AI agents · August 2026
Which AI agents the web blocks
Naming an AI crawler in robots.txt almost always means blocking it: between 58.1% and 97.3% of the sites that name one are turning it away. But these agents do different jobs — some train models, some index you so they can cite you, some fetch a page because a person just asked — so blocking them costs very different things. Measured on the same 0.1% sample as the rest of the census, 15,009 sites.
| Agent | What it does | Sites naming it | Sites blocking it | Blocking, month by month |
|---|---|---|---|---|
| GPTBot GPTBot
Crawls to train OpenAI models. Blocking it does not remove you from ChatGPT answers — that is OAI-SearchBot and ChatGPT-User, which are separate tokens. | Trains models | 8.6% | 7.6% | = +0.1 pp |
| ClaudeBot ClaudeBot
Crawls to train Anthropic models. | Trains models | 7.6% | 6.8% | = +0.1 pp |
| CCBot CCBot
Builds the public Common Crawl corpus, which many model trainers use. Blocking it is an indirect way of opting out of several models at once. | Trains models | 7.4% | 7.1% | = +0.2 pp |
| Bytespider Bytespider
Crawls to train ByteDance models. | Trains models | 7.0% | 6.8% | = +0.1 pp |
| Google-Extended Google-Extended
Controls whether your content is used to train Google’s generative models. It does NOT affect Google Search: blocking it leaves your search ranking alone. | Trains models | 7.0% | 6.5% | = +0.1 pp |
| Amazonbot Amazonbot
Crawls for Amazon’s assistant answers. | Indexes for answers | 6.9% | 6.7% | = −0.2 pp |
| meta-externalagent meta-externalagent
Crawls to train Meta models. | Trains models | 6.7% | 6.4% | = +0.4 pp |
| Applebot-Extended Applebot-Extended
Controls whether your content trains Apple’s generative models. Separate from Applebot, which indexes for Siri and Spotlight and is not affected. | Trains models | 6.4% | 6.1% | = +0.1 pp |
| anthropic-ai anthropic-ai
An older token for the same crawling. Still widely written in robots.txt files, so it is still worth counting. | Trains models | 3.2% | 2.9% | ▲ +0.5 pp |
| cohere-ai cohere-ai
Crawls to train Cohere models. | Trains models | 2.7% | 2.5% | ▲ +0.6 pp |
| YouBot YouBot
Indexes pages so You.com can cite them. | Indexes for answers | 2.5% | 2.4% | = +0.4 pp |
| PerplexityBot PerplexityBot
Indexes pages so Perplexity can cite them with a link back. | Indexes for answers | 2.3% | 1.4% | = +0.1 pp |
| ChatGPT-User ChatGPT-User
Fetches a page because a person asked ChatGPT about it. Blocking it breaks that link for a reader who was already interested. | Fetches on request | 2.0% | 1.4% | = +0.2 pp |
| OAI-SearchBot OAI-SearchBot
Indexes pages so ChatGPT can cite them. This is the one to leave open if you want to appear in its answers. | Indexes for answers | 1.8% | 1.0% | = +0.3 pp |
| Claude-Web Claude-Web
An older token for fetching a page on a user’s behalf. | Fetches on request | 1.0% | 0.7% | = −0.3 pp |
| Claude-User Claude-User
Fetches a page because a person asked Claude about it. | Fetches on request | 0.7% | 0.5% | = +0.2 pp |
| Perplexity-User Perplexity-User
Fetches a page because a person asked about it. | Fetches on request | 0.6% | 0.4% | = +0.2 pp |