How Does Claude Find Sources: Does It Use Brave Search?
Yes. When Claude searches the web, it draws on Brave Search. Anthropic lists Brave Search among the subprocessors it uses for web search, and citation data points the same way. One analysis of about 35,000 Claude citations found 79.2% came from pages in Brave's top 10 results, compared with 34% from Google's top 10. Claude does not search for every question. When it does, it turns the question into queries, reads a selection of results, and cites the pages that support its answer. For site owners, visibility in Claude depends on three things. Your pages need to rank in Brave, Anthropic's crawlers need access, and key content must be readable without JavaScript.
How does Claude find and choose sources?
- It decides whether to search. Questions about current events, prices or specific products usually trigger a search. General knowledge questions may be answered from the model alone.
- It writes search queries, sometimes several, based on the question.
- It gets results from Brave Search.
- It reads a selection of pages, fetching them to check their content.
- It writes the answer and cites the pages that support specific statements.
If web search is switched off, or the question does not need it, Claude answers from what it learned in training, and shows no sources.
How does Claude discover sources compared with traditional search?
| Traditional search | Claude with web search | |
|---|---|---|
| Index | The search engine's own, such as Google's | Mainly Brave Search's index, plus pages Anthropic's Claude-SearchBot crawls |
| What the user sees | A ranked list of links | One written answer, with citations when it searches |
| Queries | The user's own words | Queries Claude writes, sometimes several |
| Pages read | Whatever the user clicks | A handful of pages Claude opens itself |
| What wins | Ranking position | Ranking in Brave, plus a passage that answers the question clearly |
The practical difference is that ranking well in Google does not guarantee visibility in Claude. Brave's index and rankings differ from Google's. A page can rank in the top three on Google and be absent from Brave's first page.
How does Brave find pages?
Brave runs its own crawler, but it discovers many pages through its Web Discovery Project. Brave browser users who opt in contribute anonymised data about pages they visit. An analysis by MERJ found that pages discovered this way typically needed visits from about 20 distinct opted-in users. A single user was enough if the page had appeared in another search engine's results. Brave's "submit URL" option asks it to re-fetch a page it already knows. According to the same analysis, it does not add a new page on its own.
This means brand-new pages with little traffic can take time to enter Brave's index. Pages that already rank elsewhere and get real visits are found faster.
What should you check?
- Brave rankings: search your key questions on Brave Search and see where your pages appear.
- Anthropic's crawlers: Anthropic uses ClaudeBot for training, Claude-SearchBot for search, and Claude-User for fetches a user triggers. If you want to be cited, do not block Claude-SearchBot or Claude-User in robots.txt or your firewall. Blocking ClaudeBot for training is a separate choice.
- Readability: key text should be in the HTML, since most AI crawlers do not run JavaScript.
- Working URLs: Claude's fetches often hit missing pages. The Vercel and MERJ analysis found about a third of Claude's fetches ended in 404 errors, so keep redirects in place after site changes.
- Clear passages: a direct answer near the top of each section helps Claude pick your page once it is read.
Whether traditional rank trackers can check Brave is covered in whether traditional SEO trackers can check rankings in Bing and Brave. How assistants split one question into several searches is covered in what query fan-out is and how LLMs use it.
What might an illustrative check find?
Consider an illustrative accounting software company that appears in Google's AI Overviews for "best invoicing software for freelancers" but never in Claude's answers. On Brave Search, its comparison page does not appear in the first two pages of results. A newer competitor's page is third. The company's robots.txt also blocks all Anthropic crawlers, a rule added a year earlier to stop training. It changes the rule to block only ClaudeBot and allow Claude-SearchBot and Claude-User. It then works on its Brave visibility through internal links and a refreshed comparison page. Within two months, its page appears on Brave's first page, and Claude begins citing it in some answers.
What do people ask about Claude and Brave Search?
Does Claude only use Brave Search?
Not entirely. Text results line up closely with Brave's, and Anthropic lists Brave Search as a web-search subprocessor, but Anthropic also runs its own search crawler, Claude-SearchBot, Claude can fetch any page a user pastes, and Anthropic's help centre says image results are powered by Bing.
Can I submit my site to Brave?
Brave's submit option re-fetches known pages. New pages are found mainly through crawling, links and visits from opted-in Brave users.
Does blocking ClaudeBot stop Claude citing my site?
ClaudeBot is used for training. Search and user fetches use Claude-SearchBot and Claude-User, which you can allow separately.
Does Claude cite fewer sources than other assistants?
It depends on the question. Claude does not cite sources in every answer, and when it searches it can cite several. What matters is being among the pages it reads.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources