot and Claude-SearchBot fetch pages to build the answers users see right now, and they are the ones that send citations. On-demand bots (ChatGPT-User, Claude-User) fetch a single page the moment a user references it.

Why does each bot need its own rule?

Because a rule for one says nothing about the others. Allowing GPTBot does not allow OAI-SearchBot, and blocking ClaudeBot does not block Claude-User. They share a company but not a user-agent string, so you must name each one explicitly in robots.txt.

Do these crawlers actually respect robots.txt?

Mostly, but not all. The major OpenAI, Anthropic, and Google bots honor disallow rules. Bytespider (ByteDance/TikTok) and certain stealth Perplexity crawlers have been documented fetching content despite robots.txt, so treat robots.txt as a signal, not a wall.

Major AI crawlers at a glance (2026). Source: Anagram, AI Crawlers Explained.
BotCompanyPurposeObeys robots.txt?
GPTBotOpenAIModel trainingYes
OAI-SearchBotOpenAISearch retrieval + citationYes
ChatGPT-UserOpenAIOn-demand fetchYes
ClaudeBotAnthropicModel trainingYes
Claude-SearchBotAnthropicSearch retrieval + citationYes
Claude-UserAnthropicOn-demand fetchYes
Google-ExtendedGoogleGemini training only (no Search impact)Yes
Applebot-ExtendedAppleApple Intelligence trainingYes
PerplexityBotPerplexitySearch retrieval + citationMostly — stealth crawlers documented ignoring it
CCBotCommon CrawlOpen dataset (feeds many models)Yes
BytespiderByteDance / TikTokTraining / scrapingDocumented ignoring it

Use this to decide who reads you. Next, see which AI crawlers to allow and the full GPTBot glossary entry.

Which AI crawlers should I allow?

Allow the search-and-cite bots at minimum — they send citations. Training bots are a separate choice.

Read the full answer →
What is GPTBot?

OpenAI's training crawler — separate from OAI-SearchBot, which handles ChatGPT Search retrieval.

Read the full answer →
How do meta robots tags affect AI crawlers?

Some AI bots honor noindex and specific meta directives — but not all, and not identically.

Read the full answer →

Frequently asked questions

How many AI crawlers does OpenAI run?
Three. GPTBot collects training data, OAI-SearchBot powers retrieval and citation for ChatGPT Search, and ChatGPT-User fetches a page on demand when a user asks about it. Each is a separate user-agent, so a robots.txt rule for one does not cover the others.
Does Google-Extended affect my Google Search ranking?
No. Google-Extended controls only whether your content trains Gemini and related AI products. Blocking it has no effect on Googlebot, indexing, or your position in classic Google Search results.
Do all AI crawlers obey robots.txt?
No. Most major bots from OpenAI, Anthropic, and Google honor robots.txt, but Bytespider and some stealth crawlers have been documented fetching content despite disallow rules. Robots.txt is a request, not an enforced firewall.

Related reading

HTTP status codes are instructions to AI crawlers — 200 means crawl and use, 301 consolidates signals to a new URL, 404 drops a page slowly while 410 drops it fast, 429 says back off, and repeated 5xx errors can get a page removed from the answer pool entirely.

2 min read

The Core Web Vitals thresholds are fixed — LCP good under 2.5s, INP good under 200ms, CLS good under 0.1 — and you pass only when 75% of visits hit good on all three. Speed does not directly rank you in AI answers, but it keeps crawlers fetching and users landing on the pages that get cited.

2 min read

In an XML sitemap only a few fields carry weight — loc is essential, lastmod is trusted as a freshness signal if you keep it honest, while changefreq and priority are largely ignored. Include only canonical, indexable 200 URLs and split large files with a sitemap index under 50,000 URLs.

2 min read