Plumbing

What search engine does Claude use? Anthropic names three.

By Arnav Mukherjee, founder of TofuBofu · October 29, 2026

Every scan we run offers Claude its web search tool, so I care which index sits behind it. The standard answer in this category has been one word since March 2025: Brave. This week I went back to Anthropic's own pages to check it.

Brave is still there. It's no longer alone. Anthropic's subprocessor list now names three companies whose job is web search, and the newest one was added on 15 October. If your AI visibility plan for Claude says "get into Brave", it covers one of three doors.

What does Anthropic say it uses?

Anthropic publishes its subprocessors in its Trust Center. I read the list on 29 October 2026. Three entries carry the purpose "Web Search":

Anthropic's web search subprocessors, read at its Trust Center on 29 October 2026
Provider Added What it says it is Its crawler
Brave Search 19 March 2025 Its own independent web index No separate user agent; follows Googlebot's access
TurboPuffer 6 May 2026 Vector and full-text search database None of its own
Parallel Web Systems 15 October 2026 Its own web-scale index for AI products ShapBot

Anthropic's update notices describe TurboPuffer and Parallel in identical words: each "provides web search for all Anthropic products except Claude for Government." The Brave date comes from Simon Willison, who caught the listing two days after it appeared. He then asked Claude to search for pelican facts, and "the ten results it showed as citations were an exact match for that search on Brave."

Images come from a fourth place. Anthropic's web search help page says "Image search is powered by Bing." No Google search product appears among the web search entries.

What Anthropic doesn't publish is the routing: which provider answers which question, or how results get merged. Anyone who tells you the split is guessing. So am I, below, and I'll label it.

What is each provider doing?

Brave is an index with Google's gate on it. Brave's API page says "Every result comes from Brave's own independent index, crawled and ranked by Brave," with "Zero dependencies on Google or Bing." Its crawler page adds the catch: "if a domain or page is not crawlable by Googlebot, then Brave Search's bot will not crawl it either." Brave's crawler doesn't announce itself, so you can't write a Brave rule in robots.txt. Your Googlebot rule is your Brave rule.

Parallel is a second independent index. Parallel says "We run our own web-scale index with billions of pages and millions more added and updated daily." Its crawler page names the bot: "ShapBot helps discover and index websites for Parallel's web APIs." Unlike Brave, ShapBot identifies itself, so robots.txt controls it directly.

TurboPuffer is storage, and here's my read. TurboPuffer calls itself "vector and full-text search built on object storage" and lists Anthropic as a customer. It crawls nothing. A database filed under web search suggests Anthropic keeps an index of its own there. The obvious feed is Anthropic's own crawler, Claude-SearchBot, which "navigates the web to improve search result quality for users." Anthropic hasn't connected those two dots in public. I'm connecting them because the practical advice comes out the same either way: let Claude-SearchBot in.

Claude web search routing not published Brave Search own index Parallel own index, new 15 Oct TurboPuffer search storage Bing images only Googlebot access Brave follows it ShapBot allow in robots.txt Claude-SearchBot my read: likely feed Anthropic Trust Center and help centre, read 29 October 2026. Dashed lines: images, or inference.

Three doors into Claude's web search, and the crawler that opens each one.

See whether Claude names you today. Our free scan asks real buying questions across five AI engines, Claude included with web search on, and shows who gets named instead. No card, no call.

Check your AI visibility free

What should a B2B firm check this week?

Four checks, about an hour in total. None needs a developer beyond a robots.txt edit.

  1. Read your robots.txt for five names. Googlebot, Claude-SearchBot, Claude-User and ShapBot should be allowed. ClaudeBot is training, so that one's your call. Anthropic's crawler page says disabling Claude-SearchBot "may reduce your site's visibility and accuracy in user search results."
  2. Check the firewall too. A robots.txt allow means nothing if a bot-blocking rule at your CDN refuses the request. Our AI crawlers guide covers where those silent blocks hide.
  3. Search your five most important pages on search.brave.com. Use the page's own title. If one's missing, Brave takes re-fetch requests. Brave has no console beyond that.
  4. Grep last month's server logs for ShapBot and Claude-SearchBot. No hits means neither crawler has read you lately. That's your Claude gap, measured before you write a word of content.

Then ask Claude the buying questions your customers ask, with search on, and read the citations. Anthropic's help page says "Every response includes citations." Those citations came back through Claude's search tool, and Anthropic names three providers behind it. If a rival's comparison page keeps appearing and yours never does, the problem most likely starts at retrieval, before anything on your page matters. Our read of the latest citation study points the same way for ChatGPT: which pages its search retrieved explained far more than anything on the page.

One warning about the category's own advice. Plenty of AEO tools treat Claude as an afterthought, and some don't track it below their enterprise tier. A plan built on a single-provider answer from 2025 will miss the two providers Anthropic added in 2026. Re-read the source list every quarter. It moved twice this year.

Frequently asked questions

What search engine does Claude use?

Anthropic doesn't name one engine. Its Trust Center lists three subprocessors whose job is web search: Brave Search, TurboPuffer and Parallel Web Systems. Parallel was added on 15 October 2026 and TurboPuffer on 6 May 2026. Anthropic also runs its own crawler, Claude-SearchBot, which it says indexes content for search. It doesn't say which provider answers which question.

Does Claude use Google or Bing for search?

Anthropic's help centre says image search is powered by Bing. No Google search product appears among the web search entries on its subprocessor list. Brave, the longest-standing provider, says every result comes from its own independent index with zero dependencies on Google or Bing. Google still matters indirectly: Brave's crawler won't crawl a page that Googlebot can't crawl.

What is Parallel Web Systems and what is ShapBot?

Parallel Web Systems sells web search to AI products and says it runs its own web-scale index with billions of pages, updated daily. Its crawler is ShapBot, and its documentation asks site owners to allow ShapBot in robots.txt. Anthropic added Parallel to its list of web search providers on 15 October 2026.

Why is TurboPuffer on the list if it isn't a search engine?

TurboPuffer describes itself as vector and full-text search built on object storage, a database, and lists Anthropic as a customer. Anthropic files it under web search. My read is that it stores an index Anthropic builds itself, plausibly from Claude-SearchBot's crawl. Anthropic hasn't said so, and that part is my inference.

How do I get my website into Brave Search?

Brave has no webmaster console. Its crawler uses no separate user agent and won't crawl what Googlebot can't, so keep Googlebot allowed. Search your key pages on search.brave.com, and submit any that are missing through Brave's re-fetch form at search.brave.com/submit-url. Note that robots.txt doesn't stop Brave indexing a page; a noindex tag does.

Should I block ClaudeBot?

That's a training decision, separate from search. ClaudeBot gathers training data. Claude-SearchBot indexes pages for search and Claude-User fetches pages when a person asks. Anthropic says disabling Claude-SearchBot may reduce your visibility in user search results, and disabling Claude-User may reduce it for user-directed web search. Block training if you want; keep the other two allowed.

Does Claude search the web on every question?

No. Anthropic's help centre says Claude searches the web when it helps, typically for questions that need current information, and that users can nudge it by writing 'search the web'. Questions it answers from memory use training data instead. To see what a buyer sees, ask with 'search the web', and read the citations every searched answer carries.

Sources and further reading

Related reading

AI crawlers explained: who reads your site
What makes AI engines cite a page? Not your FAQ block.
How Claude decides which companies to recommend

Arnav Mukherjee

Founder, TofuBofu