How to Get Cited by Claude
Claude now searches the web and cites what it finds. Here is what its crawler needs, what kind of page it trusts, and how that differs from ChatGPT.
To get cited by Claude, allow ClaudeBot in robots.txt and at the network edge, publish pages that state facts plainly enough to be lifted as a citation, and treat its web search feature as a tool the model uses selectively rather than on every reply.
This is the Claude-specific piece of generative engine optimization; the mechanics below are what is particular to this one assistant.
On this page
Does Claude actually cite sources when it answers?
Yes, but only once its web search capability is in play. Anthropic describes this directly: when Claude searches the web, it returns citations linking back to the specific pages it used, so a reader can verify the claim rather than take the model's word for it [3]. That framing matters, because it tells you what a citation is for from Anthropic's point of view — verification, not attribution as a courtesy.
Web search is not a feature that runs on every single reply. It is a tool the model can choose to invoke when a question benefits from current information, or one a developer explicitly turns on through the API for their own product [2]. A question with a stable, well-known answer may never trigger a search at all, and no page can be cited for an answer the model never went looking for external confirmation of.
That has a direct implication for what "being cited by Claude" even means as a goal: it only applies to the subset of questions where searching genuinely helps, which in practice tends to be current facts, specific numbers, comparisons and anything time-sensitive — exactly the territory how ChatGPT chooses brands already covers for the sibling engine.
Which crawler do you have to allow, and where does it usually get blocked?
ClaudeBot, and Anthropic names it explicitly in its own support documentation along with instructions for allowing or blocking it [1]. An explicit allow line naming the crawler is worth more than a bare wildcard, because a wildcard is the first thing someone tightens later without checking what else depended on it.
The more common failure is not the robots.txt file at all. It is a CDN, a hosting platform's bot-management layer, or a security plugin that ships with AI crawlers blocked by default, on the theory that most bot traffic identifying itself as an AI agent is a scraper rather than a legitimate retrieval request on a user's behalf. That default catches ClaudeBot along with everything else, silently, with no error message and no visible change in traffic.
| Layer | How it usually happens |
|---|---|
| robots.txt | A tightened wildcard rule written for a different bot |
| CDN or WAF | A default "block AI crawlers" rule nobody reviewed |
| Hosting platform | A security plugin shipped with AI bots blocked |
| Application layer | Content gated behind a login or a form |
Name ClaudeBot explicitly in robots.txt
An explicit allow survives a later wildcard change in a way a bare permissive default does not.
Check the CDN and hosting layer separately
These sit outside robots.txt entirely and are the more common source of a silent block.
Fetch your own pages as ClaudeBot
Confirm a 200 response and full HTML, not a shell that only renders after client-side JavaScript runs.
Repeat the check after any infrastructure change
A hosting migration or a new firewall rule is the single most common cause of a citation rate quietly going to zero.
Does it matter whether someone reaches Claude through the app or through the API?
Somewhat, and it is worth knowing which one you are actually optimizing for. A person using Claude's own consumer app gets web search as a built-in option the assistant can reach for on its own judgment. A company that has built Claude into its own product through Anthropic's API only gets that behavior if the developer has explicitly turned the web search tool on for that integration, and many have not, because it adds latency and cost to every call.
That means the same question can get a completely different answer depending on which surface someone is using, independent of anything about your own site. A prospect asking a question inside a partner's Claude-powered support widget may never trigger a search at all, while the same person asking the consumer app directly might. There is no way to detect which surface a real user was on, which is another reason a single good result proves very little and a repeated pattern across many runs is what actually counts.
For most companies doing GEO work, the consumer app is the more useful thing to track, simply because it is the one you can test consistently yourself. Treat any citation you see through a third party's Claude-powered product as a bonus signal rather than the baseline you are measuring against.
What makes a page usable as a citation for Claude specifically?
The same properties that make it usable for most other assistants, with one addition worth naming: because Claude positions citations as verification for the reader, a page that states a claim alongside the evidence for it — a number next to its source, a date next to a fact — is easier to cite honestly than a page that asserts something with no way to check it.
Write the direct answer first, in plain language, before any qualification or caveat. A section that opens with three sentences of throat-clearing before it says anything concrete is a section that gets skipped over in favor of one that answers immediately. This is not specific to Claude, but it applies with particular force here because verification is the explicit point of the feature.
Keep claims specific and dated. "Pricing starts at $500 a month, checked September 2026" is citable in a way "affordable pricing" is not, because the second version gives a verification-minded system nothing concrete to attach a citation to.
How does this compare to optimizing for ChatGPT or Google's AI features?
Closely, but not identically. Getting cited by Perplexity already shows how much the target shifts between engines that cite four sources and engines that cite nineteen; Claude sits closer to ChatGPT's end of that range in practice, which means the same discipline of writing a complete, self-contained answer per section transfers with little adjustment.
Where it diverges more is timing and scope. AI Overviews vs AI Mode covers a system that runs its own search layer on effectively every query; Claude's web search is opt-in per conversation or per API call, so there is no guarantee a given question ever triggers a fetch of your page at all, however well-optimized it is. That makes consistency across many questions, not a single perfect answer, the more honest measure of progress.
The mistake is treating every assistant like it searches the same way on every question. Claude searches when it decides to, so the real test is whether it chooses to on the questions that matter to your business, not whether one clever prompt surfaces your page once.
What is the fastest way to test whether any of this is working?
Run a fixed set of twenty to thirty questions your own buyers would plausibly ask, phrased the way a person actually types them rather than as clean keyword phrases, and specifically include questions that require current information so web search is likely to trigger at all. Log whether you were named and whether a citation actually linked to your own domain, on a weekly cadence.
Do this alongside the same exercise for ChatGPT and Perplexity rather than in isolation, because the overlap and the gaps between the three tell you more than either number alone. A brand cited consistently by two of three but never the third usually has a specific, fixable reason — a crawler block, a missing profile, a page that never got indexed — and finding it is faster once you can see the pattern across engines side by side, week over week, rather than guessing from a single afternoon of manual testing. Our Claude optimization service runs exactly this cross-engine tracking as a standing weekly check rather than a one-off audit, and reports the gap in plain language, not just a score.
One file people ask about early in this process is covered in llms.txt, explained — worth reading before you spend time maintaining one on the assumption it changes citation rates.
Questions people also ask
Does Claude cite sources the way ChatGPT and Perplexity do?
Yes, when its web search tool is used it returns inline citations linking back to the pages it drew from, the same general pattern as the other two.
Which crawler blocks Claude specifically?
ClaudeBot, named explicitly in Anthropic's own documentation. A wildcard AI-blocking rule in a firewall or CDN will also catch it even if your robots.txt allows it by name.
Is Claude's web search always on?
No. It is a tool the model can choose to invoke, or that a developer explicitly enables through the API, rather than something every conversation triggers automatically.
Does Claude prefer the same kind of source as ChatGPT?
Broadly yes: primary documentation, specific and verifiable claims, and pages that answer one question cleanly, more than persuasive marketing copy.
How do I test whether Claude can actually reach my site?
Fetch your key pages with ClaudeBot's published user agent and confirm a 200 response with full HTML, the same test used for any other AI crawler.
Sources
- Does Anthropic crawl data from the web, and how can site owners block the crawler? — Anthropic (checked 23 September 2026)
- Web search tool — Anthropic (checked 23 September 2026)
- Claude can now search the web — Anthropic (checked 23 September 2026)
See whether ClaudeBot can actually reach your site
Book a free 30-minute consultation with Ibrahem and Jeremy. We look at your site while we talk, and you leave with a prioritised plan.
Who wrote this
Ibrahem HelalCo-founder
Leads strategy and delivery.
Works on: SEO, GEO
Reviewed before publishing by Jeremy Zava, Co-founder.
- Generative engine optimization: the playbookHow answer engines pick what to cite, what a quotable page looks like, which crawlers to allow, how to measure it, and a 90-day plan built on evidence.
- How ChatGPT Chooses Which Brand to RecommendChatGPT reconstructs a consensus from what other pages already say about a brand, not from the brand's own homepage. Here is what it actually weighs.
- How to Get Cited by PerplexityPerplexity cites nineteen sources per answer on average, far more than ChatGPT. Here is what its crawler needs, what it prefers, and how to check it worked.
- AI Overviews vs AI Mode: What's the DifferenceAI Overviews sits inside normal Google results; AI Mode is a separate conversational search. Here is how they differ, and which to optimize for first.