What Search Engine Does Claude Use? The Confirmed Answer (Plus What Most Guides Miss)

AI search
Written By The Danu Agency Team

AI search — Written by The Danu Agency Team

Quick answer: Claude's web search runs on Brave Search — confirmed directly by Anthropic's own Trust Center, not a third-party guess. Anthropic also lists a second, less-discussed sub-processor, TurboPuffer, whose exact role isn't fully public. Here's the paper trail, plus what most guides on this topic still get wrong.

A few weeks ago, we described Claude as working like a concierge, naming a short, reasoned shortlist instead of handing you a directory. One question we left open: who's actually staffing that research desk?

The answer is Brave Search. Anthropic's own Trust Center lists Brave Search as a sub-processor for "Web Search," scoped to all products, a disclosure they're legally obligated to keep accurate. It's been public since March 19, 2025. Within a day of Claude's web search launching, developer Simon Willison went a step further and found the receipts himself: an internal parameter called BraveSearchParams sitting right in Claude's own API tool definition. That's not an inference from behavior; that's Brave's name in the code.

There's a second name on the Trust Center list, too. Anthropic also names TurboPuffer, a vector and full-text search database built by former Shopify engineers, as a second "Web Search" sub-processor, added over a year later, on May 6, 2026. What exactly TurboPuffer handles in the pipeline — whether it's doing embedding-based retrieval on top of Brave's results, or something else — isn't publicly broken down anywhere, including by Anthropic. Its own site lists Anthropic as a customer and says nothing further. To be clear about where the certainty ends: Brave supplies the web index and organic ranking, and TurboPuffer's exact job is the one piece of this that stays genuinely unconfirmed.

On the "closely track" part, that's not just a vibe. Two separate research efforts have measured it. Profound's first test was run the same week Claude's web search launched. It found 86.7% overlap between Claude's citations and Brave's top organic results. A much larger follow-up study by MERJ, covering roughly 35,000 Claude citations across 400 queries, found 79.2% of the URLs Claude cites sit in Brave's top 10 — compared to just 34% for Google's top 10. If you're optimizing for Claude, Brave's rankings matter more than Google's do.

This is the actionable half of the story, and it's missing from nearly every "does Claude use Brave" post out there. Anthropic runs three distinct crawlers, each with its own robots.txt name, each doing a different job:

  • ClaudeBot collects content for model training.

  • Claude-User fetches a page when someone asks Claude a direct question.

  • Claude-SearchBot indexes your content for search results.

Anthropic says this directly: blocking Claude-SearchBot "may reduce your site's visibility and accuracy in user search results." Blocking ClaudeBot does nothing to that — it only opts you out of training data. If your robots.txt has a blanket "block all AI bots" rule copied from a 2023 template, there's a real chance you're blocking the one bot that actually helps you get found, while leaving the training bot — the one nobody's visibility depends on — untouched.

This isn't unique to Claude. ChatGPT runs the same three-tier structure (GPTBot, ChatGPT-User, OAI-SearchBot), and Perplexity runs a two-tier version (PerplexityBot, Perplexity-User). It's becoming the industry standard — though the tiers don't all behave the same way. Anthropic says all three of its bots, Claude-User included, honor robots.txt. OpenAI's own documentation says the opposite for its equivalent bot: because ChatGPT-User is triggered by a live user request, "robots.txt rules may not apply." Perplexity says the same about Perplexity-User. Claude is, at the moment, the strictest of the three about respecting robots.txt across the board.

Assistant Training bot User-fetch bot Search/indexing bot
Claude ClaudeBot Claude-User Claude-SearchBot
ChatGPT GPTBot ChatGPT-User OAI-SearchBot
Perplexity — Perplexity-User PerplexityBot

Common misconceptions, and where they're actually coming from

You'll still run into the claim that "Claude's search provider has never been confirmed." It's an understandable mistake — Claude's own web-search documentation genuinely doesn't name a provider anywhere, and the Trust Center is a separate page that's never linked from those docs. If you only check the search docs, "unknown" is a reasonable place to land. But it's also why the Trust Center wins as a source here: it's Anthropic disclosing its own vendor relationship, not someone piecing it together from the outside.

You'll also still see sites recommending "implement llms.txt to improve your Claude visibility." That was reasonable advice when llms.txt first started circulating. It hasn't held up since. Ahrefs analyzed 137,000 sites and found 97% of the llms.txt files on them got zero measurable traffic — not just from Claude, from anything. And on the Google side specifically, Gary Illyes said outright at a Search Central event that Google doesn't crawl or use the file at all, for rankings or for AI Overviews. Neither data point proves Claude ignores it too, but there's no evidence anywhere that llms.txt does what the advice claims, and there's now real evidence it mostly does nothing.

FAQ

Does Claude use Google or Bing to search the web?

No. Claude's web search runs on Brave Search, an independently built index — not a Google or Bing reseller — confirmed directly by Anthropic's own Trust Center.

Can I block Claude from training on my site without losing search visibility?
Yes. Disallow ClaudeBot in your robots.txt to opt out of training and leave Claude-User and Claude-SearchBot alone. That keeps you eligible to show up in Claude's search results.

Does ranking well on Google mean I'll show up in Claude?
Not automatically. Brave's crawler deliberately mirrors Googlebot's access — by Brave's own admission, if a page isn't crawlable by Googlebot, Brave won't crawl it either, specifically so it isn't blocked by sites that only allow Google in. Once it's crawlable, though, Brave ranks it with its own independent index and algorithm — it dropped its last dependency on Bing for web results back in April 2023, and there's no evidence it ever relied on Google's results for core search. So: Googlebot access is a baseline requirement, but a good Google ranking doesn't carry over.

Is TurboPuffer a second search engine Claude uses?
No, it's a database, not a crawler. It's a second named sub-processor under "Web Search," but that describes which Claude feature it supports, not what it does. Brave is the one confirmed to supply the actual organic web results.

Not sure whether your business even shows up in Brave, or whether your robots.txt is quietly blocking Claude-SearchBot? That's part of what we check. Get your free consultation, and we'll show you exactly where you stand.

Sources:

(verified, primary)

Anthropic / Claude

turbopuffer

OpenAI

Perplexity

Brave

Independent verification

Best practices for trusting anything AI tells you

AI makes things up sometimes. It hallucinates — states something with total confidence, points to "a source," and that source doesn't say it, or doesn't exist. While researching this post, one of our tools handed us a quote and a stat that sounded completely real. Neither one was.

Here's what we actually do before trusting an AI-generated answer:

  • Never take the first answer as final. Treat it as a lead to check, not a fact.
  • Make it show its sources. If an AI tool can't point to where a claim came from, that's a red flag on its own.
  • Go back and read the actual source. Not a summary of it — the real page. Summaries are where most fabrications hide.
  • Have it double-check itself. Ask it to re-verify its own claim against the source before you use it anywhere.
  • Throw out anything that doesn't hold up. If the source doesn't say it, the answer doesn't go out — no exceptions.

Simple rule: the more digging before you trust an answer, the more accurate that answer turns out to be. Most sites skip all five steps. We don't.

Want that same level of checking on your own online presence? Whether AI tools can actually find your business, and whether what they're saying about you is even true? That's exactly what we do.

Get Your Free Consultation


The Danu Agency Team

The Danu Agency team builds AI automations, SEO, and GEO strategies that help growing businesses save time, cut errors, and get found — by search engines and AI alike.

https://www.danu-agency.ai
Next

The Shortlist Test: How to Find Out If You're One of the Names AI Gives Your Next Client Body