Does Claude use Brave Search? What a site owner can and cannot check

Short answer: Anthropic lists Brave Search, but its help page names no provider

On 25 October 2026 we searched Anthropic's subprocessor list in a browser for “Brave”. It has one entry: “Brave Search”, purpose “Web Search”, products “All products”, location United States. That is all it says. It does not say how Claude picks, ranks or combines results. Anthropic's help article What is web search? describes the feature without naming a search provider, so “Anthropic documents Brave as its search engine” is more than the sources support.

Others have looked closer. In March 2025, Simon Willison and TechCrunch reported that Brave Search had appeared on the list. Willison also reported that the ten citations Claude returned for one query matched Brave's results for the same query, and that Claude's search tool definition carried a parameter set called “BraveSearchParams”. TechCrunch noted that Anthropic had not formally announced the change. These are reports, not Anthropic documentation.

What is confirmed and what is not

What Brave says about its crawler

Brave's help page on its crawler says the Brave Search crawler “does not advertise a differentiated user agent”, because Brave must avoid discriminating against sites that allow only Google to crawl them. It adds that if a domain or page is not crawlable by Googlebot, Brave Search's bot will not crawl it either. It also says robots.txt is not used to prevent a page from being indexed: to remove a page you use noindex, and Brave has to fetch the page again before it sees that. Brave describes part of its index as fed by opt-in users of its Web Discovery Project. Brave's Search API page says its results come from its own independent index.

• There is no Brave line to look for in your access log, because the crawler does not name itself.
• There is no Brave token to allow or block in robots.txt. Some bot directories list a “Bravebot” token; Brave's page does not confirm one, so we do not have a bot page for it.
• Brave visibility follows Googlebot access. A rule that keeps Googlebot out, or a noindex, affects Brave as well, according to Brave.

What Anthropic's own bots did on four new sites

Anthropic runs three bots: ClaudeBot, which it describes as collecting content that may be used for training, Claude-SearchBot, which it says improves the quality of Claude's search results, and Claude-User, which opens pages when a person asks Claude something. We counted their requests on four small sites we run, between 19 and 25 October 2026. All four domains were new.

For comparison, OAI-SearchBot made 2,360 requests in the same logs. The sample is four sites, seven days and new domains, so it says little about Claude in general. It does show one thing: on new sites, Anthropic's training crawler turned up within a week and its search crawler did not, so you should not assume that an Anthropic bot has read your new pages. The study behind the “from the published ranges” column, including how many requests borrowed a bot's name, is on the fake bots page.

Five checks you can run today

• Ask your server what it returns to Googlebot's name: curl -A "Googlebot" -I https://example.com/your-page. You want 200 and no X-Robots-Tag: noindex. This is only a header test. It sends the name from your machine, so it does not prove that Google can reach you; the URL Inspection tool in Google Search Console answers that.
• Open your robots.txt and make sure no group disallows Googlebot for the page. The checker evaluates the file for every bot we track, and robots.txt basics explains how groups are chosen.
• Look for a noindex in the HTML: curl -s https://example.com/your-page | grep -i 'name="robots"'. Nothing printed, or a value without noindex, is what you want.
• Search for your domain on search.brave.com by hand, for example site:example.com. Pages that appear are in Brave's index. A page that does not appear is not proof of a problem, because new pages take time. Run it once; do not script Brave's search.
• Ask Claude, with web search switched on, a question that only your page answers, and read the sources it cites. That is one observation, not a measurement.

To see Anthropic's bots on your own server, search the log and compare the address with the operator's list. zgrep -hE 'ClaudeBot|Claude-SearchBot|Claude-User' /var/log/nginx/access.log* | awk '{print $1, $7, $9}' prints address, path and status. Anthropic publishes one list of ranges for all its bots at claude.com/crawling/bots.json, so it cannot tell Claude-User from ClaudeBot. The script in the 48-hour log guide checks addresses against it, and the user-agent lookup tells you which bot a logged string names.

If a check fails

Checked 25 October 2026

Sources: Anthropic's subprocessor list (a JavaScript page without a stable link to one entry: search it for “Brave”), Anthropic's articles What is web search? and Does Anthropic crawl data from the web?, Brave's crawler page and Search API page, and the reports by Simon Willison, TechCrunch and Xponent21. Our log counts are from 19 to 25 October 2026. These pages change, so check the originals before you rely on a detail.