To get cited by Claude, allow Anthropic's Claude-SearchBot and Claude-User agents in robots.txt, keep specific and dated product facts on pages they can read, and earn coverage in the articles, roundups and directories Claude already cites. Anthropic says Claude searches when a question depends on current or changing information, including facts about specific products, and in our scans Claude took a larger share of its citations from articles and reference sites than any other engine did.
How does Claude find sources?
Claude decides for itself whether to search. Anthropic's developer documentation lists the cases: questions about recent events, current prices or statistics, specific organizations, people or products whose details may have changed, and explicit requests to look something up. For stable knowledge, such as established facts or coding concepts, Claude answers without searching.
When it searches, it may search more than once in a single turn. Anthropic's documentation says simple factual questions typically take one to three searches, while comparative research across several entities can take 10 or more. A buyer asking Claude to compare tools in your category is exactly that kind of question.
In the Claude apps, Anthropic's help center says Claude grounds its answer in live web content when a topic benefits from current information, and that every response that uses search includes citations. In the newer Claude experience there is no web search toggle; Claude searches when it helps. On Team and Enterprise plans, an owner has to enable web search for the workspace first.
Two details from the developer documentation matter for publishers:
- Each result carries its page age. Along with the URL and title, the search tool returns when the page was last updated, so freshness is part of what every search result reports.
- Citations quote short passages. Citations are always on for web search, and each carries up to 150 characters of the cited text.
Access runs through three agents, and Anthropic says all of them honor robots.txt.
| Agent | What Anthropic uses it for | What disabling it does |
|---|---|---|
| ClaudeBot | Collecting web content that could contribute to model training | Signals that your future content should be excluded from training data |
| Claude-SearchBot | Indexing content to improve the relevance and accuracy of search results | Stops your content being indexed for search, which may reduce your visibility in Claude's search answers |
| Claude-User | Fetching pages when a user's question calls for it | Stops Claude retrieving your pages for users, which may reduce your visibility in user-directed search |
That last row is where Claude differs. OpenAI says robots.txt rules may not apply to its user-triggered agent, and Perplexity says its equivalent generally ignores them. Anthropic's Claude-User does follow robots.txt, so a catch-all disallow rule also blocks the fetches Claude makes for users. Anthropic also says that blocking its IP addresses may not work reliably, and that opt-out rules must be set for each subdomain. How to set up robots.txt for AI crawlers has the syntax.
What did Claude cite in our scans?
Our scans cover the questions Cituna tracks about AI visibility software: 33 scans in the 30 days to 22 September 2026 (how we count). Claude gave 552 answers containing 4,377 citations to third-party sites, about 7.9 per answer, level with Google AI Overviews and behind only Perplexity's 20.2. It cited 260 different third-party sites, not counting the vendors Cituna tracks as competitors.
All five of its most-cited sites sell AI visibility or AI search software, according to their own homepages: aiclicks.io (171 citations), llmpulse.ai (170), rankability.com (116), geoptie.com (104) and getmint.ai (89).
Set aside the vendors Cituna tracks as competitors and the engines' own pages, and look at the 100 third-party sites Claude cited most. Roundups took 70% of their 1,266 citations. Articles took 22%, review sites 3% and reference sites 2%, each the highest share for its type among the seven engines in our scans. Community sites took 2%, and Reddit none at all.
The details show what Claude reaches for:
- News for "what changed" questions. For "which AI visibility tool includes Google Search Console", Claude cited articles on cmswire.com and nohacks.co about AI visibility data in Google Search Console (32 citations each).
- Explainers for how-to questions. For "how do I measure my brand's AI share of voice", the sources included articles on searchengineland.com, shadow.inc, authoritytech.io and alexbirkett.com.
- Reference and directory pages. Wikipedia drew 15 citations, including its articles on Writesonic and on AI Overviews. The software directory alternativeto.net drew 35, and no other engine in our scans cited it.
- Big roundups. The most-cited third-party sites were integrate.io (68), writesonic.com (63) and blog.hubspot.com (43), mostly for "best tools" and alternatives lists.
Claude also misread at least one question. For "how to get cited by ChatGPT", it cited the APA Style blog's page on how to cite ChatGPT 11 times. What Claude cites has the complete list.
What does that mean for your site?
Keep product facts current and dated. Claude searches precisely when facts about specific products may have changed, and its search results report each page's age. A pricing or features page with a visible last-updated date and current figures is the kind of page that question needs.
Write quotable sentences. With citations carrying short passages, a claim that fits in one plain sentence ("Plan X includes Y and costs Z a month") is easier to cite whole than a paragraph of qualifiers.
Get written about when things change. Claude went to news articles for the Search Console question. When you ship something buyers ask about, trade press coverage can become the source Claude cites.
Keep reference and directory listings accurate. Wikipedia and alternativeto.net were cited for product questions. Check what they say about your company, and correct errors through each site's own process.
Do not count on Reddit for Claude. In our category Claude cited no Reddit pages at all, so community work pays off on other engines first. Reddit, YouTube and LinkedIn in AI answers compares them.
How do you get cited by Claude, step by step?
- Allow Claude-SearchBot and Claude-User in robots.txt on every subdomain, and check that no catch-all rule blocks them.
- Decide on ClaudeBot separately. Blocking training does not require blocking search.
- Remove any IP-based blocks aimed at Anthropic, which Anthropic says may not work reliably and can stop it reading your robots.txt.
- Give each product fact buyers ask about (pricing, plans, integrations, limits) a crawlable page with a visible last-updated date.
- Put a one-sentence answer to each buyer question near the top of the relevant page.
- When you launch or change something, brief the trade publications that cover your category.
- Review your listings on software directories and reference sites, and fix anything out of date.
- Get into the roundups Claude cites for your questions. How to get into "best tools" roundups covers the pitch.
- Test in Claude with search available, ask a comparison question, and record which of your pages it cites.
What to do first
- Check robots.txt for Claude-SearchBot and Claude-User access on every subdomain today.
- Add a visible last-updated date and current figures to your pricing page.
- Ask Claude your top comparison question and note every source it cites.
- Correct one out-of-date directory or reference listing.