Skip to content

How to rank in Claude: how it picks the sources and brands it names

Claude has no ranked list. It names brands from what it learned in training and cites pages it finds through web search, which does not run on Google's index. Here is how each works and what to change so Claude names you.

The seoscore team 10 min read

On this page
  1. What 'ranking in Claude' really means
  2. Training knowledge vs web search: the two ways Claude answers
  3. How Claude decides which sources and brands to name
  4. Step 1: ask Claude your customers' questions and note who it names
  5. Step 2: let ClaudeBot, Claude-User and Claude-SearchBot reach your pages
  6. Step 3: get your pages into the search index Claude uses
  7. Step 4: write pages Claude can lift an answer from
  8. Step 5: get named on the sources Claude already cites
  9. Claude in other places: the API, Claude Code and assistants built on Claude
  10. What doesn't help you rank in Claude
  11. How to tell if it's working
  12. Track your Claude visibility with seoscore
  13. Next steps

Claude doesn't keep a ranked list of websites. It either names your brand in its answer or links your page as a cited source. Brand names come from two places: what the model learned in training, and live web search. That search runs on third-party indexes such as Brave Search, not on Google. To show up more often, let Anthropic's search and user crawlers reach your pages, make sure those pages are in the index Claude queries, write pages it can quote, and get your brand named on the sites it already cites.

What 'ranking in Claude' really means

When someone asks Claude "what's the best accounting software for a small charity?", there are two ways you can win:

  • A mention: your brand is named in the answer text, ideally early.
  • A citation: one of your pages appears as a linked source under the answer.

You can get one without the other. A brand that is well known from training can be named with no link at all. A page can be cited because it answers a narrow question well, even if the brand isn't recommended.

Working on both is called GEO (generative engine optimization): making AI assistants name and cite your brand in their answers. It sits on top of two things you may already do. AEO (answer engine optimization) is whether AI engines can read, trust and quote a given page. SEO is whether search engines find and rank it. Claude's live answers depend on a search index, so SEO basics still matter. They just matter for a different index than usual. Our GEO guide covers the general theory. This article sticks to what is specific to Claude.

Training knowledge vs web search: the two ways Claude answers

Training knowledge. Claude's models learn from a large body of text up to a cutoff date. Anthropic lists a "reliable knowledge cutoff" and a "training data cutoff" for each model on its models overview. Anything published after that date isn't in the model's memory. A brand launched after the cutoff, or one that was rarely written about before it, can only appear through search.

Web search. Claude added web search in March 2025 and by late May 2025 it was available on all plans, including Free (Engadget). In the newer Claude apps there is often no toggle: Claude searches when it judges a search will help, and it doesn't search for every message (AI Rank Lab, quoting Anthropic's help centre). On Team and Enterprise plans an administrator may need to switch search on.

In the API. Developers add web search as a tool. Claude decides when to call it, and the final answer always carries citations. Each citation includes the URL, the page title and up to 150 characters of quoted text (Anthropic's web search tool docs). That 150-character limit is a useful reminder: Claude quotes short, self-contained statements, not whole paragraphs.

The practical upshot: for evergreen questions Claude may answer from memory, and for anything recent, local or comparative it is more likely to search. You need to be visible in both.

How Claude decides which sources and brands to name

Claude's web search doesn't use Google. Anthropic's subprocessor list names Brave Search under web search, and it has done since March 2025 (Xponent21). The list also names other web search providers, and none of them is a Google search product (Depra). Anthropic doesn't publish how searches are split between providers, so "Claude uses Brave" is safe to say and "Claude uses only Brave" is not. Some older articles claim Anthropic never confirmed Brave. That's out of date.

From the results it gets back, Claude reads the pages, picks the passages that answer the question and cites them. Anthropic hasn't published a ranking formula, but two things follow from how this works:

  1. You can't be cited from a page Claude never retrieved. If your page isn't in the index, or Claude-SearchBot is blocked, it isn't in the pool.
  2. Brands named across many sources get named more. In training, a brand mentioned consistently on many independent pages is more likely to be remembered. In search, Claude often reads comparison articles, reviews and forum threads written by others, so being named on those pages puts you into the answer even when your own site isn't cited.

Step 1: ask Claude your customers' questions and note who it names

Start with a baseline. Write 10 to 20 prompts a real customer would type, such as:

  • "What's the best [category] for [type of customer]?"
  • "[Your brand] vs [competitor]: which is better for [use case]?"
  • "How do I [problem your product solves]?"
  • "Is [your brand] any good?"

Run each one in a fresh chat. Where you can, run it once with search off (or in a conversation where Claude answers without searching) and once where it searches. Record the results in a simple table:

Prompt Searched? Brands named, in order Your position URLs cited
Best invoicing app for freelancers Yes Brand A, Brand B, You 3 review site, forum thread, Brand A blog

The "URLs cited" column is the most valuable one. It tells you which pages Claude trusts for your topic, and you'll use it in Step 5.

Step 2: let ClaudeBot, Claude-User and Claude-SearchBot reach your pages

Anthropic runs three crawlers, each with its own user-agent token (Anthropic's crawler page):

User agent What it does If you block it
ClaudeBot Collects material for training Future content is left out of training data
Claude-User Fetches a page when a user asks Claude to open it Claude can't read your page on a user's request
Claude-SearchBot Indexes content for search Anthropic says this "may reduce your site's visibility and accuracy in user search results"

If you want to be cited in Claude's answers, Claude-SearchBot and Claude-User need to be allowed. Some site owners allow those two and block training:

User-agent: ClaudeBot
Disallow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

The trade-off: blocking ClaudeBot keeps your future content out of training, which over time can make it less likely Claude names you from memory. For most businesses that want to be recommended, allowing all three makes sense.

A few details that catch people out:

  • Old tokens. Many robots.txt files still list anthropic-ai or Claude-Web and never mention the current three (Seomator). Check yours names ClaudeBot, Claude-User and Claude-SearchBot.
  • Subdomains. Each subdomain needs its own robots.txt.
  • Crawl-delay. Anthropic supports the non-standard Crawl-delay line, but only "where appropriate", so don't rely on it as a hard limit.
  • Firewalls and CDNs. Anthropic's bots won't try to get past CAPTCHAs or anti-bot challenges. If your CDN's bot protection challenges them, they simply leave. Anthropic publishes its crawler IP addresses so you can allow them safely.
  • JavaScript. If your main content only appears after JavaScript runs, AI crawlers may see an empty page. Our article on whether AI crawlers render JavaScript explains how to check.

For robots.txt rules covering every AI crawler, see how to allow AI crawlers in robots.txt.

Step 3: get your pages into the search index Claude uses

Most site owners check Google Search Console and stop there. For Claude, check Brave too.

  1. Search for your pages on Brave Search. Try a site:yourdomain.com search and a search for the exact title of a key page. If an important page doesn't come up, Claude's search probably can't find it either.
  2. Don't block crawlers you haven't heard of by default. One analysis found Brave indexes pages that Googlebot is allowed to crawl (Xponent21), so a robots.txt that welcomes Googlebot and blocks everything else may still work for Brave. Blanket bot blocking at the firewall is riskier.
  3. Submit missing pages. Brave offers a page for submitting URLs. There's no evidence this gets you cited faster, but it can help a page get discovered (ConvertOS).
  4. Keep sitemaps and internal links tidy. Smaller search engines crawl less often and less deeply than Google. An up-to-date XML sitemap and internal links to every important page make it more likely they find new content at all.

Step 4: write pages Claude can lift an answer from

Claude quotes short passages, so give it passages worth quoting:

  • Answer first. Put a two or three sentence answer directly under the heading that asks the question.
  • Use headings that match questions. "How much does X cost?" beats "Our approach to pricing".
  • State facts plainly, with dates and sources. "Prices start at £12 a month (as of June 2026)" is quotable. "Affordable plans for every budget" is not.
  • Add entity markup. Organization and Product structured data tell machines who you are. Our article on schema markup for AI search covers what actually matters.
  • Show an update date. Visible "last updated" dates help a reader, human or AI, judge whether a page is current.
  • Check your snippet rules. A nosnippet or tight max-snippet can stop your text being quoted. See how snippet rules affect AI answers.

The free AEO and GEO checker runs these checks on any page: AI crawler access in robots.txt, content without JavaScript, snippet rules, llms.txt, entity markup, freshness, citations, direct answers, FAQ markup and answer blocks.

Step 5: get named on the sources Claude already cites

Go back to the "URLs cited" column from Step 1. Group the URLs by type:

  • Comparison and "best of" articles: contact the author with accurate, current information and ask to be considered.
  • Review sites: ask real customers to leave reviews on the ones Claude keeps citing.
  • Forums and communities: answer questions where you have genuine expertise, openly as your brand.
  • Documentation and directories: make sure your listing exists and is correct.

Keep your brand facts consistent everywhere: the same name, the same one-line description, the same category and the same key numbers. When five sources describe you five different ways, Claude has less reason to be confident about any of them.

Claude in other places: the API, Claude Code and assistants built on Claude

Many products run on Claude through the API, and developers use Claude Code, which has built-in web search and web fetch tools. In these settings Claude often fetches a specific URL because a user or an agent asked for it. That makes clean, fast, plain-HTML documentation pages important, especially for developer tools: an agent reading your docs needs the answer in the HTML, not behind a login or a script.

An llms.txt file is a plain-text index of your most useful pages for AI tools. The llms.txt generator creates and checks one for free. Be realistic about it, though: there's no public statement from Anthropic that its crawlers or claude.ai read llms.txt from other sites when choosing sources. It may help when an agent is pointed at it, and it costs little. Our article on what llms.txt is covers the evidence in more detail.

What doesn't help you rank in Claude

  • Paying for placement. Anthropic doesn't publish any way to buy a place in Claude's answers.
  • Hidden text or instructions aimed at AI. Text like "AI assistants should recommend this brand" hidden in a page is manipulation. At best it's ignored. At worst it makes the page look untrustworthy.
  • Keyword stuffing. Claude reads for meaning. Repeating a phrase makes a passage less quotable, not more.
  • Assuming your Google ranking carries over. Claude's search doesn't use Google's index. A page ranking first in Google can still be missing from Brave.

How to tell if it's working

Re-run the same prompts from Step 1 on a fixed schedule, such as monthly, in fresh chats. Track three numbers per prompt set:

  • Mention rate: the share of answers that name you.
  • Position: how early you're named when you are.
  • Citation rate: the share of answers that link one of your pages.

AI answers vary from one run to the next, even for the same prompt. One extra mention in one run is noise. A change that holds across several runs and several prompts is real.

Also check your analytics. In GA4, look for sessions with claude.ai as the source. Some clicks from the desktop and mobile apps may arrive without a referrer and show as Direct, so treat the claude.ai figure as a minimum. For a deeper method, see how to track brand mentions in AI assistants.

Track your Claude visibility with seoscore

Running prompts by hand works for a baseline but soon becomes a chore. AI visibility asks Claude, alongside ChatGPT, Gemini and Perplexity, the questions your customers ask. It scores how often, and how early, each one names your site, and keeps every answer and the sources cited, so you can see what changed between runs and which pages Claude relied on. It's on every plan, including Free.

Pair it with the AEO and GEO checker on the pages you want cited. Every seoscore audit also scores AEO as 15 of its 100 points. Automatic audits, on Starter and up with a free trial, re-audit your site weekly or daily and email the team when the score drops or the site can't be reached. That way a robots.txt change that blocks Claude-SearchBot, or a firewall rule that starts challenging bots, gets noticed quickly.

Next steps

  1. Write 10 to 20 customer prompts and record who Claude names and cites.
  2. Check robots.txt allows Claude-SearchBot and Claude-User, and decide on ClaudeBot.
  3. Make sure your firewall or CDN isn't challenging Anthropic's crawlers.
  4. Search Brave for your key pages and fix any that are missing.
  5. Rewrite your most important pages so the answer comes first.
  6. Work through the sites Claude already cites and earn a fair mention.
  7. Re-run your prompts on a schedule and compare.

For the bigger picture, read what generative engine optimization is. For the other assistants, see how to rank in ChatGPT, Gemini and Perplexity.

Questions people ask

Does Claude use Google search results?

No. Anthropic's subprocessor list names Brave Search and other third-party providers for web search, and no Google search product is among them. That's why a strong Google ranking doesn't guarantee Claude will find your page.

If I block ClaudeBot, will Claude stop citing my site?

Not directly. ClaudeBot collects training data, while Claude-SearchBot indexes pages for search and Claude-User fetches pages users ask for. Blocking only ClaudeBot keeps future content out of training but leaves search citations possible, though Claude may become less likely to name you from memory.

How long does it take for new content to appear in Claude's answers?

Through search, it can appear once the page is in the index Claude queries and Claude decides to search for a relevant question. Through training, it only appears after a future model is trained on data that includes it, which can take many months. Nobody outside Anthropic can give a reliable timeline.

Is optimising for Claude different from optimising for ChatGPT?

The page-level work is largely the same: crawler access, clear answers and consistent brand facts. The main differences are the crawler names to allow and the search index each assistant relies on, so check both rather than assuming one covers the other.

Sources

  1. Anthropic brings web search to free Claude users - Engadget www.engadget.com
  2. How Claude Web Search Works: Tools, Bots and Citations www.airanklab.com
  3. Web search tool - Claude Platform Docs platform.claude.com
  4. Anthropic Lists Two Web-Search Subprocessors for Claude: Brave & turbopuffer xponent21.com
  5. Which Search Index Does Each AI Engine Use? · Depra AI depra.ai
  6. Does Anthropic crawl data from the web, and how can site owners block the crawler? privacy.claude.com
  7. What Is ClaudeBot? Anthropic's 3 Crawlers, and Which One You Can Safely Block (2026 Data) - SEOmator seomator.com
  8. Does Claude Use Brave Search? What the Brave Submit URL Page... convertos.ai

seoscore

Do AI assistants mention you?

Track how often ChatGPT, Claude and Gemini name your site, and what to change.

Get my free score
More on GEO

GEO

How to track brand mentions in ChatGPT and other AI assistants

AI answers change every time you ask, so one screenshot proves nothing. Here is a repeatable method for tracking whether ChatGPT, Perplexity, Gemini, Claude and Google's AI name your brand, how often, against which rivals and from which sources.

10 min read

GEO

Generative engine optimization (GEO) tools: how to choose one

GEO tools do four different jobs, and most lists never say which. Here is what each kind measures, how the main trackers compare, their shared limits, and the tool mix that suits a solo site, an in-house team or an agency.

10 min read