Claude Web Search Runs on Brave. Here's How to Rank

A content team watches their Google traffic slide. AI answers are eating the clicks. So the obvious question comes up in the meeting: if not Google, where do we optimize now?

The answer almost nobody is ready for: Brave.

Yes, the privacy browser. It quietly powers the web search behind Claude, one of the most used AI assistants on the planet. Which means a search engine most of your customers have never opened now helps decide whether an AI recommends you. If you have spent a decade learning Google, this is the plot twist of the year, and it is very fixable once you see it.

Why does Claude’s web search run on Brave?

When you ask Claude a live question like “best CRM for a small agency” or “top project management tools,” it does not crawl the open web in real time. It sends your question to a search API and reads the results back. That API is Brave Search.

This is on the record, not a rumor:

• Anthropic added Brave Search to its public subprocessor list on March 21, 2025 (Simon Willison, TechCrunch). A subprocessor is a partner that handles data on Anthropic’s behalf. Brave is on that list.
• Claude’s web-search tool contains a parameter literally named BraveSearchParams, and testers have watched Claude return the exact citations Brave returns for the same query.
• Brave is doing this on purpose. Its LLM Context API is sold as “the leading search tool for applications that use Claude,” and Brave even publishes a guide for wiring its search into Claude.

So Brave is not a side character. It is the index Claude reads from. If a page is not in Brave, it is not in Claude’s web answers.

How much does Brave decide what Claude cites?

Enough that you should treat Brave rank and Claude citation as nearly the same thing.

The cleanest number comes from Profound, surfaced by analyst Josh Blyskal and reported by Search Engine Land: 79.2% of Claude’s citations come from Brave’s top 10 results. The overlap between what Claude cites and what Brave ranks is around 86.7%, with little reranking in between.

Sit with that for a second. Claude is not running secret magic to pick its sources. For roughly four out of five citations, it is handing back what Brave already put on page one. The old skill of ranking on a search engine did not die. It moved to a search engine you have probably never optimized for.

If Brave ranks you, Claude tends to cite you. If Brave never indexed you, you do not exist to Claude’s web search.

How is this different from classic SEO, AEO, and GEO?

Search optimization has moved through three eras, and Brave adds a fourth layer on top. Here is the honest map.

The big shift is the number of referees. Classic SEO had one: Google. AEO taught us to write answers, not just pages. GEO told us to think about how AI models understand a brand, which is right, but it stayed abstract, and a lot of people turned it into “add schema and an llms.txt and hope.” Google itself has pushed back on much of that advice, which is worth reading about in why Google killed the GEO hack industry.

The Brave discovery makes GEO concrete again. Each AI assistant has a search backend. Claude’s is Brave. So “optimize for AI” stops being a vibe and becomes a specific, testable task: rank on Brave. You can search it, see your position, and fix it. That is a far better job than guessing what a language model thinks of you.

How does Brave decide what to rank?

Brave is not a Google or Bing mirror. It removed the last piece of Bing from its results and now runs a fully independent index. That changes which levers move the needle, and the most interesting one is genuinely new.

It is called the Web Discovery Project. It is an opt-in, anonymous system inside the Brave browser. When users turn it on, Brave collects anonymous signals about the pages they visit, including the URLs and the time spent on each page, and uses that to decide which pages are worth indexing and how they should rank (Brave Help Center). The data is unlinkable, with no user or session ID attached.

Put plainly: Brave’s index has a built-in human signal. Pages that real people open and actually read get a vote. Thin pages stuffed with keywords do not. On top of that, Brave’s search API is built to feed AI cleanly. It does more than turn a page into text. It pulls structured data like JSON-LD and tables, extracts forum discussions, and handles video captions, then ranks the most useful chunks for the model to read.

So the modern playbook is not “chase more backlinks.” It is: be genuinely useful, be machine-readable, and be present where people actually discuss your topic.

How do you actually get into Brave’s index?

Here is the part that trips everyone up, because it is the opposite of how Google works. Brave has no submit button. There is no Brave Search Console, no sitemap upload, no “request indexing.” Site owners have been asking Brave for webmaster tools for years, and the answer is still a pile of open feature requests, not a product.

So how does a page get in? Two ways, and you control both:

1. Brave’s crawler finds it. Brave runs its own crawler over the open web (site owners know it as Bravebot). If your robots.txt blocks it, you are invisible. Let it through.
2. Real Brave users visit it. Through the Web Discovery Project, a URL is only pulled into the index after it has been “visited independently by a large number of people,” anonymized through Brave’s STAR protocol (Brave). One person opening your page, even you, does not move the needle. It takes a crowd. That is the human gate, and it is why low-traffic pages quietly never show up.

Brave also throws pages out. It will not index a page that sits behind a login or paywall, a page the owner marked noindex, or a messy one-time link with tokens in the URL (Brave). Keep the pages you care about public, clean, and indexable, or they never make the cut.

One more trap to avoid: Goggles and the Google fallback are not back doors into the index. Goggles only re-rank pages Brave already has. The Google fallback is a stopgap Brave borrows when its own results are thin, and Brave admits it is still refining its index. Neither one puts a new page in front of Claude. Only the index does that.

Put it together and the rule is blunt: Brave does not index sites you submit, it indexes sites people visit. To exist in Claude’s web search, you need a public, crawlable page that enough real humans actually open.

How do you get ready for Brave and Claude?

Here is a simple checklist anyone can run. None of it is hard, and the first two steps are free and take five minutes.

1. See what Claude sees. Go to search.brave.com and run the query a customer would type. That result page is roughly Claude’s source list.
2. Check your coverage. Search site:yourdomain.com on Brave. If your important pages are missing, that is the first problem to fix, before any on-page tweak.
3. Skip the sitemap. There is no submit button. Brave has no webmaster tools, so you cannot submit your way in. Instead, make sure the crawler can reach you (step 7) and that real people visit your pages (step 6). Coverage on Brave is earned, not requested. A great page nobody opens stays out of the index.
4. Make pages machine-readable. Server-render your content so it appears without JavaScript, add JSON-LD schema, and include at least one comparison table on important pages. Brave extracts structured data and tables cleanly, so a table is free citation surface.
5. Connect your entity. List every profile you own, including YouTube and social accounts, in your schema sameAs block so the engine knows they are all the same person or brand.
6. Earn the human vote. Post genuinely useful answers on Reddit and niche forums in your field. Brave pulls forum threads and weighs time-on-page, so real engagement off your site feeds your ranking on it.
7. Welcome the crawlers. Make sure your robots.txt allows search and AI bots. If you block them, none of the above matters.

If you already keep up with answer-style content, you are partway there. The same instinct behind adapting when Google killed FAQ rich results applies here: the platforms change, but clear, structured, genuinely helpful pages keep winning.

The takeaway

For a decade we optimized for one search box and called it SEO. AEO and GEO were the right instincts that got a little lost in jargon. The Brave story snaps it back into focus: the AI your customers trust is reading from a real search index, and for Claude that index is Brave.

The good news is that this is the most checkable thing in modern search. You do not need a budget or a consultant to start. Open Brave, search the thing your buyers would ask, and look. What you find there is close to what Claude already believes.

If you want a deeper read on how AI engines see a website end to end, that is exactly what the free AI discoverability audit was built to check.