Blog By Esteve Castells
Many AI assistants that cite the web don't run a full search index of their own. They rent search results from a company that does, and since Microsoft switched off the Bing Search APIs in August 2025, Brave has pitched its index as the only independent one still for rent at scale. Brave says that index holds about 40 billion pages. It sells access by the request to AI labs, enterprise software companies and people building their own agents, and the evidence points to Anthropic's Claude as one of the buyers.
That business is now in court. News Corp countersued Brave in July 2026, and on November 2, 2026, The New York Times and three other newspapers filed suits in Manhattan aimed at the index itself and the API that sells it. This guide covers how Brave builds its index, what the license lets buyers do, who is reported to use it, what the lawsuits claim, and what all of it means if you want your pages cited in AI answers.
What Brave Search is, and why its index matters #
Brave Search started in 2021. Brave bought Tailcat, a search engine built by the former Cliqz team, in March 2021 and opened Brave Search as a public beta that June. For its first two years it still filled some results from Bing. In April 2023 Brave removed all Bing API calls, which it said had supplied about 7% of query results, and image and video results moved to its own index later that year.
Brave has said it is the smallest of three US companies that operate independent search engines at scale, behind Google and Microsoft's Bing. Most other search engines and most AI products either license one of those three, scrape them, or run a smaller index of their own.
The numbers Brave gives about itself:
- Index size: about 40 billion pages, with more than 100 million pages added or refreshed each day (Brave, February 2026 ). Older Brave pages, and the November 2026 lawsuits, use "over 30 billion".
- AI answers: Ask Brave, its own AI search, handles about 22 million answers a day.
- Audience: Brave's API page cites signals from more than 130 million users of its browser and search engine, and the publishers' complaints put its 2025 revenue above $100 million.
These are self-reported figures, and Brave's own pages don't always agree with each other, so treat them as approximate.
Timeline: from search engine to AI data supplier
| Date | What happened |
|---|---|
| March 2021 | Brave buys Tailcat, the base of Brave Search |
| June 2021 | Brave Search public beta |
| April 2023 | Brave removes Bing API calls from its web results |
| May 2023 | Brave Search API opens to developers, with separate plans for search use and AI use |
| July 2024 | Meta's Llama 3 paper says the model was trained to use Brave Search as a tool |
| March 2025 | Claude gets web search; Anthropic adds Brave Search to its subprocessor list |
| March 2025 | Brave sues News Corp, asking a court to declare its indexing lawful |
| August 11, 2025 | Microsoft retires the Bing Search APIs |
| September 29, 2025 | Brave launches Ask Brave, its AI chat search |
| October 2025 | Brave Search API gets SOC 2 Type II attestation |
| December 2025 | Final judgment in the US search case against Google, including a one-time index disclosure to qualified rivals |
| February 2026 | New API pricing replaces the free tier with a $5 monthly credit; LLM Context endpoint launched; Snowflake integration announced |
| April 2026 | Brave reports nearly 700,000 OpenClaw users signed up for the API |
| July 21, 2026 | News Corp countersues Brave |
| November 2, 2026 | The New York Times and three Alden-owned papers sue Brave |
| November 12, 2026 | Brave accepts per-request payments from AI agents over the x402 protocol |
| January 1, 2027 | Google's Custom Search JSON API shuts down for existing customers |
How Brave builds its index #
Brave's index comes from two sources: its own crawler, and data from browser users who opt in to the Web Discovery Project.
A crawler with no name
Brave's crawler help page says the crawler "does not advertise a differentiated user agent". Brave's stated reason is fairness: many sites allow Google and block everyone else, and a named Brave crawler would be shut out of them. Instead, Brave follows the rules you set for Googlebot. If Googlebot can't crawl a page, Brave's crawler won't either.
For site owners this has three practical consequences:
- You can't block Brave without blocking Google. There is no Brave token to put in robots.txt.
- robots.txt doesn't remove a page from the index. Brave says so directly. To delist a page you add a
noindexdirective, which also removes it from Google, and then ask Brave to re-fetch it through its submit-URL page. - Your AI crawler rules don't apply. Blocking GPTBot, ClaudeBot or other AI agents by name in robots.txt has no effect on what Brave crawls, stores or sells.
The publishers suing Brave make the same point from the other side: their complaints argue that sites have no way to block the crawler that feeds an index sold to AI companies.
The Web Discovery Project
The second source is the Web Discovery Project, an opt-in setting in the Brave browser that sends anonymized data about the pages people visit. It is off by default, and the code is open source. Brave says these signals help it find pages and judge which ones matter, and its API page describes the index as built from the signals of "130+ million real humans" plus its own crawler. In practice, pages that people visit have a second route into the index besides links.
How Brave licenses the index: the Search API #
Brave sells its index through the Brave Search API, which opened to developers in May 2023. The plan you buy sets what you may do with the results: show them, feed them to a model, or keep them.
Plans and prices (November 2026)
| Plan | Price | What it covers |
|---|---|---|
| Search | $5 per 1,000 requests | Web results, LLM Context, news, images, video and places, up to 50 queries per second |
| Answers | $4 per 1,000 requests plus $5 per million input and output tokens | AI-written answers with citations and streaming, OpenAI SDK compatible |
| Spellcheck | $5 per 10,000 requests | Query spelling correction |
| Autocomplete | $5 per 10,000 requests | Query suggestions |
| Agent payments (x402) | $0.005 per request, paid in USDC | The same search endpoints with no account or API key |
| Enterprise | Custom | Zero data retention, custom terms, invoicing, storage rights |
Each plan comes with $5 of free credit a month, roughly 1,000 searches, on one condition: you credit the Brave Search API on your website or about page. Brave requires a credit card at sign-up. These prices come from Brave's February 2026 launch post and its API page, and Brave has changed its plans several times, so check the live page before you budget.
The x402 option, added on November 12, 2026, is the newest and the least conventional. An AI agent sends a request, gets an HTTP 402 "payment required" reply with the price, signs a USDC payment on the Base network and retries, with no account or API key involved. Brave pitches it at developers and autonomous agents.
Three levels of rights: show, ground, train
The pricing has simplified, but the licensing logic underneath has stayed the same since 2023. There are three things you might want to do with a search result, and Brave has always priced them differently.
| Right | What it means | How Brave handles it |
|---|---|---|
| Display | Show results to a person, as a search engine or a site search would | Included in the Search plan |
| AI inference (grounding) | Pass results to a model at answer time so it can write a cited answer | Included in the Search plan since February 2026; before that it needed an AI plan |
| Storage and training | Keep copies of results and train or fine-tune models on them | Custom plan with storage rights |
In the API's early plans, the free "Data for Search" tier came with terms stating that they "prohibit using responses for AI inference". AI use needed a "Data for AI" plan, and even then, according to plan terms quoted by Brave staff in a 2025 GitHub thread, inference was allowed "for the duration of your subscription only". Keeping results to train models required a third tier, "Data with storage rights", which let a customer "retain copies of any API responses" and "train and deploy AI learning models based on those stored API responses".
In February 2026 Brave folded inference into the basic Search plan, so any paying developer can now ground a model on Brave results. Storage and training stayed behind a custom contract. The current API terms, last updated on September 1, 2026, bar customers from storing results "other than transient storage" (clause 3(b)(i)) and from using them to "create, evaluate, train, re-train, fine-tune, benchmark" AI models (clause 3(b)(xiii)). That line between grounding and training sits at the center of the lawsuits, because the complaints quote Brave's own marketing, which said the API's most common uses were "in training and grounding several of the Web's biggest large language models".
LLM Context and Answers: results shaped for models
Two newer products package results for models. The LLM Context endpoint, launched in February 2026, runs a search, converts the HTML of the top pages into ranked "smart chunks" (including structured data, code, forum discussions and YouTube captions) and returns them within a token budget the developer sets, from about 1,000 to about 33,000 tokens. Brave says it answers in under 500 milliseconds at the 90th percentile.
The Answers plan goes one step further and returns a finished answer with citations, written by Brave's own system. Brave says LLM Context already powers Ask Brave, at about 22 million answers a day. Brave publishes benchmarks in which its answers beat ChatGPT, Perplexity and Google AI Mode, but it runs those tests itself and hasn't released the questions, so treat them as marketing.
LLM Context matters most if you track how your content appears in AI answers: a model grounded on it reads the chunks Brave chose to extract from your page, ranked by Brave's relevance model and cut to fit a token budget.
Who uses Brave's index #
Brave doesn't publish a customer list with contract details, and most buyers don't advertise where their search results come from. What is public comes from three places: Brave's own announcements and logos, technical evidence, and the court filings.
| Company | What is public | Source |
|---|---|---|
| Anthropic (Claude) | Brave Search added to Anthropic's subprocessor list in March 2025, a "BraveSearchParams" parameter in Claude's search tool, and matching citations in side-by-side tests. The November 2026 complaints name Anthropic as a customer and cite a 2023 Brave blog post saying Anthropic used the API to train Claude 2. Anthropic hasn't described the arrangement publicly. | TechCrunch , court filings |
| Perplexity | Named as a customer in the complaints and, according to Search Engine Journal, on Brave's About page. Perplexity also runs its own index and sells its own search API. | Court filings, Search Engine Journal |
| Mistral AI | Named in the complaints; logo on Brave's API page | Court filings, Brave |
| Cohere | Named in the complaints; logo on Brave's API page | Court filings, Brave |
| xAI | Named in the complaints | Court filings |
| Meta | The July 2024 Llama 3 paper says the model was trained to use Brave Search as a tool for recent events | Meta's Llama 3 paper |
| Snowflake | Brave powers web search in Snowflake Intelligence, Cortex Code and Cortex Agents (announced February 3, 2026) | Brave |
| AWS | Brave Search API listed in AWS Marketplace since July 2025; Brave published a guide to using it with Claude Cowork on Amazon Bedrock in June 2026 | Brave |
| You.com, Kagi, Together AI | Logos on Brave's API page | Brave |
| OpenClaw users | Nearly 700,000 users of the open-source agent platform had signed up for the API by April 2026; Brave's API page now claims more than a million | Brave |
None of this shows what these companies buy today. The five AI companies in the complaints are named in the allegations but are not defendants, the filings don't say which of them still use the API, and none has confirmed a contract in public. The link to Claude rests on evidence from 2023 and March 2025. A logo on a vendor page doesn't tell you what a company buys or whether it still does. Brave's broader claim, that the API supplies "most of the top-10 LLMs" and is the only index behind some of them, is its own.
Brave also has its own consumer AI products built on the index: AI answers at the top of Brave Search results, and Ask Brave, its AI chat search. LLM Pulse tracks Ask Brave as Brave AI.
Why AI companies buy from Brave #
Brave's pitch works because the alternatives shrank. In 2024 a developer who needed general web results through an API could buy them from Microsoft or Google. By the end of 2026 neither will sell them in that form.
- Microsoftretired the Bing Search APIs on August 11, 2025 and pointed customers to Grounding with Bing Search inside Azure AI Agents. That product feeds Bing results to an agent, but developers don't get the raw results back, so it can't stand in for a search API. ChatGPT's search still draws on third-party providers that OpenAI says include Bing, as we explain inwhere ChatGPT gets its data .
- Googleclosed its Custom Search JSON API to new customers , and existing customers lose it on January 1, 2027. The suggested replacement, Vertex AI Search, covers up to 50 domains. Full web search is by request only.
- Scraping got harder. Google sued a search-results scraping company in December 2025, and more recently began routing result clicks through google.com/goto. Brave says scrapers then turned to Brave Search, with spikes of several thousand requests per second, and that it is starting tocloak its own result links to make bulk scraping more expensive.
Brave sells raw results from a large independent index, and few others do. Perplexity launched a search API on its own index in September 2025 at a similar $5 per 1,000 requests. Exa, Tavily, You.com and a few others sell search built for AI, with their own crawls of varying size or a mix of sources. Common Crawl offers a free archive that is useful for training and useless for anything that needs today's web.
Google's court-ordered index sharing
One change could reopen the market. In the US government's search case against Google, Judge Amit Mehta's final judgment of December 2025 requires Google to make a one-time disclosure of its search index to "qualified competitors", share some user interaction data, and syndicate search results and text ads to rivals. Most of the judgment took effect on February 3, 2026, for six years, and the court indicated that generative AI companies planning to compete in search could qualify.
Google appealed in January 2026. The district court refused to the data-sharing remedy, calling the request premature because the license terms, privacy safeguards and the list of qualified competitors were "far from established" (PPC Land). If the remedy survives and the terms get settled, AI companies could get a second large index to build on, and Brave would lose its position as the obvious choice.
The lawsuits against Brave #
Three publisher groups are now in court with Brave, and all of them target the index itself as well as Brave's AI answers.
Brave v. News Corp, and the countersuit
In March 2025, after News Corp sent a cease-and-desist letter, Brave sued first, asking a court to declare that its use of News Corp's articles is not copyright infringement. Its argument is fair use: indexing makes the content searchable, and Brave shows users only snippets and "high-level summaries". The defendants include Dow Jones, the New York Post and News Corp's UK and Australian businesses. Brave filed a revised complaint in May 2026 after what News Corp called failed negotiations over a "fair, market-based agreement".
On July 21, 2026, News Corp countersued in federal court in Oakland, California, accusing Brave of "flagrant theft" by copying Wall Street Journal and New York Post articles and selling them to AI companies. It wants an injunction and damages of up to $150,000 per infringement (Reuters, via The Spokesman-Review). Brave, for its part, has accused News Corp of threatening advances in generative AI, which it said many consider "the most important innovation so far this century".
The New York Times and the Alden papers
On November 2, 2026, two suits landed in the US District Court for the Southern District of New York: The New York Times Co. v. Brave Software (1:26-cv-09549) and Daily News LP v. Brave Software (1:26-cv-09554), the second brought by the New York Daily News, the Chicago Tribune and the Orlando Sentinel (heise).
The complaints make four main allegations (The Next Web):
- The index is unlicensed copying. Brave built an index of more than 30 billion pages with a crawler that never identifies itself, so publishers had no way to block it.
- Brave resells that copy to AI companies. The complaints name Anthropic, Perplexity, Mistral, xAI and Cohere as API customers. None of them is a defendant.
- Ask Brave reproduces articles. The papers say Brave's AI answers return passages word for word and give detailed summaries of their articles.
- The browser defeats paywalls. The complaints point to a paywall-blocking feature in the Brave browser, which is the basis for the DMCA anti-circumvention claims.
The Alden papers bring claims for direct, contributory and vicarious copyright infringement, DMCA circumvention and unjust enrichment under New York law, and say more than two million registered articles are involved. The papers ask for damages, Brave's profits, a permanent injunction and a jury trial. Brave hadn't responded publicly when the suits were reported.
What could change
None of these cases has reached a ruling, and copyright suits of this size take years. Three outcomes would matter well beyond Brave:
- If indexing for resale isn't fair use, every company selling search results to AI would need licenses or exclusions, and AI products would pay more for grounding data or use less of it.
- If courts focus on the unnamed crawler, Brave could be pushed to identify itself, giving sites a way to block it without blocking Google.
- If injunctions remove publishers from the index, AI answers built on Brave would cite those publishers less, and other sources more.
For the wider picture of which publishers chose deals over lawsuits, see our map of every AI content licensing deal.
What Brave's index means for your AI visibility #
If you work on getting your brand mentioned and cited in AI answers, the index layer deserves attention: an assistant that grounds on Brave can only cite pages Brave has, and only reads the parts of them Brave extracts.
1. Check that Brave has your important pages
Search site:yourdomain.com on search.brave.com, then search for the pages that matter one by one: product pages, pricing, comparison pages, the guides you want cited. A homepage in the index doesn't mean the rest is there. If a page is missing, check for noindex tags, canonical tags pointing elsewhere, and robots.txt rules that block Googlebot, since Brave follows them.
2. Know which crawler rules reach which AI products
Your AI crawler settings and your Brave visibility are separate decisions. Blocking Anthropic's crawlers by name in robots.txt doesn't take your pages out of Brave's index, and Brave's index is what Claude's search was reported to use in 2025. The reverse also holds: the only way to keep a page out of Brave is a noindex that also takes it out of Google, so a publisher can't keep Google search traffic while opting out of Brave.
3. Make your pages easy to chunk
LLM Context splits pages into chunks and ranks them, so a page that states its main facts plainly, in HTML text, near the relevant heading, gives Brave something clean to extract. Content that only appears after JavaScript runs, inside images or behind interactions is harder to pick up. Structured data, forum threads and video captions are among the formats Brave says it extracts.
4. Test your queries in Brave and Ask Brave
Brave ranks differently from Google. Run your priority questions on Brave Search and Ask Brave and note which domains come up, because those results decide what a Brave-grounded model gets to read. A page ranking third on Google can be on the second page of Brave, or missing.
5. Measure the answers, not just the index
The same index feeds models that behave very differently. When we asked the Claude.ai app 400 buying questions 18 times each, between October 31 and November 1, 2026, Claude Sonnet 5.5 searched the web in every answer about smartphones, earbuds and robot vacuums, and in 1.4% of answers about skincare and language apps. None of the 15,937 sources it and Claude Haiku 5.5 cited came from Reddit, Wikipedia or YouTube (our Claude consistency study).
6. If you're a publisher, pick a position on licensing
Publishers now face a three-way choice with Brave: accept being indexed and cited, license on negotiated terms, or remove pages with noindex at the cost of Google traffic. The lawsuits may create a fourth option, a Brave-only opt-out, but nobody can plan around that yet.
How LLM Pulse tracks the answers built on Brave's index #
Neither Brave nor the AI companies show you which indexed pages fed a given answer. What you can measure is the result: whether AI answers mention your brand, which pages they cite and how that changes over time.
LLM Pulse runs your prompts across ChatGPT, Perplexity, Gemini, Google AI Mode and Google AI Overviews, plus models such as Claude as paid add-ons. Ask Brave is tracked as Brave AI through Talk to sales: our team sets it up for approved paid accounts, and it then runs on your tracking cadence, weekly by default or daily, like the rest of your prompts.
The citations report shows which domains each model cites for your prompts, side by side. If Claude cites a review site that Perplexity ignores, or your own pages appear in one model and not another, you see it here, and you can compare it with what you find in Brave's results.
Each answer opens in a response details view with the prompt, the model, the brands mentioned, sentiment and the full list of cited pages, which is where you check whether a specific URL made it into the answer.
See which pages Claude and Ask Brave cite
Track brand mentions, citations, sentiment and share of voice across ChatGPT, Claude, Perplexity, Gemini and Google's AI results.
What to watch in 2027 #
- January 1, 2027: Google's Custom Search JSON API stops working for existing customers, who will need another source of web results.
- Motions in the New York cases: an early ruling on Brave's fair use defense would signal how courts see indexing for AI resale.
- The Google remedy: whether qualified competitors, including AI companies, actually receive Google index data while the appeal runs.
- A named Brave crawler: the clearest sign that the lawsuits are changing how Brave operates.
- Agent payments: whether AI agents paying per request over x402 become a meaningful share of Brave's API traffic.
Frequently asked questions #
Brave runs its own web search index and sells access to it through the Brave Search API. The plan you buy decides what you can do with the results: show them to users, pass them to an AI model at answer time (included in the Search plan since February 2026), or store them and train models on them, which needs a custom plan with storage rights.
As of November 2026, the Search plan costs $5 per 1,000 requests and the Answers plan $4 per 1,000 requests plus $5 per million input and output tokens. Spellcheck and Autocomplete cost $5 per 10,000 requests. Each plan includes $5 of free monthly credit if you credit Brave on your site, and AI agents can pay $0.005 per request in USDC over the x402 protocol. Enterprise pricing is custom.
It did as of 2025, and nothing public says it has switched, though Anthropic hasn't described the arrangement. Anthropic added Brave Search to its subprocessor list when Claude got web search in March 2025, Claude's search tool used a "BraveSearchParams" parameter, and side-by-side tests found matching citations. The November 2026 lawsuits against Brave also name Anthropic as an API customer.
The New York Times and Alden newspaper complaints name Anthropic, Perplexity, Mistral, xAI and Cohere as customers, though none is a defendant and none has confirmed a contract publicly. Brave has announced integrations with Snowflake and AWS Marketplace, shows logos including Cohere, Mistral AI, You.com, Kagi and Together AI on its API page, and says nearly 700,000 OpenClaw agent users had signed up by April 2026.
No. Brave's crawler doesn't use a user agent of its own and follows the robots.txt rules you set for Googlebot. robots.txt also doesn't remove a page from Brave's index: you need a noindex directive, which removes the page from Google as well.
Only under a custom agreement that includes storage rights. Brave's current API terms bar storing results beyond transient storage and using them to create, train, fine-tune or benchmark AI models. The standard Search plan lets you pass results to a model at answer time but not keep them.
In suits filed in Manhattan on November 2, 2026, The New York Times and the Daily News, Chicago Tribune and Orlando Sentinel allege that Brave copied their articles into an index of more than 30 billion pages with an unnamed crawler, sold that content to AI companies through its API, reproduced passages in Ask Brave and let browser users get past paywalls. They seek damages, profits and a permanent injunction. Separately, News Corp countersued Brave in July 2026.
Search site:yourdomain.com on search.brave.com, then check your most important pages one by one. If a page is missing, look for noindex tags, canonical tags pointing elsewhere and robots.txt rules that block Googlebot, since Brave follows them. You can ask Brave to re-fetch a page through its submit-URL page.
Discover your brand's visibility in AI search effortlessly