Brave Search builds its index from anonymous signals contributed by opted-in Brave browsers, not from anything you submit. There is no search console, no sitemap intake, and no IndexNow. A page enters Brave's index either when about 20 opted-in users visit it, or when one opted-in user sees it rank on Google, Bing, Yahoo, or DuckDuckGo. Because Claude's live web search appears to lean on Brave's independent index, whether Brave has discovered your page increasingly decides whether Claude can cite you at all. This is part of the crawlers-and-technical series.
The mechanics below come from Merj's mid-2026 deep dive, which read the open-source Web Discovery Project (WDP) client line by line, corroborated against Brave's own published documentation. Treat the exact thresholds as one consultancy's careful reading of open-source code. Directionally reliable, not a Brave spec sheet.
Why does Brave's index matter for AI visibility?
Because it is the discovery layer underneath Claude. As covered in how to get cited in Claude, independent evidence points to Claude's live search being powered by Brave's index. Anthropic listed Brave Search as a subprocessor in March 2025, and Google Cloud's documentation describes Brave Search as the third-party service behind Claude's web search tool.
The citation overlap is the striking part:
- Profound reported in June 2026 that 79.2% of Claude's citations sat in Brave's top-ten results for the same query, versus 34% for Google's own top ten: across roughly 35,000 citations over about 400 queries.
- That makes Brave discovery a precondition for Claude citation: a page Brave has never indexed is one Claude's search is unlikely to reach, no matter how well it ranks on Google.
Why care now? Because Claude is where business buyers are heading. Ramp's AI Index (April 2026 data, published 13 May 2026, tracking spending across 50,000+ U.S. businesses) showed Anthropic overtaking OpenAI in business adoption for the first time. 34.4% of businesses versus 32.3%, up from under 8% a year earlier. Ramp's figures measure paid business usage, not consumer share, and single-source spending data is directional, but the direction is clear enough that Claude's plumbing now matters commercially.
Whether Brave has discovered your page increasingly decides whether Claude can cite you at all. And unlike Google, you cannot submit your way in. You have to be found.
How does a page get into Brave's index?
Through two channels with very different bars, per Merj's reading of the WDP client. Neither involves submitting a URL.
| Channel | What triggers it | The bar | Best read as |
|---|---|---|---|
| Page channel (the hard door) | Opted-in users visit your URL directly | ~20 different users on different networks before Brave can decrypt the URL | A strong quality signal |
| Query channel (the easy door) | One opted-in user views a results page where you rank | A single search on Google, Bing, Yahoo, or DuckDuckGo | The lowest-friction path in |
The important takeaway: classic search visibility seeds Brave visibility, which seeds Claude's. If you already rank on a mainstream engine, an opted-in Brave user searching that term is enough to hand your URL to Brave. Being reachable on the open web is upstream of everything else.
What is the page channel (the hard door)?
The page channel discovers URLs that enough real people visit. Its privacy design is why the bar is high. Per Merj. Corroborated by Brave's own STAR writeup. Each visited URL is encrypted with STAR secret sharing (a scheme based on Shamir's Secret Sharing) before it leaves the browser. Brave's servers cannot recover the URL until roughly 20 distinct users have reported the same page; below that threshold, decryption is mathematically impossible.
Two anti-gaming details matter:
- Network deduplication. Reports are deduplicated by the last octet of the user's IP address, so an office full of people on one network counts as roughly one voice. You cannot manufacture a quorum from a single location.
- Exemptions. Bare homepages and pages a user reached by clicking a Brave Search result are exempt from the full quorum. The system already has reason to trust those.
For most brands, the page channel is not the practical lever: gathering 20 independent Brave users on distinct networks is not something you can engineer. The query channel is.
What is the query channel (the easy door)?
The query channel is where a single search puts your URL in front of Brave. Opted-in Brave browsers ship recognisers for Google, Bing, Yahoo, and DuckDuckGo results pages. Per Merj's analysis, when an opted-in user searches on one of those engines, their browser waits a random 1-20 minute delay, anonymously re-fetches the same results page with no cookies or identity, and reports the listed URLs. Provided at least four clean results survive sanitisation. There is no quorum: one user, one search, and a URL that ranks can reach Brave.
A server-side pattern file (cryptographically signed, polled hourly) controls which engines are harvested. Brave can switch a source on or off without shipping new code. The practical implication is durable regardless of that file's exact state:
- If you rank on any mainstream engine, you are already discoverable by Brave through the query channel.
- If you rank nowhere, the page channel's ~20-user bar is your only remaining route. A much harder path.
This is why classic answer-engine and search fundamentals still underpin AI visibility: they are what feed Brave, and Brave is what feeds Claude.
What can silently keep your page out of Brave?
Even a visited page must survive a verification fetch with strict, testable edges before it qualifies. Per Merj, these are the filters that quietly disqualify pages:
| Filter | Rule (per Merj's WDP reading) | Why it bites |
|---|---|---|
| JavaScript | The verification fetch reads raw HTML off the wire and does not execute JavaScript | Client-side-rendered content is invisible to discovery |
| noindex / canonical | Honoured only inside the HTML <head>; the X-Robots-Tag and Link HTTP headers are ignored |
Header-only directives are neither obeyed nor read |
| Redirects | Any redirect makes the URL ineligible. It must return 200 at the exact URL | A "helpful" redirect can drop you from discovery |
| Size and time | Responses over ~2 MB or slower than ~10 seconds are refused | Heavy, slow pages fail the fetch |
| URL cleanliness | Long query strings, many parameters, and risky path words (login, checkout, admin) are rejected |
Parameterised URLs may never qualify |
Then there is the crawler that fills gaps. Brave's own crawler, per its documentation, is gated on Googlebot-crawlability: "if a domain or page is not crawlable by Googlebot, then Brave Search's bot will not crawl it either", and it deliberately sends no distinctive user agent, to avoid being treated differently from Google. The unintended consequence, which Merj highlights: a "block all unknown bots" rule silently cuts Brave off, because Brave does not announce itself the way named AI crawlers do. This connects directly to whether your CDN is blocking AI crawlers. A default-deny bot rule can quietly remove you from Claude's reachable web without any error you would notice.
What should you do about it?
The levers are unglamorous and mostly things good technical GEO already covers. Sharpened by how Brave behaves:
- Rank somewhere mainstream. Google, Bing, Yahoo, or DuckDuckGo visibility is the query channel's fuel. This is the single highest-leverage move.
- Serve server-rendered HTML. Because the fetch does not execute JavaScript, critical content and facts must be in the raw HTML.
- Put directives in the
<head>. If a page should be indexable, keepnoindexout of the HTML and canonical tags accurate in the markup. Header-only signals are ignored. - Don't default-block unknown bots. Confirm your CDN and robots rules don't silently exclude a crawler that sends no special user agent.
- Keep URLs clean. Short paths, few parameters, no redirects on the pages you want discovered.
None of this is submission. Brave has no front door. It is being reachable and rank-worthy so the network of opted-in browsers finds you on your behalf.
Brave discovery is one link in a longer chain. Reachable, then discovered, then retrieved, then quoted, and Claude is one engine among several, each with its own plumbing. The work is to be discoverable and extractable across all of them, then measure where you actually appear. Tracking your citation share across Claude, ChatGPT, Perplexity, and Google's AI surfaces, sampled over time, is exactly what Buffy Intel does.