Claude Citations: How to Get Your Site Cited by Claude

Claude citations are the source links Claude attaches to an answer when it pulls information from the web. They let a reader check where a fact came from, and they put your page and your brand name inside the answer itself. Earning them depends on Claude being able to reach your page and finding something on it worth quoting.

Quick answer: When Claude uses web search, it cites the pages it drew from so people can check the source. To be one of those pages, your site first has to be reachable by Anthropic's search and user-triggered fetchers, Claude-SearchBot and Claude-User, which follow robots.txt. After that, the page needs clear, specific, well-sourced answers that are easy to quote accurately. Then you confirm with server data that Anthropic's agents are actually fetching the pages you care about, and fix the ones returning errors.

What a Claude citation actually is

Anthropic launched web search in Claude on March 20, 2025, and expanded it to all Claude plans globally on May 27, 2025. In the announcement, Anthropic says that when Claude incorporates information from the web into its responses, it provides direct citations so people can fact check sources. That is the citation: a visible link from the answer back to your page.

The same idea runs through Anthropic's developer platform. The web search tool documentation states that citations are always enabled for web search, and each one carries the source URL, the page title, and a short excerpt of the cited text. Developers who display those outputs directly to end users are required to include the citations. So a page cited through the API can show up with its link inside third-party apps built on Claude, not only on claude.ai.

For a site owner, a citation does three jobs at once. It can send a referral click, it signals that your page was judged relevant to the question, and it puts your name in front of someone who may never click at all.

Citations are not rankings

There is no leaderboard to check. Claude assembles an answer from sources that fit the question in front of it, so the same page can be cited in one conversation and skipped in the next. That volatility is why measuring what reaches your site matters as much as polishing the copy. If you want the broader framework first, our guide to answer engine optimization covers how assistants choose and credit sources in general.

Which Anthropic agents fetch your pages

Anthropic's crawler documentation lists three agents, and they matter to citations in different ways.

The takeaway is simple. Citations come from the search and user-triggered side, not from the training crawler. A site that blocks everything with "Claude" in the name to keep content out of training may also be removing itself from cited answers. If you want the details on the training crawler specifically, read our breakdown of ClaudeBot. Anthropic also states that its bots honor robots.txt directives, so your file is the first place to check. Our guide to robots.txt for AI crawlers walks through agent-by-agent rules.

Structure pages so they are easy to quote

Claude is trying to produce an answer it can back up. A page that already reads like a defensible answer is easier to lift than one that buries the point under a sales pitch.

In practice that means a few habits:

• Put the direct answer near the top of the page, then the reasoning, then the caveats.
• Use headings that match the questions people actually ask.
• Define your terms the first time you use them.
• Keep each paragraph about one idea, so a short excerpt still makes sense out of context.
• Show who wrote the page and why they are qualified to write it.

That last point is cheap to fix and often skipped. Author names, credentials, and a clear organization behind the content make a page attributable rather than anonymous. For more on page layout, see our guide to content structure for AI citations.

Sourcing rules that keep you citable

Fabricated references are a known failure mode for language models, and Anthropic has been on the receiving end of it. TechCrunch reported in May 2025 that a lawyer representing Anthropic apologized in court after Claude produced an incorrect citation in a legal filing. The lesson for publishers is not that Claude is unreliable. It is that assistants prefer sources that make verification easy.

Apply the same discipline to your own pages. If you name a study, link the study. If you quote a figure, say where it came from. If a claim cannot be sourced, say so plainly instead of presenting it as settled fact. A model that quotes you inherits your sourcing, and a page with clean references is a safer page to cite.

Keep your markup and your prose telling the same story

Structured data does not earn citations on its own, but inconsistency costs you. If your article text says one thing and your JSON-LD says another, you have created ambiguity where a clean signal should exist. The fix is boring: treat the visible content as the single source of truth, then build metadata and schema for AI citations from it rather than writing them separately.

• Update the visible page first, then the schema that describes it.
• Keep author, organization, and topic details identical wherever they appear.
• Check every external reference before it goes live.
• Remove or correct stale claims instead of layering new ones on top.

Most sites never bother with this, which is exactly why it helps. A page that agrees with itself is easier to trust and easier to quote accurately.

Check whether Anthropic's agents are actually visiting

You cannot earn a citation from a page Claude never fetches. Most teams have no idea whether Claude-SearchBot or Claude-User has ever requested a specific URL, so they end up guessing from prompt-based tools that report inferred visibility rather than observed traffic.

Your server logs hold the real answer. Look for three things:

1. Frequency. If Anthropic's agents visit regularly, your content is in play and the work is about quality and structure. If they never appear, the problem is access and discovery, and no rewrite will fix it.
2. Which pages. If the agents only fetch old blog posts and ignore your key pages, that tells you where your authority currently sits.
3. Status codes. A fetch that ends in a redirect chain, a 403, or a 404 is a lost chance at a citation. Our guide to HTTP status codes for AI crawlers explains what each response means.

Remember that a user agent string is only a claim. Anthropic publishes its crawler IP addresses at claude.com/crawling/bots.json, and matching requests against that list is the standard way to tell a real crawler from a spoofed one.

A practical checklist for earning Claude citations

1. Confirm robots.txt allows Claude-SearchBot and Claude-User on the pages you want cited.
2. Decide on ClaudeBot separately, as a training question rather than a citation question.
3. Rewrite key pages answer-first, with one idea per paragraph.
4. Add author and organization details to the pages you most want cited.
5. Link every study and figure you mention to its original source.
6. Sync your visible content, metadata, and JSON-LD so they describe the same thing.
7. Track Anthropic's fetches over time, per page and per status code.

None of this depends on a trick that stops working next quarter. It depends on being a source an assistant can quote without embarrassing itself, and then confirming with server data that the assistant is actually showing up to read.

If you want to see which Anthropic agents reach each page and what response they get, citAEOtion is a $99 Lifetime Deal, or you can book a 20-minute walkthrough to see the per-page crawler view first.

Frequently Asked Questions

Does Claude always show citations?

When Claude uses web search, Anthropic says it provides direct citations so people can check sources. Through the developer API, citations are always enabled for the web search tool, and apps that show outputs directly to end users must include them. Answers drawn only from the model's training, with no search, do not come with web citations.

Which Anthropic crawler matters most for Claude citations?

Claude-SearchBot and Claude-User. Anthropic says blocking Claude-SearchBot prevents indexing for search, and blocking Claude-User prevents retrieval in response to user queries, both of which may reduce visibility. ClaudeBot is the training crawler, so blocking it is a separate decision about model training.

Can I block Claude from training on my content but still get cited?

Anthropic's documentation treats the three agents separately, and robots.txt rules are written per user agent. You can disallow ClaudeBot while allowing Claude-SearchBot and Claude-User. After you change the file, confirm in your server logs which agents keep requesting which pages.

Do I need special schema markup to get Claude citations?

There is no required tag. What helps is consistency between your visible text and your structured data, so the page never contradicts itself. Build JSON-LD from the content on the page rather than writing it separately.

How long does it take before Claude cites my page?

There is no set timeline. It depends on whether Anthropic's agents can reach the page, whether the page answers questions people actually ask, and how clearly it does so. Verify access first, publish sourced answer-first content, then watch crawler activity per page and adjust based on what the data shows.