Updated on 13 November 2026, by João Leite, Founder of murmur.
To appear in Claude, your company needs pages that Anthropic’s search crawler is allowed to access and that answer the buyer’s question. Anthropic, the company that makes Claude, writes on the help page about its crawlers that blocking Claude-SearchBot prevents the system from indexing the site for search, “which may reduce your site’s visibility and accuracy in user search results”. When Claude searches the internet, the answer shows the sources it used, and your company’s page can be among them. The path there has six steps, from robots.txt to the sources Claude cites.
- Claude-SearchBot is the crawler that brings pages into Claude’s search. ClaudeBot collects text to train the models, and Claude-User opens the page when someone asks a question.
- With search, every Claude answer brings the sources. That is where you check whether your company’s page has already made it in.
- Position in Brave Search counts, according to Profound. It measured the connection between the top results of that search engine and the pages Claude cites.
- murmur, a GEO agency in Brazil that puts brands into the answers of ChatGPT, Claude, Google AIO and other AI agents, monitors Claude every week. In the questions about GEO and GEO agencies that it measures, Claude cited company pages and two agency profiles in directories, one on Sortlist and one on Crunchbase.
- The timeline depends on where the company starts. On average across murmur’s clients, significant improvement came from 60 days onward, depending on the starting point and the competition.
To appear in Claude, the page must be found by its search and answer the question
To appear in Claude, the page must be found when it searches the internet and must directly answer what the person asked. Claude answers with what it learned in training or with what it searches at the moment, and your company’s new page gets in through search. In Claude’s new experience there is no search button: Anthropic writes that it searches when that helps, as in subjects that call for current information. The guide about training or internet search covers when it searches and when it answers from training, in the new experience and in the API.
Anthropic has three AI crawlers (crawlers), the programs that visit sites and read pages for it. Claude-SearchBot reads pages for Claude’s search, ClaudeBot collects text for training and Claude-User opens a page when someone asks a question. To get into the answer with search, the ones that matter are Claude-SearchBot and Claude-User. A table with the effect of blocking each crawler, side by side with OpenAI’s, appears in the guide that explains llms.txt.
| Control | What it does in Claude | Where to change it |
|---|---|---|
| Claude-SearchBot | Reads pages for Claude’s search | In the site’s robots.txt |
| Claude-User | Opens the page when someone asks a question | In the site’s robots.txt |
| ClaudeBot | Collects text for training the models, not for search | In the site’s robots.txt |
| Search in the conversation | In the old experience, the person turns on the Web search option in the + menu; in the new one, Claude searches when that helps | In Claude, at the user’s end |
| Citations | Every answer with search shows the sources used | In the answer itself, to check |
In an August 2026 measurement, Claude showed fewer sources per answer than ChatGPT, Gemini and Perplexity. In Bill Widmer’s study for Orbit Media, a company that sells website creation with a focus on SEO, Claude cited an average of 3.6 sources per answer, against 4.5 for ChatGPT, 8.1 for Gemini and 19.2 for Perplexity. The sample is 13,184 citations in 1,765 answers to 72 purchase questions about three brands that sell to businesses, collected up to 23 August 2026. The author declares Semrush, one of the three brands measured, as a client. How Perplexity chooses its sources is in the guide about how to appear in Perplexity.
What Claude cited in the questions about GEO that murmur measures
In the questions about GEO and GEO agencies that murmur asks Claude, the sources brought pages from the companies themselves and two agency profiles in company directories, one on Sortlist and one on Crunchbase. GEO is the work of making the company appear in the answers of ChatGPT and the other AI models. murmur asks Claude 90 questions about this subject every week, and the study brings together four rounds, from 12 October to 2 November 2026. There were 358 answers, 151 of them with the sources recorded.
One agency’s profile on Sortlist appeared as a source in 9 of the 90 questions, and another agency’s profile on Crunchbase in 7 of the 90. In both directories, the source was a single page, that agency’s profile, and not the directory as a whole.
The result holds for these 90 questions and these four rounds. In another sector, Claude may cite other sources, and the right calculation uses your buyer’s questions. What other measurements found about company sites in Claude’s citations, and the search engine behind the pages, is in the guide about how the company becomes a source for AI models.
How to appear in Claude in six steps
To appear in Claude, follow six steps, from the crawler that reads the site to reading the sources in the answer.
- Allow Claude-SearchBot and Claude-User in robots.txt. In robots.txt, the rules file at the root of the site, each robot gets the list of what it can and cannot read. Anthropic writes that blocking Claude-User prevents Claude from searching the page when someone asks a question and can reduce the site’s visibility in those searches. The robots.txt example is in the llms.txt guide.
- Have the text ready in the HTML. The text that matters must be in the code the server delivers, not only in what the browser assembles afterward with JavaScript. One way to check: in Chrome, Ctrl+U shows the source code, and the main sentence must be there. The rest of the technical part is in the checklist of what the site needs to have for AI models.
- Publish one page for each buyer question. Each page answers a genuinely different question, with the answer in the first sentence and the words the buyer uses. Numbers, prices, timelines and the city are written in the text, not only in an image. How to build each page is in the guide about how to write a text to be cited by AI.
- Keep name, service and city the same in directory profiles. The company name always goes with what it does and where it serves, the same way on the site, on LinkedIn and in the sector’s company directories. A similar name confuses the AI: in June 2026, with the name Visio in the question, the AI talked about the Microsoft program 41% of the time, and not about Visio, a murmur client that sells operations software for retail. The guide about Reclame AQUI and customer ratings shows how much review sites count in Claude.
- Check the page’s position in Brave Search. Ask the same buyer question in Brave Search and see whether your page appears among the top results. Profound, a company that sells AI monitoring, published a measurement of the link between the top results of Brave Search and the pages Claude cites.
- Ask Claude and open the sources in the answer. Ask Claude your buyer’s questions, each one in a new conversation, and open the sources it shows. Note whether your company appears, whether its site is among the sources and which companies appear in its place.
How to know whether Claude already cites your company
To find out whether Claude already cites your company, ask your buyer’s questions with search turned on and open the sources for each answer. Anthropic writes, on the help page about web search in Claude, that “Every response includes citations, so you can easily verify sources yourself”. In the old experience, the person turns on search in each conversation, through the Web search option in the + menu. In the new one, there is no button, and Claude searches when that helps.
Asking once is not enough: Claude can answer the same question differently on each attempt. Open a new conversation for each question, so that previous ones do not influence it, and return to the same questions on different days. The size of the list also matters. In a study updated in September 2026, based on 320 questions, Otterly, which sells AI monitoring, saw the variation in presence drop by close to 73% when going from a list of 10 to 100 questions. Otterly notes that 100 does not work as the right number for every brand.
The number that sums up this reading is presence: of the answers recorded, the share in which your brand appears. The market uses the English name, share of voice, and the calculation is explained in the guide about share of voice in AI models. To track Claude alone, the full method is in the guide that measures Claude’s citations.
At murmur, Claude receives the same questions every week, alongside ChatGPT, Perplexity and Google AIO, the AI-written summary that Google shows at the top of many searches. With a fixed list, one week can be compared with the next, and the dashboard included in the service shows the brand’s curve next to the competitors’ curve.
murmur tracks Claude every week, alongside the other three AI models
murmur tracks Claude every week, on the same questions, alongside ChatGPT, Perplexity and Google AIO. The content it produces works for any AI model that reads the open web, and these four are where murmur tracks the result. One person studies the company and its sector, analyzes what Claude and the other AI models answer to the buyer and who appears in the company’s place. The same person takes care of the content that changes that answer and closely follows the result, from start to finish.
At Visio, a murmur client, the progress also showed up in Claude. Visio sells operations software for retail chains, franchises and food service, with margin control, COGS, per-store reconciliation and loss prevention. The starting point was the name shared with Microsoft’s diagramming program, and the work was to get the AI to link the name Visio to retail operations. On the same questions, Visio appeared in 7% of Claude’s answers in July 2026, against 1% in June; across the four AI models measured, it went from 2% to 36%.
The pages murmur writes for the client sit on a subdomain with the client’s brand, set up and maintained by murmur, and the main site stays as it is. When the contract, monthly and with no minimum term, ends, the content remains the client’s.
Frequently asked questions
Does blocking ClaudeBot remove my site from Claude’s answers?
ClaudeBot is the bot that collects text to train Anthropic’s models, and Claude’s search depends on another bot, Claude-SearchBot. For answers with search, Anthropic writes that blocking Claude-SearchBot or Claude-User can reduce the site’s visibility. Those who want to stay out of training write a rule only for ClaudeBot and leave the other two allowed. The table of bots, with the effect of blocking each one, is in the llms.txt guide.
Does Claude search the internet before every answer?
No: in Claude’s new experience, it searches when that helps answer, and there is no button to turn search on. Anthropic writes that, when the subject calls for current information, Claude triggers a search tool and bases the answer on what it found on the web. Outside that, it answers with what it learned in training. When Claude searches, in the new experience and in the API, is in the guide about training or internet search.
Does a directory profile help a company appear in Claude?
A company directory profile can become a source for Claude. In the questions about GEO and GEO agencies that murmur measures, Claude cited one agency’s profile on Sortlist and another’s on Crunchbase, across four rounds in October and November 2026. Each was a single page, and the result holds for those questions, not for every sector. That is why the profile carries the same name, the same service and the same city as the site.
How many sources does Claude show in each answer?
Few, in Bill Widmer’s measurement for Orbit Media, a company that sells website creation with a focus on SEO: Claude cited an average of 3.6 sources per answer, against 19.2 for Perplexity. The sample is 1,765 answers to 72 purchase questions about three brands that sell to businesses, up to 23 August 2026, and the author declares Semrush, one of the brands measured, as a client. When Claude searches the web, Anthropic writes that every answer includes the citations, and that is where you see whether the company’s page made it in.
Where to start
The first step is to see what Claude and the other AI models already say to people searching for what your company sells. In murmur’s free diagnostic, you type the site address and, without signing up, see those answers on screen in less than two minutes; the full report arrives later in your email. In murmur’s sector studies, each market measured appears with the brands the AI models named, in a round of 25 questions per sector, with the date of each one.
Read also
- How to get your company cited by AI models: ChatGPT, Claude and Google AIO
- When Claude searches the internet and when it answers from training
- llms.txt: What It Is, What It Is For, and Whether It Works to Get Cited by AI
- The same path in Perplexity, from the bot to the sources cited
- Reviews and Reclame Aqui influence what AI says about your company
References
- Anthropic, “Does Anthropic crawl data from the web, and how can site owners block the crawler?”, Claude Help Center, undated page. Blocking Claude-SearchBot or Claude-User can reduce the site’s visibility in answers with search. https://support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler (accessed on 7 November 2026).
- Anthropic, “Enable and use web search”, Claude Help Center, dated 7 October 2026. In the new experience there is no search button, and every answer with search includes the citations. https://support.claude.com/en/articles/10684626-enabling-and-using-web-search (accessed on 7 November 2026).
- Bill Widmer, Orbit Media, “I Tracked 13,184 AI Citations Across 4 LLMs: Here’s What I Learned”, 2 September 2026. Average sources per answer across four AI models, in 13,184 citations from 1,765 answers to 72 purchase questions about three brands that sell to businesses, up to 23 August 2026; the author declares Semrush, one of the brands measured, as a client. Orbit Media sells website creation with a focus on SEO. https://www.orbitmedia.com/blog/ai-citation-sources/ (accessed on 7 October 2026).
- Josh Blyskal, of Profound, in “How Brave Search discovers new pages: A deep dive into the Web Discovery Project”, 9 July 2026. The link between Brave Search’s top results and the pages cited by Claude. Profound sells AI monitoring. https://www.merj.com/blog/how-brave-search-discovers-new-pages (accessed on 3 October 2026).
- Otterly, “We Measured How Stable AI Citations and Brand Mentions Really Are (and Why 10 Prompts Isn’t Enough)”, updated on 11 September 2026. From 10 to 100 questions, the variation in measured presence dropped by about 73%, in a sample of 320 questions. Otterly sells AI monitoring. https://otterly.ai/blog/ai-search-visibility-stability/ (accessed on 20 October 2026).
- murmur, Visio case study, published with the client’s authorization. https://murmur.marketing/en/cases/visio/ (accessed on 6 November 2026).