Claude cites a business when it runs a web search for a question and one of the pages it retrieves is yours. The work is to let Anthropic's crawlers reach your site, publish pages on your own domain that answer the exact questions your buyers ask, and then measure every week whether Claude names you, because nobody can guarantee that it will.
How does Claude find the pages it cites?
When web search is on, Claude does not answer only from what it learned in training. It searches, reads the results and writes an answer with sources attached. Anthropic's API documentation for the web search tool states that the response includes citations for sources drawn from the search results it retrieves, with a final response and cited sources at the end of its turn. The Claude support documentation describes the same thing from the user's side: when Claude invokes web search, it processes multiple sources and returns a response with direct citations, source links and relevant quotes where appropriate.
Two things follow from that. A test we ran on 23 September 2026 illustrates both, with one caveat: it measured a search-enabled model, not Claude specifically, so read it as an example of how assistants with search can behave rather than as a measurement of Claude.
First, a citation only happens if the assistant decides to search. In that test of 20 German business questions, the model answered 10 without searching at all and cited nobody.
Second, when it does search, it cites what the search turned up and what was worth quoting. Where the model searched, it cited primary sources such as laws, ministries, the OECD and large accounting firms on 9 of 10 questions. Firms were named only on questions like "who should I hire" or "how do I choose". Two runs of the same question named two different sets of firms.
In that test, then, the openings for a business were the questions where a buyer is choosing. The other route is to be the most specific, quotable source on a question, which is what the documentation above implies: Claude cites the sources it retrieves and draws on.
Which Anthropic crawlers must be allowed on your site?
According to ALM Corp's summary of Anthropic's documentation, Anthropic runs three separate crawlers, and all three honour standard robots.txt Disallow rules as well as the non-standard Crawl-delay directive. Each does a different job, so blocking one is not the same as blocking another.
| Crawler | What it does | What blocking it means for citations |
|---|---|---|
| ClaudeBot | Collects pages for model training | Your content is left out of what future models learn; it does not by itself stop live search |
| Claude-SearchBot | Indexes pages for Claude's search | Your pages have a weaker chance of appearing in the results Claude retrieves |
| Claude-User | Fetches a page in real time when a person's question needs it | Claude cannot open your page while answering, even if it has the URL |
If you want to be cited, all three should be allowed. A blanket rule such as User-agent: * with Disallow: /, a CDN bot-protection setting or a firewall rule can quietly shut out all of them.
ClaudeBot is not a rare visitor when nothing blocks it. Across the full retained nginx logs of pickaclass.com, an edtech platform we work with, from 28 July to 11 August 2026, ClaudeBot made 19,272,125 requests. GPTBot made 2,744,985 and Googlebot 52,292 in the same period. If your own logs show no Anthropic crawlers at all, treat that as a sign that something is in the way.
We made this mistake ourselves. A robots.txt rule on our own site was meant to hide a single redirect page, /b. Because robots.txt rules match prefixes, from 11 August to 21 October 2026 it also closed /blog, /business and every article in our business blog to all crawlers. Bing Webmaster Tools listed our business page as "discovered but not crawled" from 11 August 2026, and Google had none of the business pages indexed. We changed the rule to /b$ on 21 October 2026. The lesson: check what your robots.txt actually blocks, not what you meant it to block. Our guide on whether your website is blocking GPTBot, ClaudeBot and PerplexityBot walks through the checks.
Can Claude find your pages today?
Access is necessary, but it is not the same as being findable. On 19 October 2026 we asked assistants for the exact titles of 18 articles, each older than three weeks, published on three client sites. Claude found 7 of 18. ChatGPT found 1 of 18. Google had indexed 29 of 30 articles of one of those sites. Being in Google's index did not mean an assistant could find the page.
Claude did better than ChatGPT in that test, but 7 of 18 still means most pages were not found even when we asked for them by name. A buyer will never ask for your title; they ask a question in their own words.
The size of your site matters more than its markup. A site with complete markup (server rendering, Article, FAQPage and Organization JSON-LD, a sitemap) but only 75 URLs was still absent from answers on its own article's exact topic. A site with about 190,000 crawlable URLs was crawled constantly. Google's own guidance on AI features and your website points the same way: it lists crawlable robots.txt, internal linking, good page experience, textual content and accurate structured data as the basics, and says no special schema.org markup is required.
A local example shows how this plays out by question type. On 20 October 2026, for an orthopaedic practice in Bonn, the four assistants named the practice in 12 of 80 answers to 20 patient questions. Claude named it in 5 of 20, the most of the four (Gemini 4, ChatGPT 2, Perplexity 1). Across all four assistants, the directory jameda.de was cited in 12 answers, and competing practices' own websites were the most cited sources. Questions about who to choose named practices; questions about costs, risks and treatments named none.
What should you publish so Claude has something to cite?
Claude can only quote a page that exists and says something specific. In practice that means:
- One page per buyer question, on your own domain. Put the direct answer in the first two sentences, then the detail. A buyer asking "how do I choose a freight dimensioning tool" needs a page that answers exactly that.
- Answer the choosing questions honestly. Since firms were named mostly on "who should I hire" questions in our tests, write pages that set out selection criteria, who you are a good fit for, who you are not for, and what things cost.
- Make pages quotable. The paper that coined the term, GEO: Generative Engine Optimization, reports that across 10,000 queries on its GEO-BENCH dataset, adding citations, quotations from relevant sources and statistics to content could boost a source's visibility in AI-generated answers by over 40%. Cite the law, the standard or the official figure your claim rests on.
- Build surface, not just polish. Dozens of thorough answer pages, linked to each other, give a search far more chances to land on you than a perfect handful.
- Show up where assistants already look. In our 21 October 2026 measurement of GEO questions across ChatGPT, Claude, Gemini and Perplexity, the most cited sources were the blogs of GEO agencies and tool vendors. In the Bonn example, across the same four assistants, it was a medical directory and competitors' sites. Find the sites cited in your category and make sure you are accurately listed or mentioned there.
- Keep it in plain text. Content hidden in images, PDFs or scripts that need a browser to render is harder for any crawler to read. Files like llms.txt are not a substitute; see what llms.txt is and whether you need one.
Live fetches spread across a whole site, not just the homepage. In the fortnight of logs mentioned above, live retrieval agents from other assistants (ChatGPT-User, OAI-SearchBot, Perplexity-User) fetched content pages on the edtech platform 1,918 times, and a client blog of a few dozen articles received 56 such fetches across 24 different articles. Every answer page is a separate chance.
How do you know whether Claude is citing you?
You ask it, repeatedly, and count. Asked the same question three times, an assistant often names different businesses. We count a business as named for a question if any of three asks names it, and we only compare weeks on the same questions with the same number of asks.
Do not wait for referral traffic to tell you. In about two weeks of logs from three of our course websites, human visitors arriving from AI assistants totalled one from Copilot and zero from ChatGPT, Perplexity, Gemini, Claude, Meta AI and the rest. Assistants mostly answer in place rather than send a click, so citations have to be measured in the answers themselves.
Measuring Claude is the expensive part. With API prices of October 2026, one question asked once with web search costs us about $0.069 with Claude, against $0.010 with ChatGPT, $0.009 with Gemini and $0.0013 with Perplexity. A report of 20 questions across four assistants, three asks each, costs about $5.40.
Start from honest numbers. When we measured ourselves on 21 October 2026 with 20 English and 20 German questions about getting a business into AI answers, ItGrows was named in 0 of 160 answers and none of our pages was cited. We wrote up what that measurement showed. Progress, when it comes, tends to arrive one citation at a time: in our weekly measurement of 20 October 2026, Perplexity cited an article we had published on a client's site, a German company-formation advisory. Clients report results too: FreightSizer, a B2B SaaS for freight dimensioning, reported 2 new B2B customers in one month, both from ChatGPT; MagiScan, an iOS 3D-scanning app, reports hundreds of installs a month from ChatGPT, Claude and Gemini; appslift, an app store optimisation service, reports 2 to 5 leads a month from AI assistants.
If you want this done for you, ItGrows publishes answer articles on your domain, fixes crawler access and indexing, and every week puts your buyer questions to ChatGPT, Claude, Gemini and Perplexity, three times each, for $499 a month (plus a one-time $499 setup, free with 3 months or a year). To see where you stand first, the free check at itgrows.ai shows in about 30 seconds whether ChatGPT and Perplexity name your business for five of your buyers' questions. For Claude itself, put the same five questions to it with web search on, three times each, and note who gets named and which sites get cited. Our guide on how long it takes to start appearing in ChatGPT answers covers what to expect over the first months.
Frequently Asked Questions
Does blocking ClaudeBot stop Claude from citing my site?
Not on its own. Anthropic runs ClaudeBot for training, Claude-SearchBot for search indexing and Claude-User for live fetches, and each respects robots.txt separately. A wildcard Disallow or a CDN bot block, however, can shut out all three, which does remove you from Claude's answers.
Do I need special schema markup or llms.txt for Claude to cite me?
No special markup is needed. Google's guidance for its AI features says no special schema.org data is required, and in our measurements a site with complete JSON-LD but only 75 URLs was still absent from answers on its own topic. Pages that answer real questions in plain text matter far more.
Why does Claude name my competitors but not me?
Usually because their pages answer the question and yours do not exist or cannot be reached. In our Bonn practice example, competing practices' own websites were the most cited sources across the four assistants. Check crawler access first, then publish a page for each question where competitors appear.
How quickly can I expect Claude to cite my business?
There is no fixed timeline and no guarantee. Claude has to be able to crawl and find the page, and it has to decide to search for the question at all; in a 23 September 2026 test of a search-enabled model, half of 20 questions were answered without any search. Measure the same questions weekly so you can see the change when it happens.