Verify ClaudeBot IP Addresses:
Real Anthropic Crawler, or Just Claiming To Be?
Paste an address from your access log to verify a ClaudeBot IP against Anthropic's published CIDR list. Anthropic states that a crawler with a source IP on that list is coming from Anthropic, so a match settles it. The tool then adds something no other ClaudeBot IP check does: whose cloud address space that prefix sits in — because eighteen of the twenty-three are Google Cloud, and that trips people up badly.
Quick Answer: How Do You Verify a ClaudeBot IP?
To verify a ClaudeBot IP, match the address against Anthropic's published CIDR list at claude.com/crawling/bots.json. Anthropic states that a crawler with a source IP on that list is coming from Anthropic, so a match is conclusive. Use robots.txt to opt out, not an IP block.
Jessica Wright
Cybersecurity Threat Researcher
Jessica works on bot verification and IP reputation, separating traffic that is what it says from traffic that only claims to be.
Anthropic generates a particular kind of support ticket. Somebody sees ClaudeBot in their logs, runs the address through an ASN lookup, gets back Google Cloud, and concludes the user-agent is forged. Then they block it. It was real: eighteen of the twenty-three prefixes on Anthropic's published list sit in Google Cloud space and three more in Azure. The ASN lookup was not wrong, it just answered a different question than the one being asked.
The distinction worth holding onto is between checking an address and blocking one. Anthropic is clear that a source IP on its list means the crawler is coming from Anthropic, so a ClaudeBot IP check is a legitimate thing to do. What it warns against is blocking those addresses as a way to opt out, for a reason that is easy to miss: if Anthropic cannot reach you, it cannot read your robots.txt, and the opt-out you actually wanted never registers. Verify with the address. Opt out with the file.
Last reviewed 22 August 2026 · List read from Anthropic's published feed · No lookups logged
View all articles by Jessica WrightWhat Anthropic Actually Says About Verifying ClaudeBot by IP
Most write-ups on this get it backwards, so it is worth quoting the source. From Anthropic's crawler support article, last updated 7 April 2026:
Alternate methods like blocking IP address(es) from which Anthropic Bots operates may not work correctly or persistently guarantee an opt-out, as doing so impedes our ability to read your robots.txt file. If a crawler has a source IP address on this list, it indicates that the crawler is coming from Anthropic.
Two separate statements, and people routinely collapse them into one.
The second sentence is an endorsement. A source IP on the published list indicates the crawler is coming from Anthropic — which is exactly what a ClaudeBot IP check is for, and it is the strongest thing Anthropic says about identification anywhere.
The first sentence is a caution about something else entirely: using an IP block as your opt-out mechanism. The reason given is mechanical rather than a doubt about the list. If you firewall the addresses, Anthropic cannot fetch your robots.txt, so the preference you actually wanted to express never gets read. You end up with a rule that stops today's traffic and no registered opt-out at all.
The split to hold onto: verify with the address, opt out with the file. A ClaudeBot IP match tells you the request is genuine. A robots.txt directive tells Anthropic what you want. Neither does the other's job, and swapping them is the mistake this page exists to prevent.
What else the article commits to
The same page states that Anthropic's bots honour robots.txt, that they support the non-standard Crawl-delay extension — which Google ignores — and that they will not attempt to bypass CAPTCHAs. It also lists the three crawlers separately, a clarification that landed around 20 February 2026, before the IP list itself was published.
For the general verification method across nineteen operators, including the forward-confirmed reverse DNS that Anthropic does not offer, the universal bot verifier is the page that covers it properly. This one stays on Anthropic.
Check what you are testing first. Behind a CDN or load balancer, the connecting address in your logs is your own edge and the real client sits in a forwarded header. Running your load balancer through a ClaudeBot IP check will fail every time and tell you nothing.
Which brings us to the part of the list that surprises almost everyone who reads it.
Why Does a Real ClaudeBot IP Show Up as Google Cloud?
Because most ClaudeBot addresses are cloud space rather than Anthropic's own. Read the published list and the pattern is immediate.
| Address space | Prefixes | Addresses | What an ASN lookup returns |
|---|---|---|---|
| Google Cloud34.x and 35.x | 18 | 18 | Google — not Anthropic |
| Microsoft Azure20.x and 40.x | 3 | 48 | Microsoft — not Anthropic |
| 216.73.216.0/22the only block bigger than a /28 | 1 | 1,024 | Not obviously cloud space |
| Unclassifiedone /32 we do not guess at | 1 | 1 | Run the lookup and see |
| TOTAL | 23 | 1,091 | Zero IPv6 prefixes |
Eighteen of twenty-three prefixes — 78% of the list — sit in Google Cloud ranges. So the sequence that trips people up runs like this: ClaudeBot appears in the log, the owner checks the address, every lookup says Google Cloud, and the obvious conclusion is a forged user-agent. The address is genuine. The lookup is also correct. They are answering different questions.
One honest caveat on the classification above, and on the badge this tool shows next to a match: it is worked out from the address block, not from a live registry query. It is a strong hint rather than a lookup result. An ASN lookup names the network that actually announces an address, and a datacenter check confirms whether it is rented space.
The practical consequence: never verify a ClaudeBot IP by ASN, and never block one that way. Blocking AS15169 to stop ClaudeBot would take Googlebot and a large part of the internet with it — and, per Anthropic's own note, would stop your robots.txt being read into the bargain.
The total is worth noting too. About 1,091 addresses cover the entire published fleet, most as single /32 entries. That is small: Google publishes over a hundred IPv6 prefixes for its common crawlers alone. To see how many addresses a matched prefix covers, the CIDR calculator expands any prefix into its full range.
No IPv6, at all
The list holds twenty-three IPv4 prefixes and not one IPv6 prefix. Google, OpenAI and Common Crawl all publish IPv6. Anthropic does not, which means an IPv6 address cannot match whether it belongs to Anthropic or not. This tool reports an IPv6 ClaudeBot IP check as inconclusive rather than as a non-match, because a list that cannot contain your address is not evidence that your address is fake.
Knowing the request is Anthropic's is where the list stops being useful, though, because it will not tell you which Claude asked.
One List, Three Bots: What a ClaudeBot IP Match Cannot Tell You
Anthropic runs three documented crawlers and publishes one combined list for all of them. Google publishes five files for five categories. OpenAI publishes four. Anthropic publishes one, so a match confirms Anthropic and stops there — even though the three want completely different decisions from you.
ClaudeBot Model training
Collects web content that could contribute to training Anthropic's models. Restricting it signals that your future material should be excluded from training datasets.
If you block it: A content-rights decision with no visibility cost. It does not remove you from Claude search answers or from live user fetches - those are separate agents.
Claude-User Live user-triggered fetch
Fires when a person asks Claude a question and Claude opens a page to answer it.
If you block it: Anthropic states this prevents its system retrieving your content in response to a user query, which may reduce your visibility for user-directed web search.
Claude-SearchBot Search index quality
Analyses content to improve the relevance and accuracy of Claude search responses.
If you block it: Anthropic states this prevents your content being indexed for search optimisation, which may reduce your visibility and accuracy in Claude search results.
anthropic-ai Legacy token
An older user-agent still seen in access logs. It is not a separately addressable crawler today.
If you block it: Harmless to keep in robots.txt for older log entries, but it buys nothing that blocking ClaudeBot does not already cover.
Those tokens are separate in robots.txt, so the split most people want — refuse training, keep search and live fetches — is achievable there. At the firewall it is not, because one address list covers all three. An IP rule aimed at ClaudeBot takes Claude-User and Claude-SearchBot with it every time.
The rule that follows: use robots.txt to separate Anthropic's crawlers, and use IP rules only if you intend to block all of Anthropic — knowing that doing so also stops your robots.txt being read. The crawler blocker writes both layers from this same live list and flags the collision as soon as your selection creates one.
OpenAI has a milder version of the same problem — four files that partially overlap rather than one covering everything — and the GPTBot verifier covers how that plays out. Anthropic is the strictest case: there is no partial separation to find.
Where third-party bot databases get ClaudeBot wrong
Search for ClaudeBot IP ranges and you will find sites publishing an ASN and three /16 blocks for it, alongside a claim that reverse DNS resolves under an Anthropic domain. None of those ranges appear on Anthropic's actual list, and Anthropic documents no reverse-DNS scheme. Some of it looks like stale information from Claude-Web, the earlier crawler; some looks invented.
When a third-party database and the vendor disagree, take the vendor. A rule built from three fabricated /16 blocks would cover 196,608 addresses that have nothing to do with Anthropic, and would still miss every genuine ClaudeBot address on the real list.
What this tool cannot do
Anthropic gives you one signal to check against, so the limits on a ClaudeBot IP check are wider than for most operators. A tool that lists only its strengths is no help when you are about to block real traffic.
There is no second check to run
Anthropic documents no reverse-DNS scheme, so forward confirmation is not available for any Claude agent. The published list is the whole of it, and any tool claiming a stronger verdict is inventing one.
It cannot name which Claude bot
One list covers ClaudeBot, Claude-User and Claude-SearchBot. A match confirms Anthropic and nothing finer. The answer lives in the user-agent on the same log line, once identity is settled.
IPv6 cannot be checked at all
Anthropic publishes zero IPv6 prefixes, so an IPv6 address cannot match whether it is genuine or not. That is reported as inconclusive here, which is the only honest answer available.
A miss is weak evidence
About 1,091 addresses cover the whole published fleet, and Anthropic revises the list as infrastructure moves. Treat a match as conclusive, as Anthropic does, and a miss as a prompt to look at behaviour.
The cloud badge is a hint, not a lookup
Address space is worked out from the block rather than from a registry query. It is right often enough to be useful and it is not authoritative. Confirm it with an ASN lookup before you act on it.
It cannot see your log line
You paste an address. The tool has no view of the user-agent that came with it, the path requested, or the request rate — and all three matter as much as identity when you decide what to allow.
When an address is not on the list, the next useful question is whose network it is — and with Anthropic that question is worth asking anyway, because for a genuine ClaudeBot IP the answer is a cloud provider. A geolocation and network lookup settles what an address is in one step, the range export lists every prefix an operator announces, and the guide to IP spoofing covers why a header is never identity.
Frequently asked questions about ClaudeBot IP verification
How do I verify a ClaudeBot IP address?
Check the address against Anthropic’s published list at claude.com/crawling/bots.json. Anthropic states that a crawler with a source IP on that list is coming from Anthropic, so a match settles it. One combined file covers ClaudeBot, Claude-User and Claude-SearchBot. The user-agent proves nothing on its own, because the client writes it.
Does Anthropic support checking ClaudeBot by IP?
Yes, for identification. Its support article says a crawler with a source IP on the published list “is coming from Anthropic”. What Anthropic cautions against is blocking by IP as a way to opt out, and the reason is mechanical: block the addresses and Anthropic cannot fetch your robots.txt, so your opt-out never registers. Verify with the IP, opt out with robots.txt.
Why does a real ClaudeBot IP show up as Google Cloud?
Because most of those addresses are cloud space. Eighteen of the twenty-three prefixes on Anthropic’s list sit in ranges announced by Google Cloud and three in Microsoft Azure, so every ASN lookup and WHOIS on a genuine ClaudeBot IP names the cloud provider rather than Anthropic. It is expected, not a sign of spoofing, and it catches out a lot of people.
Can I tell which Claude bot sent the request?
Not from the address. Anthropic publishes one combined file for all three agents, so a ClaudeBot IP match confirms Anthropic and nothing finer. Google publishes five files and OpenAI four, which lets those be separated. Read the user-agent on the same log line once the address itself is confirmed.
Does Anthropic publish IPv6 ranges for ClaudeBot?
No. The list holds twenty-three IPv4 prefixes and zero IPv6 prefixes. An IPv6 address cannot match it whether it is genuine or not, so this tool reports IPv6 lookups as inconclusive rather than as a non-match. Every other major crawler operator publishes IPv6, which makes this a real gap.
Can I block ClaudeBot training but allow Claude search?
In robots.txt, yes - the three agents have separate tokens and Anthropic documents the consequence of each. At the firewall, no. One file covers all three, so an IP rule aimed at training takes Claude-User and Claude-SearchBot with it. Anthropic also supports the non-standard Crawl-delay directive, which Google ignores.
My address is not on the list. Is it definitely a fake ClaudeBot?
Not definitely. The list is short - about 1,091 addresses - and Anthropic revises it as infrastructure moves, so a genuine new address can sit outside it for a while. This tool only says “not listed” when the file was actually readable on that request. A match is strong evidence; a miss is weak evidence.
Other sites give different ClaudeBot IP ranges. Which is right?
The file Anthropic publishes. At least one widely-cited bot database gives an ASN and three /16 blocks for ClaudeBot that appear nowhere on Anthropic’s own list, and claims a reverse-DNS scheme Anthropic does not document. When a third-party database and the vendor disagree, take the vendor.
Related IP & crawler tools
This page answers one question about Anthropic. These answer the rest.
Browse the full set on the TrustMyIP tools directory.
Decided how to handle Claude?
The blocker writes robots.txt and firewall rules from this same live list, and tells you when a rule aimed at training would take Claude search down with it.
Last updated 22 August 2026 · List read from Anthropic's published feed · No lookups logged