How to get cited by Claude
Claude cites web pages when its search retrieves them for a user's question, and Anthropic splits its bots the same way OpenAI does: 1 crawler for training, 1 for search, 1 for user-initiated fetches. The names are ClaudeBot, Claude-SearchBot, and Claude-User, each with its own user-agent token and its own robots.txt consequences, documented in our Claude-SearchBot glossary entry. Getting cited is 2 jobs: let the right bots in, and give them something safe to quote. Both are checkable on your site today.
Job 1: let the right bots in
The robots.txt mistake that costs Claude citations is the same one that costs ChatGPT citations: a blanket rule written to block AI training that also blocks search and user fetches. The 3 Anthropic agents do different work. Blocking ClaudeBot is a training opt-out, a legitimate choice that does not protect visibility. Blocking the search agent removes you from consideration. Blocking the user agent stops Claude from opening your page when a human asks about you specifically, which is the strangest own goal in the list, since that human is often a customer mid-decision.
Read your robots.txt now and look for any rule matching Claude agents, including wildcards above them. Name each bot separately and give each its own rule, the same per-bot pattern shown in our robots.txt walkthrough. Then verify against your server logs that the agents you allowed are actually fetching.
Job 2: give it something safe to quote
- Exist in raw HTML. Retrieval reads the page as delivered. Content that appears only after JavaScript runs is content that does not exist for this purpose. Test and fix per the rendering guide.
- Open with the answer. A self-contained first passage that resolves the page's question gives the model a liftable quote with your name on it. Pattern in adding answer blocks.
- State checkable facts. Models attribute specifics: numbers, dates, named sources. Vague copy is unquotable. If your page contains nothing a fact-checker could verify, it contains nothing a model will cite.
- Be one unambiguous entity. Organization schema, consistent name and details, connected profiles. See entity clarity. Uncertain attribution loses to a competitor with clean identity data.
The honest part: what nobody controls
Which page Claude cites for a given question changes with the question's wording, the moment, and the model version. That is the nature of retrieval plus generation, and it is why we call our checks readiness rather than ranking. Industry reporting puts month-to-month churn in AI-cited sources at 40 to 60 percent (Search Engine Land). Readiness is the stable half of the equation and the only half for sale honestly. Anyone offering guaranteed Claude citations is offering something the system does not support.
Measure with your own data
Claude sends real referral traffic when it cites you and users click through. Check your analytics for referrals from claude.ai. Those visits, not a tracker's sampled score, are the ground truth of whether this work paid. The same measurement habit covers every engine, detailed in how to check your AI visibility.
Frequently asked questions
Is Claude optimization different from ChatGPT optimization?
The site-side checklist is nearly identical, which is good news: one round of fixes serves every engine. The engine-specific part is the bot names in robots.txt and the index behind each engine.
Does Claude use a search index like Bing?
Anthropic operates its own search crawler. The reliable strategy does not depend on index trivia: be reachable, readable, and quotable, and every retrieval system can use you.
How do I know if Claude can read my site right now?
Check robots.txt rules for the Claude agents and check server logs for their fetches. Our audit runs the access checks per bot as part of the 73 and reports exactly which engine's crawlers you block.