Most brands working on AI visibility are running the right plays for the wrong system. They publish, they structure, they cite, they wait for ChatGPT and Claude to start surfacing them. ChatGPT eventually does, because ChatGPT searches Bing. Claude often doesn't, because Claude doesn't search Bing. Claude searches Brave.
Anthropic has never officially confirmed this. The link came out in March 2025 when TechCrunch reported that Brave Search had appeared on Anthropic's subprocessor list. A day later, Profound ran a citation overlap study using 15 representative queries and found that 13 of the 15 URLs Claude cited matched the top results Brave returned for the same query. The overlap, 86.7%, was statistically significant at p < 0.0001. The sample is small and the study is from the week Claude web search launched, so it should not be treated as gospel. But Claude's web search has continued to operate in a manner consistent with that finding, and our own GA4 data on Claude referrers points to the same pattern. The working assumption is: Claude's citations track Brave's organic rankings closely.
Why this matters more than the share figures suggest
Brave handles about 43 million search queries per day at last reliable count. Google handles roughly 8.5 billion. By volume, Brave is rounding error in the broader search market. If you only cared about direct organic traffic, you would barely think about Brave at all.
For Claude visibility, Brave is the gating condition. The chain that runs every time Claude generates a response with citations looks like this: user asks Claude something, Claude formulates an internal query, Claude sends that query through its search backend (Brave), Claude reads the top results, Claude summarises and cites them. The page either appears in Brave's top results for that internal query, or Claude doesn't see it. The bit most brands miss is that this happens whether or not they have invested in AI visibility at all. Brave indexation determines whether the rest of the chain even runs.
Brave isn't a smaller Google
This is the part where the operational guidance gets practical, because the assumption that Brave optimisation is just Google optimisation at a smaller scale is wrong.
Brave runs its own crawler and its own index, around 40 billion pages, with roughly 100 million pages added or refreshed every day. They phased out their Bing fallback completely by 2024, so the index now operates independently of both Google and Bing. The index is deliberately smaller than Google's, and Brave is explicit that this is to reduce SEO spam and prioritise content quality. Sites that lean on programmatic SEO scale, syndicated thin pages, or aggressive link-graph manipulation will see less of that work pay off in Brave than in Google.
Part of Brave's index comes from the Web Discovery Project, an opt-in feature that lets Brave browser users anonymously share data about pages they visit. Pages that real Brave users actually open get refreshed and re-ranked faster than pages the crawler only reaches occasionally. This creates a quiet network effect favouring content that draws traffic from Brave's user base, which skews technical, professional and privacy-aware.
The other meaningful difference is editorial. Brave's published policy says the index downweights mass-produced content. In practice this means original analysis, named authors, primary source citations and varied prose all appear to help. The usual AI-content tells (uniform paragraph rhythm, generic listicle scaffolding, transition-phrase padding) appear to hurt. This is harder to confirm with controlled testing from outside the system, but it lines up with what multiple independent SEO analyses have observed and what Brave themselves say they reward.
One detail worth knowing: there's no Brave Search Console. No indexation dashboard, no impressions data, no canonical declarations. The two levers brands actually have are the manual submission form at search.brave.com/submit-url and the rule that Brave's crawler defers to Googlebot, so if Google can crawl a page, Brave can too.
What to do about it
Submit your priority URLs to Brave directly
The form at search.brave.com/submit-url takes one URL at a time and uses Cloudflare Turnstile to keep submission manual. Ten to twenty URLs takes about ten minutes. Submit the homepage, the core product pages, and the pillar content you actually want Claude to cite. There is no API for bulk submission.
Audit which of your pages Brave actually has
Run site:yourdomain.com in Brave and compare against the same query in Google. Pages indexed in Google but missing from Brave are blind spots for Claude. Pages present in both but ranking materially lower in Brave are the next thing to look at.
Read your content as if Brave's algorithm is allergic to AI tells
Because it might be. The pattern is consistent enough across recent analyses that it's worth applying as operational guidance even without controlled testing. Look for the obvious markers: uniform sentence and paragraph length, listicle scaffolding without varied transitions, generic mechanism-light explainers that could have been written about any topic. Original first-person analysis with specific examples, named authors and primary citations is the version that holds up.
Build citations from sites Brave indexes deeply
Brave's smaller index means a handful of authoritative inbound links from sites it indexes well (Wikipedia, Reddit, established trade publications, GitHub) carries more relative weight than the same links would in Google. A targeted citation strategy aimed at sites Brave already trusts is more efficient than broad link building.
Track Claude referrals directly in GA4
Claude does pass referrer information when users click out from its responses. Set up GA4 segments for claude.ai and platform.claude.com as session sources, then look at which of your pages Claude is actually sending traffic to. We've found that the pages Claude cites are often not the ones a brand expects.
Write the kind of content Brave rewards
This is the slowest lever and the most durable. Mechanism-led posts that explain how something works, with verifiable claims and named sources, perform structurally better in Brave's index than best-practice content that could have been written by anyone. This post is the kind of thing we mean.
The broader point
Most AI search optimisation advice assumes the AI engines source their citations from broadly the same web, with broadly the same signals, in broadly the same proportions. They don't. ChatGPT's web search uses Bing. Claude uses Brave. Perplexity built its own index. Gemini sources through Google. Each engine has a different upstream gatekeeper with different ranking preferences. Treating "AI visibility" as one discipline misses the structural differences that determine which engine cites which sources.
For brands that care about Claude specifically, the practical answer is to run Brave optimisation as its own workstream. Different submission process, different content style, different competitive set. The work is materially less crowded than Google SEO because almost no one is doing it deliberately. The window for getting in early is open right now and will not stay open indefinitely.
What we are doing about it. We are running this audit on our own sites first. Submitting our priority URLs to Brave, comparing our Brave indexation against our Google indexation, tracking which of our pages Claude is actually citing. We are also building a Brave-specific check into our /check tool so brands can see their Brave index position alongside the chat-output results most AI visibility tools focus on.
Sources
- TechCrunch, "Anthropic appears to be using Brave to power web search for its Claude chatbot"
- Profound, Claude-Brave citation overlap analysis
- Anthropic, Web search tool API reference
- Brave Search, Crawler documentation
- Brave Search, URL submission
- Brave, Web Discovery Project
- Brave, Search API growth post
- Colorlib, Search engine statistics