How to rank in Claude and get cited when it searches the web

What does it mean to rank in Claude?

You rank in Claude when it cites your page or names your brand in an answer it wrote after searching the web. There's no position one. Claude writes a single answer and attaches citations to the passages it used, so the goal is to be among the sources it draws on, and for commercial questions, among the brands it recommends.

Two things make Claude its own project. It searches more often than people assume: web search has been available on every Claude plan worldwide since 27 May 20251, and in the newer Claude apps there's no toggle at all, because Claude "searches the web when it helps"2. And its results come from a different index than ChatGPT's or Google's, so pages that do well elsewhere can be missing here.

If you want the cross-engine picture first (which index each assistant uses and which crawler feeds it), read how AI engines choose their sources. This guide goes deeper on Claude alone.

How does Claude decide which pages to cite?

Claude cites pages its search tool returns and, in some cases, fetches in full. It doesn't browse freely. The sequence is search, select, read, cite, and you can fail at each step.

Step 1: Claude decides whether to search

Anthropic's web search tool documentation says "Claude determines when to search based on the prompt." It searches for recent events, current prices and statistics, and organizations, people or products that might have changed. It skips search for stable facts, creative writing and greetings3. Simple factual queries typically use 1 to 3 searches; comparative or multi-entity research "can use 10 or more"3.

That list is useful. Questions about tools, vendors and prices are exactly the ones Anthropic says trigger search, so commercial prompts in your category will usually be answered from live results. A question like "what is double-entry bookkeeping" may not be, and then the answer comes from training data.

Step 2: the results come from an index Anthropic doesn't name

Anthropic's product and API docs don't name a search provider. Brave Search was added to Anthropic's subprocessor list in March 2025, and Simon Willison found that Claude's citations for a test query matched Brave's top results for the same search6. TechCrunch reported the same finding; neither company commented7. Treat Brave as the strongest available evidence, not a confirmed contract.

Brave runs its own index, which it describes as "40+ billion pages" kept fresh by more than 100 million page updates a day9. Two details on its crawler help page matter for you. The crawler "does not advertise a differentiated user agent," and if a page "is not crawlable by Googlebot, then Brave Search's bot will not crawl it either"8. You can't allow or block Brave by name. You control it through your Googlebot rules.

Anthropic also runs its own crawlers, documented in its crawler help article5:

Step 3: Claude picks results and may read them in full

When both search and fetch are available, Anthropic's web fetch documentation describes Claude searching, selecting "the most promising results," then fetching the full content4. The same page says the fetch tool doesn't support sites "dynamically rendered with JavaScript" and respects robots.txt4. Those docs describe the developer API; Anthropic hasn't published how the consumer app's fetcher differs, so treat a server-rendered page as the safe assumption for both.

Step 4: Claude quotes a short passage

"Citations are always enabled for web search," per the API docs, and each citation carries up to 150 characters of cited text3. Every search result also comes with a page_age field, described as when the site was last updated3. Anthropic doesn't say how much weight that field gets. A visible, accurate date costs nothing, so keep one.

What does the data say Claude cites?

The best public dataset is Slate's study of 1,000 B2B SaaS questions across 40 software categories, run with web search on and captured on 3 August 2026, using Claude Sonnet 4.6 through the API10. Here's what it found for Claude:

• Claude linked a source on 978 of 1,000 answers, citing 8,933 different pages on 3,972 sites.
• It quoted 35.0% of pages ranking first on Google for the question and 17.4% of pages ranking ninth. Google rank helps, but far less than in Perplexity (86.9% at first).
• Page-quality signals mattered more than rank. Word count had the strongest effect size (0.186), and Slate notes Claude is "the only engine where what is on the page changes its odds at all."
• Reddit, Quora and Stack Overflow pages that ranked for a question were cited 0.0% of the time by Claude, against 74.2% by Perplexity.
• It cited 34.5% of ranking pages on research questions and 21.0% on buying questions.

By page format, Claude's citation rates looked like this10:

Two caveats. This is one study, one answer per question, through the API, which may not match what a logged-in user sees in the Claude app. And "position" in the study means Google rank, not Brave rank. Still, the pattern is clear enough to act on: Claude is choosier than Perplexity, it rewards substantive pages, and it doesn't lean on forums.

The 7-step plan to get cited by Claude

Steps 1 to 3 take a week. Steps 4 to 6 are the ongoing work, and step 7 keeps you honest.

Worked example: a restaurant scheduling app

This is a hypothetical company, "Rostra," which sells shift scheduling software to restaurant groups. We haven't run these prompts; the example shows how the checks and decisions fit together.

The access check. Rostra wants to stay out of model training but in Claude's search answers. Its robots.txt groups look like this:

User-agent: ClaudeBot
Disallow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

User-agent: Googlebot
Allow: /

The trap is a leftover Disallow: /pricing under User-agent: * from an old A/B test. Googlebot has its own group here, so it's fine, but any page blocked for Googlebot would also be skipped by Brave.

The render check. Rostra runs curl -s https://rostra.example/pricing | grep -c "per location" and gets 0. The plans load from a JavaScript widget. That's the first fix, because Claude's top-cited format after definitions is pricing.

The baseline. 25 prompts, 75 runs. Reading the log:

• If "best scheduling software for restaurants" cites three listicles and none names Rostra, step 7 is the job. Rostra's own page isn't in contention for that prompt.
• If "how to schedule split shifts legally in California" cites Rostra's blog, Claude already trusts that format. Rostra writes more how-to pages on labor rules.
• If "Rostra vs 7shifts pricing" cites the competitor's comparison page, Rostra has no comparison page of its own. It builds one with a server-rendered price table.
• If a Rostra page ranks well on Google but never appears in Claude, Rostra searches it on Brave. Missing there means submitting the URL and checking for Googlebot rules or noindex.

The log sheet. One row per run:

prompt_id | type | run | searched? | cited_urls (in order) | rostra_cited? | rostra_named? | competitors_named | in_brave_top10?

What doesn't get you cited by Claude

Paying for placement

There's nothing to buy. In February 2026 Anthropic said "Claude will remain ad-free"12. Citations come from search results, not sponsorship.

Blocking ClaudeBot and expecting to vanish (or to appear)

ClaudeBot is the training crawler5. Blocking it doesn't remove you from Claude's search answers, and allowing it doesn't put you in them. The bots that matter for citations are Claude-SearchBot and Claude-User, plus Googlebot because of Brave.

Using robots.txt to keep a page out of Brave

Brave says it doesn't use robots.txt to prevent indexing and points site owners to noindex instead8. If you need a page gone from Claude's results, robots.txt alone may not do it.

Seeding Reddit threads

It works for Perplexity and Google. For Claude, the Slate data shows no Reddit citations at all across 1,000 questions10. Put that effort into your own pages and trade press.

Schema as a Claude ranking factor

Anthropic has published nothing saying Claude reads structured data when choosing sources, and the citation is a quoted text passage. Keep schema for Google's sake. For Claude, the words in the HTML are what count.

How do you measure your visibility in Claude?

Track three numbers on your fixed prompt set each month: citation rate (share of runs citing your domain), mention rate (share of runs naming your brand) and the top cited URLs. Show run counts next to every percentage. Our guide to measuring AI visibility covers sample sizes, and the AI visibility tools comparison lists which trackers include Claude.

Then watch traffic. In GA4, look for sessions with claude.ai as the source and see which landing pages they hit. It's worth the five minutes: Goodie's 2026 report found Claude's share of AI referral sessions rose from 1.4% in May to August 2025 to 18.5% in March and April 2026, on a mostly B2B panel11. For a quick first read across engines, the free Arobis AI visibility checker shows whether AI engines name your brand for your category's questions.

How long does it take to show up in Claude?

Access fixes are quick to verify: once Brave has crawled a page you'll see it in Brave's results, and Claude can cite it from then on. Content changes show up as fast as Brave recrawls the page, which is why submitting updated URLs is worth doing. Getting named on other people's pages runs on their schedule, so give step 7 a quarter. Answers that come from training data change only with new models.

The same work carries over. Claude and Perplexity shared 13.6% of cited pages in the Slate test, the highest overlap of any pair10, so see our guides to ranking in ChatGPT, Gemini and Google AI Overviews for what differs.

Frequently asked questions

Sources