Key takeaways
- Claude knows a brand either from its training data or by searching the web at that moment; because training data has a cutoff date, a new brand can only be found through web search.
- When we read it on 21 November 2026, Anthropic's subprocessor list named three companies for web search: Brave Search, TurboPuffer and Parallel Web Systems. Which query is answered from which is not stated.
- Independent evidence that Brave powers Claude's web search surfaced in March 2025; Anthropic still lists Brave by name today, but does not say it is the only source.
- Brave's Web Discovery Project only passes on an address once a large number of people have visited it independently, so a new domain with few visitors can be slow to appear in Brave's index.
- IndexNow does not reach Brave: it feeds Bing, and through Bing, Copilot; Google uses Search Console. To check Brave, a site: search on search.brave.com is enough.
Claude can know a brand in two ways: from its training data, or by searching the web at that moment. Training data has a cutoff date; for a brand founded after that date, or one that little has been written about, web search is the only way in. When Claude searches, it takes results from search providers, and Anthropic's own list names Brave Search among them. A new domain that is missing from Brave's index, that blocks Claude's bots in robots.txt, or that only a handful of sites link to may therefore not appear in Claude's answers. GeoMono is a platform that measures how brands appear in AI answers; in this post we explain, with sources, how Claude reads the web, and show how to check where your brand gets stuck along the way.
This post is not a measurement report. The information below rests on Anthropic's, Brave's and other companies' own documentation and on dated reporting; the sources were read on 21 November 2026. Where a fact is not confirmed by a source, we say so plainly.
When does Claude search the web?
Claude does not go to the web for every question. The web search tool page in Anthropic's developer documentation says search exists to reach current information beyond the model's knowledge cutoff, and that Claude decides when to search based on the prompt. The same page explains that decision with two lists:
| Claude usually searches | Claude usually answers directly |
|---|---|
| Recent events, news, announcements | Established facts, math, science and coding fundamentals |
| Current prices, rates, statistics | Creative writing and brainstorming |
| Organizations, people and products that might have changed | Analysis of content already given in the conversation |
| Explicit requests to search | Greetings and small talk |
For brands, the key row is the third: questions about specific organizations and products are the kind that trigger a search, because the information may have changed. But for a general question such as "Which accounting software should I use in Istanbul?", whether Claude searches depends on how the question is put; if it does not, the answer comes from training data, and a new brand is not there.
The Claude app works much the same way. According to the Claude Help Center (updated 9 November 2026), Claude invokes a search tool for topics that benefit from current information and includes source links in every answer. The page notes that the new Claude experience has no web search toggle and that Claude searches the web when it helps; in the older interface, search can be switched on and off per chat. The same page gives two more details: for requests that call for local results, Claude may use the user's location inferred from their IP address, and image search is powered by Bing.
Web search was first announced on 20 March 2025 as a preview for paid users in the United States; according to an update dated 27 May 2025, it is available on all plans and worldwide.
Which search infrastructure does Claude read from?
Anthropic's announcement and help page do not name a provider for web search. The providers' names appear in the subprocessor list, where the company lists the third parties that process customer data.
What was seen in March 2025? According to a TechCrunch report of 21 March 2025, Anthropic added Brave Search to its subprocessor list in the same days it announced web search. Software developer Simon Willison listed three pieces of evidence in a note published the same day: the list showing Brave Search as of 19 March 2025; the ten sources Claude showed for a query matching Brave's results for the same query exactly; and the parameter schema of Claude's web search tool being named "BraveSearchParams". At the time the report was published, neither Anthropic nor Brave had made a statement confirming it.
What does the list say today? When we read the subprocessor list in Anthropic's Trust Center on 21 November 2026, there were three entries whose purpose was given as "Web Search":
| Subprocessor | Products covered | Purpose |
|---|---|---|
| Brave Search | All products | Web search |
| TurboPuffer | All products except Claude for Government | Web search |
| Parallel Web Systems | All products except Claude for Government | Web search |
Anthropic's own bot should be added to this. According to Anthropic's crawler page, Claude-SearchBot navigates the web to improve search result quality for users, and when it is blocked, content is not indexed for search. In other words, Anthropic does some indexing of its own on the search side.
What is not confirmed? We found no Anthropic document explaining how these three providers and Anthropic's own index combine in an answer, which query is answered from which, or how much of the results come from Brave. The March 2025 evidence shows the situation at that time; no source today says Brave is the only source. What we can say is this: Brave Search is the only index that Anthropic lists by name for web search and that anyone can check from the outside.
Brave itself writes on its Brave Search API page that every result comes from its own independent index and that the API has no fallback to other search engines. This means a page missing from Brave's index will not arrive through Brave's API from another engine either.
robots.txt and Claude's three bots
The second condition for Claude to reach your page is that robots.txt allows it. Anthropic's crawler page (updated 4 November 2026) defines three bots:
| Bot | Role | If you block it |
|---|---|---|
| ClaudeBot | Collects content that could contribute to training | The site's future content is excluded from training data |
| Claude-SearchBot | Navigates the web to improve search result quality | Your content is not indexed for search; your visibility in search may drop |
| Claude-User | Opens a page at the moment a user asks a question | Claude cannot retrieve your content when answering the user |
Anthropic writes that its bots honor standard robots.txt directives, that each bot is managed separately, and that blocking IP addresses may not be a reliable route because it also stops the bot from reading robots.txt. If you want to turn off training and stay visible in search, blocking ClaudeBot while allowing Claude-SearchBot and Claude-User is enough. A sample file for all bots, and the details, are in our robots.txt guide.
On Brave's side, the rule works differently, and this detail is often missed. According to Brave's crawler page, Brave's bot does not use a distinctive user agent name, because it does not want to be excluded by sites that allow only Google. In return, if a domain or page cannot be crawled by Googlebot, Brave's bot does not crawl it either. The same page says robots.txt is not used to keep a page out of Brave's index; that takes a noindex directive on the page.
The practical consequence: allowing Claude's bots in robots.txt is not enough on its own. A rule that blocks Googlebot, or blocks all bots through the wildcard group, also stops Brave's bot and can close the path to Claude through Brave. These are the lines to check:
# Claude's search and user-request bots
User-agent: Claude-SearchBot
User-agent: Claude-User
Allow: /
# Blocking Googlebot also stops Brave's bot
User-agent: Googlebot
Allow: /
Next step
Is your site open to AI crawlers?
Enter your domain. The free scan measures whether AI crawlers can reach your site and whether your pages are quotable. No signup, no card, no email.
Why might a new domain be missing from Brave?
For a newly launched or rarely linked domain, the real bottleneck is not robots.txt but discovery. A search engine has to know a page exists before it can index it, and Brave does this differently from Google.
The Web Discovery Project. According to the announcement Brave published in October 2021 and updated in March 2024, the Web Discovery Project builds Brave's independent index from anonymous search and browsing data contributed by Brave browser users who opt in. The key rule in the announcement: for an address to be sent, it has to have been visited independently by a large number of people. Brave's crawler page also says its bot is partly powered by this system, and that the system is off by default. A new domain with few visitors may take time to pass this threshold.
Links. Brave does not publish in detail how its bot finds new pages outside the Web Discovery Project. The general way search engine crawlers work is to follow links on pages already in their index, so links from established sites would be expected to help Brave's discovery too. But we found no Brave document showing how much weight this carries at Brave; it is an expectation, not a confirmed rule.
There is no search console. Brave has no panel like Google Search Console or Bing Webmaster Tools where a site owner sees index status and submits a sitemap. Its crawler page only links to a form for submitting an address to be re-fetched, and mentions that form in the context of delisting pages that carry noindex or no longer exist. We found no Brave document saying the form adds a new domain to the index or speeds that up.
IndexNow does not reach Brave. IndexNow is an open protocol that notifies search engines of new and changed addresses instantly. The IndexNow FAQ lists Bing, Yandex, Naver, Seznam.cz, Yep and Amazon as participants; Brave and Google are not on the list. So think of IndexNow as a tool for Bing, not for Claude.
| Channel | What it reaches | Which AI products it opens a path to |
|---|---|---|
| IndexNow and Bing Webmaster Tools | Bing and the other IndexNow participants | Microsoft Copilot; ChatGPT search also points to Microsoft among its search providers |
| Google Search Console | Google Search, AI Overviews and AI Mode | |
| Brave's URL submission form | Brave | Re-fetching only; no documented effect on Claude |
The basis for the Bing row: a Bing Webmaster Blog post of 31 July 2025 describes using sitemaps and IndexNow together as the way to keep content fresh in both classic search and AI-powered experiences such as Copilot. OpenAI's ChatGPT search help page says ChatGPT works with search providers and links to Microsoft's privacy statement for how providers process queries. For Google, the route is Search Console. According to Google's AI features and your website document there are no additional requirements for AI Overviews and AI Mode, and standard SEO practice applies; according to its recrawl document, individual addresses are submitted with the URL Inspection tool and large numbers of addresses with a sitemap, and crawling can take from a few days to a few weeks.
How do you check whether your brand appears in Claude?
The steps below do not depend on any tool; you can do each of them yourself.
- Search Brave. On search.brave.com, search for your brand name and for
site:yourdomain.com. According to Brave's search operators page,site:returns results from that site only. If there are no results, you are most likely not in Brave's index. - Turn off Google mixing. According to Brave's help page, the Brave browser has an optional setting that mixes in Google's results for queries where Brave may not return enough results. If it is on, what you see may not come from Brave's own index. Check with this setting off, or from a different browser.
- Ask Claude with web search on. First ask a question your buyer would ask that does not contain your brand name; then ask one that does. Look at the sources under the answer: is your site cited, and if not, which sites are?
- Read your robots.txt. Check one by one which group applies to Claude-SearchBot, Claude-User and Googlebot. As explained above, a rule that blocks Googlebot also stops Brave.
- Look at your server logs. You can see in the logs whether Claude-SearchBot visits your site. Anthropic publishes its bots' IP addresses; not every request carrying a bot's name comes from Anthropic.
A warning: a single Claude answer is not a measurement. The same question can be answered with different sources at different times. To reach a conclusion, you need to ask the same questions repeatedly and count how many answers mention you; we describe the same diagnosis for ChatGPT in our post Why doesn't ChatGPT recommend my brand?
Our scan, which needs no sign-up, does the fourth step for you: it reads your robots.txt rules separately for each AI bot, Claude's included.
Frequently asked questions
Does Claude use Google's index?
We found no Anthropic document saying so. Anthropic's subprocessor list names Brave Search, TurboPuffer and Parallel Web Systems for web search; the help page says image search is powered by Bing. Google does not appear in these lists as a web search provider.
How do I add my site to Brave Search?
Brave has no search console for site owners. Brave's help page links to a form for submitting an address to be re-fetched, but does not say the form adds a new domain to the index. The discovery route Brave describes is the Web Discovery Project, fed by visits from participating users; links would also be expected to help, but Brave has not documented this.
Does IndexNow reach Claude?
Brave is not on IndexNow's participant list, so an IndexNow notification does not go to Brave's index. IndexNow feeds Bing and the other participating engines, and Bing describes sitemaps and IndexNow as the way to stay fresh in AI-powered experiences such as Copilot. For Google, new pages are submitted through Search Console.
What happens if I block Claude-SearchBot?
According to Anthropic, your content is not indexed for Claude's search, and your visibility in search results may drop. Blocking ClaudeBot, by contrast, affects only training; it does not block Claude-SearchBot or Claude-User. If you want to turn off training and stay in search, block ClaudeBot only.
Sources
What every figure in the post rests on. Web sources with their publisher and date, our own measurements with their date and scope.
- 1.Claude can now search the web · Anthropic, 20 March 2025, updated 27 May 2025
- 2.Enabling and using web search · Claude Help Center, updated 9 November 2026
- 3.Web search tool · Claude developer documentation, accessed 21 November 2026
- 4.Subprocessors · Anthropic Trust Center, accessed 21 November 2026
- 5.Anthropic appears to be using Brave to power web searches for its Claude chatbot · TechCrunch (Kyle Wiggers), 21 March 2025
- 6.Anthropic Trust Center: Brave Search added as a subprocessor · Simon Willison's Weblog, 21 March 2025
- 7.Does Anthropic crawl data from the web, and how can site owners block the crawler? · Claude Help Center, updated 4 November 2026
- 8.Brave Search Crawler · Brave Search Help, accessed 21 November 2026
- 9.Privacy-preserving Brave Search replaces Google as the default search engine in the Brave browser · Brave, 19 October 2021, updated 27 March 2024
- 10.Brave Search API · Brave, accessed 21 November 2026
- 11.Google fallback mixing · Brave Search Help, accessed 21 November 2026
- 12.Search Operators · Brave Search Help, accessed 21 November 2026
- 13.IndexNow FAQ · IndexNow, accessed 21 November 2026
- 14.Keeping Content Discoverable with Sitemaps in AI Powered Search · Bing Webmaster Blog, 31 July 2025
- 15.ChatGPT search · OpenAI Help Center, accessed 21 November 2026
- 16.AI features and your website · Google Search Central, updated 10 December 2025
- 17.Ask Google to recrawl your URLs · Google Search Central, updated 10 December 2025
Co-founder & CTO of GeoMono, and also of GOGOGO LLC and Gotonom. He has worked on chatbots and AI systems since 2015: AIML at Turkcell BiP first, then Messenger, Telegram and WhatsApp workflows. Online he has used the name Ezagor since 2007.
Measurements: GeoMono Research Team
Next step
You have read it, now see it on your own brand
The first of the steps in this post is measuring where you stand today. The tool below shows where that measurement starts.
Scan for free



