Findings · 2026-11-13 · By VSNARY | Emmanuel Orta
AI crawler IP ranges from 18 operator lists, in one file
Checking whether a crawler is real means comparing its address with the operator's list. Those lists live at 18 different addresses in several formats. We now mirror them in one file.
CrawlCheck now publishes one JSON file that mirrors 18 address lists published by crawler operators: OpenAI, Anthropic, Perplexity, Google, Microsoft, Apple, Amazon, Mistral and Common Crawl. It is refreshed hourly. Each source shows whether the last read worked, its HTTP status and a SHA-256 of what was read. On 13 November 2026 it held 3,963 IPv4 and 408 IPv6 prefixes from 18 of 18 sources.
A user agent is a claim. Anyone can send a request that says GPTBot. The usual way to check the claim is to compare the request's address with the list the operator publishes. The trouble is the lists: each operator keeps its own, at its own address, in its own shape, and two of Amazon's are web pages of single addresses.
The file is at crawlcheck.io/ai-crawlers/ip-ranges.json. It is free and needs no key.
What is in it #
| Operator | Lists mirrored | Crawlers covered |
|---|---|---|
| OpenAI | 4 | GPTBot, OAI-SearchBot, ChatGPT-User, OAI-AdsBot |
| Anthropic | 1 | ClaudeBot, Claude-SearchBot, Claude-User |
| Perplexity | 2 | PerplexityBot, Perplexity-User |
| 3 | Googlebot and its common crawlers, Google-Agent, Google-GeminiNotebook | |
| Amazon | 3 | Amazonbot, Amzn-SearchBot, Amzn-User |
| Mistral | 2 | MistralAI-User, MistralAI-Index |
| Microsoft | 1 | bingbot |
| Apple | 1 | Applebot |
| Common Crawl | 1 | CCBot |
How to read a source #
Every source carries the address we read, the crawlers it covers, whether the read worked, the HTTP status and a SHA-256 of the response. If an operator's server fails, the source is marked as failed and keeps its last good list with the time that list was read, flagged as stale. You can tell a fresh list from an old one without guessing.
What it is not #
It is a mirror, not an authority. The operator's own page is the source, and each entry links to it. The file is cached for an hour, so a change on an operator's side can take up to an hour to appear.
It also holds only what operators publish as crawler lists. Anthropic's on-demand fetcher can arrive from a second published list that is not a crawler list, which is how we got it wrong ourselves.
Using it #
Fetch the file on a schedule, load the prefixes for the crawlers you care about, and compare each request's address before you trust its user agent. If you block at the server, pair the address check with a user agent rule: one stops the named crawler, the other stops things pretending to be it.
CrawlCheck’s dataset when this page loaded: 570,959 sites measured.
Share this finding:LinkedInXEmailFollow Emmanuel on LinkedIn
Questions this post answers
Where can I get AI crawler IP ranges in one place?
At crawlcheck.io/ai-crawlers/ip-ranges.json. It mirrors 18 lists published by OpenAI, Anthropic, Perplexity, Google, Microsoft, Apple, Amazon, Mistral and Common Crawl.
How often is the file updated?
It is re-read from the operators' own pages and cached for one hour.
What happens when an operator's list is unreachable?
That source is marked as failed and keeps its last good list, flagged as stale, with the time it was last read successfully.
Is an address check enough to verify a crawler?
It is the strongest check most sites can run. A request whose address is on the operator's list for that crawler is very likely real; one that is not is either forged or relayed through something else.
Related findings
The main product
Found this on your own site? We fix it for $100.
Scan free to see where you stand. The fix is one site, every finding implemented and re-measured, with a sealed before and after.
Free browser extension
Check this on any site you open, without leaving the page.
CrawlCheck for Chrome grades every site you visit A–F and shows, from your own browser, what search engines and AI assistants are actually served. Free, no account.
Write for the library
Measured something yourself? Submit an article.
Send your finding, the data and the method. A person reads every submission and replies by email; accepted articles are published under your name with a link to your study. Free to submit. Articles are accepted on merit; no one can pay to have one published.
How it works · a reply within about a week
Spotted a mistake or a bug on this page? Report an error. Confirmed errors are fixed and logged on the corrections page.
Comments
Comments are read before they appear. Nothing is published automatically, and no account is needed. An AI agent with a published signing key can post without the queue: it signs the request and the signature is checked against its own domain (how). Those appear at once, marked verified.
Writing about this? Facts, live figures and marks — every number on that page is dated and traceable to a scan.