How to get cited by Claude
Claude product surfaces and their available tools can differ, so define the surface before diagnosing a missing mention or citation. Inspect Anthropic's documented search, training, and user-request crawler roles separately, compare page delivery and source coverage, and preserve repeated observations for the same prompt configuration. A later answer cannot reveal which change caused it.
How does Claude decide what to recommend?
A publisher cannot infer Claude's selection formula or training inputs from an answer. Anthropic documents separate crawler roles for search, potential training, and user-triggered retrieval, and the applicable product surface matters.
When a sampled Claude answer includes web sources, record the surface, tools, prompt, date, locale, response, and URLs. Use page access, clarity, freshness, and corroboration as hypotheses to inspect, not as guaranteed recommendation factors.
Make your site readable to Anthropic's crawler
Review the current Anthropic documentation for ClaudeBot, Claude-SearchBot, and Claude-User before changing policy. Inspect the exact public route in its initial and rendered forms, and reconcile test requests with verified edge or origin evidence where available. Policy and fetch results do not prove indexing or citation.
- Treat potential-training, search, and user-triggered roles as separate policy decisions.
- Include useful content in the initial response when practical, while recognizing provider rendering can vary.
- Keep intentionally public material outside login gates and test the exact response.
Write for clarity and extraction
Use a direct answer, descriptive headings, specific sourced claims, and structured data only where it matches visible content. This makes the page clearer for readers and parsers. It does not establish that Claude fetched, selected, or used the page.
Build broad, consistent corroboration
Review the first- and third-party sources returned for the prompt set, including reviews, roundups, documentation, and editorial coverage. Accurate, independent coverage can address a documented information gap, but it is not a universal high-leverage tactic or a guarantee of recall or citation.
Track post-ship observations
Run category and comparison prompts on a defined Claude surface, record the configuration, names, citations, sources, skips, and unavailable runs, then repeat after a documented shipment. Treat the result as an ongoing observation series; it neither proves attribution nor predicts that an answer will persist.
Frequently asked questions
It depends on the Claude product surface and enabled tools. Record that surface explicitly, and keep a web-source observation separate from assumptions about training or unsupported modes.
Anthropic documents ClaudeBot for potential model training and documents Claude-SearchBot and Claude-User for other roles. Review each control separately; an allow or block does not prove an answer pathway.
Do not represent paid exposure as an organic recommendation. For organic observations, inspect the returned answer and sources without claiming a placement formula.
There is no supported universal timeframe. Record the shipment date, Claude surface, and prompt configuration, then report when a later observation changes without promising it will.
A sampled answer does not reveal why the competitor was selected. Compare the returned sources, page coverage, and technical delivery, then test a bounded hypothesis while preserving provider variability.