Updated
Anthropic Claude
Claude refuses to make claims it cannot back up.
Anthropic built Claude to be cautious about citation. Where ChatGPT might confidently paraphrase a community thread, Claude is more likely to hedge or refuse if the source is weak. That means structured, fact-led content with verifiable sources wins. The brands cited most often are the ones that read like a real reference, not a marketing landing page.
The 30-second answer: Claude rewards content that looks like a reliable reference. Named author, primary sources cited inline, structured facts, schema. Conservative on contested claims. The brands that get cited are the ones that earned the structural credibility, not the ones with the most content.
What Claude weights when it picks a source.
Four signals that drive whether Claude trusts your page enough to quote it.
Source verifiability
Claude checks whether a claim can be backed up against multiple sources. Sites with strong internal consistency and external corroboration get cited more. Outlier claims get hedged or skipped.
Primary over secondary sources
If your page IS the primary source - the company that built the tool, the firm that ran the study, the consultant who did the work - Claude prefers it over a third-party summary. Establish primacy clearly through Person and Organization schema.
Structured factual content
Claude is good at reading well-structured pages - tables, lists, FAQ blocks with clear answers. Marketing prose with claims buried in adjectives gets less citation weight than structured, declarative content.
Long-context awareness
Claude's larger context window means it can read full long articles, not just snippets. Long-form content with internal coherence and consistent terminology benefits. Short, fragmented content less so.
Wikipedia, Wikidata + the open-web entity graph
Claude leans on institutional sources more heavily than ChatGPT does. STRIDEC's 2026 analysis and Tilio's UK-focused breakdown both put Wikipedia and Wikidata at the centre of the inclusion-signal picture - if your entity is verifiable in the open-web entity graph (Wikipedia article, Wikidata Q-number, sameAs in your Organization schema), Claude treats you as a real entity worth referencing. No Wikidata entry, and you are often invisible at the entity layer even with strong on-page content.
Where Claude visibility usually breaks down.
The common things that keep Claude away from otherwise good sites. The first three are structural; the fourth is about content depth.
Claims without sources
The page makes a confident assertion but does not link the underlying study, dataset or expert. Claude declines to repeat unverifiable claims and quotes a competitor who did link a source.
Summarising someone else's work
The page rewrites a third-party study without adding new analysis. Claude prefers the primary source and skips the rewrite. The fix is to add original analysis, original data, original framing.
Marketing prose hides the facts
The page has the right facts but they are wrapped in three paragraphs of brand voice. Claude can read it but the extraction is messy. Restructure with clear answer blocks and the citation odds shift.
ClaudeBot blocked
Robots.txt blocks ClaudeBot or anthropic-ai explicitly, or by accident through a deny-all rule. The site is invisible to Claude's training and live web search inputs.
What Claude treats as a trustworthy source.
1. Source-linked content restructure
Every load-bearing claim gets a source - inline link to a study, dataset, or named expert. The structural pattern Claude treats as verifiable expertise.
2. Primary source positioning
Where your firm IS the primary source, the page makes that explicit through Person schema, Organization schema, and clear "original research" framing. Claude rewards primacy when it can confirm it.
3. Structured factual layer
Key facts get pulled into tables, lists and FAQ blocks. The marketing prose stays but the citable substance is structurally separate and easy to extract.
4. ClaudeBot crawl confirmation
Verify ClaudeBot and anthropic-ai can crawl. Monitor logs to confirm citation-eligible content is being indexed.
5. Wikipedia / Wikidata entity establishment
Get the brand (or founder) verifiably present in Wikipedia/Wikidata where notability justifies it. Add the Wikidata Q-number to Organization schema as sameAs. The single biggest entity-layer signal Claude weights - missing it caps your citation ceiling regardless of on-page work.
AI Bot Robots.txt Tester
Checks ClaudeBot and anthropic-ai access to your site.
Test your robots.txtAI Search Visibility Tool
Scores your brand's visibility across Claude, ChatGPT, Perplexity and Gemini.
Score your visibilityAI Schema Validator
Validates schema for AI extraction - Person, Article, Organization in particular.
Validate schemaWhat clients ask about Claude visibility.
How is Claude different from ChatGPT and Perplexity for citation?
Claude is conservative about citation - it prefers verifiable sources and tends to hedge on claims it cannot back up. Where ChatGPT might paraphrase a Reddit thread, Claude is more likely to cite a primary source or refuse to answer. The implication: structured, fact-led content with clear sources wins citations from Claude.
Does Claude have a live web crawler?
Claude has web search through Claude.ai (when enabled) and operates the ClaudeBot crawler for training. Live citations rely on web search returning sources that Claude then evaluates and quotes. Without web search enabled, Claude only knows what was in training data up to its cutoff.
Why does Claude refuse to answer some questions other AIs answer?
Anthropic builds Claude with a Constitutional AI approach. It is more willing to decline an answer when sources are weak or contested. For SEO and GEO work this is useful intel - if Claude refuses to recommend or compare in your category, your content lacks the structured proof needed to be confidently cited.
Should I optimise differently for Claude versus ChatGPT?
Some overlap, some divergence. Both reward clear authoritative content. Claude penalises unverifiable claims more aggressively, so the work skews toward fact-based content with primary sources cited. ChatGPT rewards community presence more, so Reddit and similar matter more for ChatGPT than Claude.
Can I block ClaudeBot from crawling my site?
Yes, via robots.txt with the ClaudeBot user-agent. Doing so removes you from future training data and from current web-search citations through Claude. For most sites, the citation reach outweighs the principle, but enterprise sites with sensitive data have reasons to block.
How important is Wikipedia or Wikidata for getting cited by Claude?
Very. Both STRIDEC's 2026 analysis and Tilio's UK-focused study put Wikipedia and Wikidata at the centre of the inclusion-signal picture. Claude leans heavily on the open-web entity graph - if your brand or founder is verifiably present in Wikipedia/Wikidata with a Q-number and your Organization schema includes the Wikidata URL in sameAs, Claude treats you as a real entity. No Wikidata entry caps your citation ceiling, especially for category-level queries where Claude is choosing between several similar-looking brands.
Do I need to publish original data and research to get cited?
It dramatically increases the odds. Claude rewards primary sources with original data and clear methodology. A page that cites a study you can verify gets weighted higher than a page that re-summarises someone else's. If you run any kind of survey, audit, benchmark or cohort analysis, publishing the methodology and numbers properly turns that into a citation magnet across Claude, Perplexity and AI Overviews simultaneously.
Find out whether Claude cites your brand.
Three queries you wish Claude recommended you for. Send them over, we run them together on a call, and dissect what Claude actually trusts - what it quotes, what it hedges and what it refuses to recommend in your category.
Book the test session