Brave sells its index to Cohere, Mistral, Perplexity

Brave Search has roughly 30 billion pages in its index and, per Search Engine Journal, has been quietly licensing that index to Cohere, Mistral, and Perplexity. There was no press cycle. No brand-safety deck. No opt-in email asking webmasters whether they'd like to feed a frontier model's retrieval layer. Your content is either in there or it isn't, and Search Engine Journal's suggested fix is a 30-second site: search, which tells you everything about how much warning anyone got.

This is the part of the AI supply chain nobody puts in a slide. Model-makers need current web data and crawling it all themselves is expensive, legally fraught, and increasingly blocked at the robots.txt level by publishers who've wised up. Buying an index that already exists solves that problem at a stroke. Brave built its index to compete with Google; it turns out the more durable business was never the consumer search product, it was the underlying crawl, sold wholesale to whoever needs to ground an LLM in something resembling the current internet.

Three customers, one index, no visibility into which answer came from where

Cohere sells enterprise AI to banks and governments. Mistral is the model many European public-sector buyers default to for sovereignty reasons. Perplexity is the answer engine that has made "citation" into a marketing claim. All three now draw, at least in part, from the same underlying Brave crawl. That matters because brand visibility teams tend to think about model presence one platform at a time: are we cited in ChatGPT, do we show up in Perplexity, does Gemini know who we are. That framing assumes each platform has its own independent view of your site. Increasingly, several of them are looking through the same lens, acquired wholesale rather than crawled bespoke. A content gap in Brave's index isn't a Brave problem. It's a Cohere problem, a Mistral problem, and a Perplexity problem, simultaneously, and you won't see which one surfaces it first.

For a financial institution or a multilateral agency, that collapses a diversification assumption that doesn't hold up. A bank optimising its newsroom for Perplexity citations because that's where its comms team tracks share of voice may think it has covered its exposure. If Perplexity's answer for "who sets Basel III capital buffers" draws on a Brave-sourced snippet that predates a policy update, the same stale fact can surface across Cohere-powered enterprise tools and Mistral-based sovereign deployments without anyone on the policy team knowing a single crawl was the common point of failure. Multilaterals publishing guidance documents, in particular, should treat Brave's index the way they'd treat Wikipedia a decade ago: unglamorous, rarely the subject of board discussion, and disproportionately influential in how summarisation tools downstream actually describe you.

The index itself is also idiosyncratic in ways that matter. Brave Search was built from a privacy-first crawl with its own crawler behaviour, its own rate limits, its own blind spots around pages that gate content behind consent walls or aggressive bot detection. A site that renders cleanly for Googlebot can render as a near-empty shell for Brave's crawler if the JavaScript or cookie-consent implementation differs even slightly. That means the set of brands well-represented in Brave's index is not the same set well-represented in Google's, and the gap between the two is now a gap in what three separate AI products can say about you. Industrial groups with product documentation locked behind portal logins, a common pattern in B2B manufacturing, are a plausible blind spot here: the spec sheet exists, it's just never been crawlable by anything outside Google's own infrastructure deals.

None of this requires conspiracy thinking. It requires updating a mental model that was built for an era when "search visibility" meant one index, one ranking algorithm, one set of best practices. The AI answer layer is now assembled from licensed, resold, and recombined crawls, and the commercial relationships behind them are disclosed, if at all, in a vendor's data-sourcing page rather than a headline. Brave's arrangement is confirmed by the companies involved; the next one may not be announced at all.

The practical move is unglamorous and overdue: run the site: check, audit for consent-wall and JavaScript-rendering blind spots that would exclude pages from a privacy-oriented crawler, and stop assuming that good standing in Google's index implies good standing everywhere an AI model goes looking for facts.