How to Get Cited by Claude

Claude cites a page when it can retrieve it, understand it, and justify choosing it over the alternatives. Getting cited means clearing all three: your page must be crawlable and not blocked to Anthropic's crawler, structured clearly enough that a model can extract a clean answer from it, and credible enough that the model prefers it to competing sources. Freshness, a named author, tables, and direct answers under the exact question all raise the odds. There is no submission process and no paid placement.

Buyers increasingly ask an assistant before they open a search engine. Someone evaluating software now types a question into Claude and reads the answer with three or four citations under it, and the brands in those citations get the consideration. Nobody is bidding for that slot, which is exactly why it is worth understanding how the slot gets filled.

How Claude picks sources

Claude does not have an index of its own the way Google does. When a question needs current information, it runs a web search, retrieves a set of candidate pages, reads them, and cites the ones it actually used. Third-party analyses of Claude's cited results consistently find its retrieval leans heavily on Brave Search, which means classic search visibility remains upstream of citation: a page nobody's index can find is a page Claude will never see.

From there, the selection breaks into three stages, and a page can fail at any one of them.

Stage	What Claude is doing	What makes a page fail	What makes a page win
1. Retrieval	Finding candidate pages for the question	Blocked crawler, thin internal linking, content rendered only by JavaScript	Indexed, crawlable, linked from pages that already rank
2. Understanding	Reading the page and extracting a usable answer	Buried answers, vague headings, walls of prose, no structure	Clean HTML, the question as a heading, a direct answer under it, tables, schema
3. Selection	Choosing which sources to cite	Unattributed claims, stale dates, generic content-farm feel	Named author, dated content, specific evidence, recognizable publisher

Most pages that miss out are not losing at stage three on authority. They are losing at stage two, because the answer is technically on the page but takes four paragraphs to arrive, and a competing page said it in one sentence.

Do not block the crawler

Start here, because it is the only step that can silently zero out everything else. Anthropic operates named crawlers, and if robots.txt disallows them, or a bot-fight setting at your CDN challenges them, you are not a candidate at any price. Check that ClaudeBot is allowed alongside the other AI crawlers, and check it at the CDN layer too, since plenty of sites allow bots in robots.txt while a firewall rule quietly blocks them.

The same applies to how your content is delivered. If the substance of the page only appears after JavaScript runs, you are relying on a fetcher to behave like a browser, and many do not. This is the same reason pipelines built to turn a web page into clean, LLM-ready text struggle with client-rendered sites: the markup arrives empty. Server-render the content that matters.

The seven changes that make a page citable

1. Put the answer in the first 60 words. Open with a direct, self-contained answer to the page's core question, written so it can be lifted out and still make sense with no surrounding context. Do not open with what this article will cover. That is the single highest-leverage change on most pages.

2. Use the exact question as a heading. Match the phrasing people actually type, verbatim, and answer it immediately underneath in 40 to 60 words. Models match questions to headings; a clever heading loses to a literal one.

3. Put comparisons in tables. Models extract structured data far more reliably than prose. A pricing table, a feature table, or a stage-by-stage table like the one above is much more likely to be quoted than the same information in paragraphs, because there is no ambiguity about what maps to what.

4. Be specific and cite your evidence. Numbers, dates, named sources. A page saying most companies pay $1,500 to $5,000 a month, attributed to a named survey, is citable. A page saying pricing varies is not, because there is nothing in it to quote.

5. Name a real author and date the page. Claude's cited sets skew toward content with visible attribution and a clear publisher. An author byline with genuine credentials, plus an honest last-updated date, both feed the credibility signal and the freshness signal.

6. Keep the facts fresh. Recency-dependent questions are exactly the ones that trigger a web search in the first place, so a page whose numbers are two years old gets passed over for one that says 2026 and means it. Update the figures rather than reprinting them.

7. Ship correct structured data. Article, Organization with sameAs, and BreadcrumbList help a model resolve who you are and what the page is. Schema does not force a citation, but it removes ambiguity, and ambiguity is what loses at the understanding stage.

What you cannot control

Two things matter and are not on-page. The first is that AI assistants over-cite recognizable publishers and primary sources relative to unknown domains, which is the same authority problem as classic SEO wearing a different hat. The second is third-party corroboration: for commercial questions, assistants lean on review platforms and comparison sites, so what G2, Capterra, and independent roundups say about you feeds directly into whether you get recommended. You cannot write your way into those. You earn them.

Which leads to the thing worth being honest about: there is no trick. The pages Claude cites are, overwhelmingly, the pages a careful human editor would also cite, because they answer the question directly, show their evidence, and come from somewhere credible. That is not a coincidence, and it is why the work of getting cited by Claude, ChatGPT, and Gemini turns out to be almost the same work as ranking, done with more discipline about structure.

How to tell if it is working

There is no Search Console for AI citations. The practical approach is to ask the assistants the questions your buyers ask, on a regular schedule, and record who gets cited. Track the handful of queries that would actually send you a customer, not vanity phrases. Over a few months you will see whether new pages start appearing, and which page shapes get pulled most often. In our experience it is the ones with tables and a hard number in the first sentence.

Last updated July 2026. Claude's retrieval and citation behavior is not fully documented by Anthropic; the mechanics above reflect Anthropic's published guidance on its crawlers plus consistent third-party analysis of Claude's cited sources in 2026, and it changes. Re-test rather than assuming.

Related reading: AI search optimization, answer engine optimization, how to get cited by Gemini, and how to rank on ChatGPT.