Claude's citation behavior: what content earns an Anthropic AI mention
OtterlyAI's June 2026 study of 379,321 citations found company and product domains earn 64% of Claude citations. Here's what structural signals drive Claude citation eligibility.
B
Stephan Charles
15 min read
August 13, 2026
Share
json
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "How does Claude decide what content to cite?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Claude cites content that meets three conditions at once: the claim attributes to a named, verifiable entity; the passage is self-contained within roughly 200 tokens without requiring context from adjacent paragraphs; and the source appears in Brave Search's index. OtterlyAI's June 2026 analysis of 379,321 Claude citations found that company and product domains account for 64% of all citations. Social media earns 3.99%."
}
},
{
"@type": "Question",
"name": "What is the difference between Claude, ChatGPT, and Perplexity citations?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Claude surfaces roughly 7.92 sources per response. Perplexity surfaces 21.87. Claude's social citation rate is 3.99%, the lowest of any major AI engine, versus Perplexity's 11.3% and Google AI Overviews' 15.3%. A cross-platform study found that only 11% of domains cited by one AI platform appear across multiple platforms, meaning platform-specific optimization matters."
}
},
{
"@type": "Question",
"name": "Do you need to rank on Brave Search to get cited by Claude?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Anthropic confirmed in July 2025 that Claude's web search tool routes queries through Brave's index. An independent analysis found an 86.7% overlap between Claude's cited pages and Brave's top organic results, with a p-value below 0.0001. Pages absent from Brave's index are invisible to Claude's search tool regardless of their Google ranking."
}
},
{
"@type": "Question",
"name": "Does Constitutional AI affect what Claude will cite?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Anthropic's Constitutional AI training builds a prior against promotional, vague, or unverifiable content. Claude is less likely to cite self-promotional brand content without third-party corroboration and favors sources with visible methodology. CAI operates at training time rather than as a real-time content filter, but produces consistent citation avoidance of purely promotional material."
}
},
{
"@type": "Question",
"name": "Which content types does Claude cite most often?",
"acceptedAnswer": {
"@type": "Answer",
"text": "OtterlyAI's June 2026 study of 379,321 Claude citations found company and product domains account for 64% of citations, followed by News and Media at 14.9% and Educational and Academic at 9.1%. Social media accounts for 0.9%, and Reddit was cited zero times in their sample. Claude prioritizes primary brand content, official documentation, and practitioner explainers with specific claims."
}
}
]
}
Track your AI visibility now
See exactly how AI engines like ChatGPT, Perplexity, and Gemini perceive your brand.
The BrandCited team covers GEO, AI search optimization, and brand visibility strategy. We publish research, practical guides, and product updates every week.
Spot an error in this article? Report it.
By Stephan Charles | Last fact-checked: <time datetime="2026-08-13">2026-08-13</time>
Claude cites content differently from every other major AI platform, and what works for Google AI Overviews doesn't transfer. OtterlyAI's June 2026 analysis of 379,321 Claude citations found that company and product domains account for 64% of all Claude citations, while social media earns 3.99%, making Claude the most brand-favorable and social-averse engine on the market. The underlying driver is Brave Search: an independent analysis found an 86.7% overlap between Claude's cited pages and Brave's top organic results. BrandCited is an AI brand visibility platform that monitors Claude citation rates alongside 8 other AI engines and runs 30+ structural audit checks per brand to surface exactly which content signals your brand is missing.
Run a free BrandCited scan at brandcited.ai to see your Claude citation rate in 30 seconds.
Claude cites content that meets three conditions at once: the claim attributes to a named, verifiable entity; the passage is self-contained within roughly 200 tokens without context from adjacent sections; and the source appears in Brave Search's index.
When Claude uses web search, it routes queries through Brave's index via Anthropic's search tool integration. Claude's retrieval mechanism selects passages with high semantic similarity to the query, weighted by entity coherence within the passage. A passage where "BrandCited" appears as the entity noun in every sentence outperforms a passage that alternates "BrandCited," "the platform," "the tool," and "this service," even when the underlying content is identical.
OtterlyAI's June 2026 study of 379,321 citations across 16,406 domains found that company and product domains account for 64.0% of all Claude citations. News and Media accounts for 14.9%. Social media accounts for 0.9%, and Reddit was cited zero times in OtterlyAI's sample.
Claude favors stable, established content over fresh content. Unlike Perplexity's proprietary index, which weights recency aggressively, Claude's Brave-sourced retrieval favors authority and structural credibility over publication date. A well-structured page from a high-authority domain published 18 months ago will outperform a week-old blog post on the same topic.
Claude evaluates passages at the chunk level, not the page level. Since March 2026, Claude attributes citations to specific passages within a page rather than to the page URL. A page with one excellent 200-token section gets cited for that section. Nine mediocre sections don't help.
Why does Claude's citation rate skew so heavily toward brand-owned content?#
Claude cites brand-owned domains at 64% because Anthropic's Constitutional AI training built a preference for primary sources with named entity attribution over aggregated, summarized, or socially-sourced content.
Anthropic's Constitutional AI framework trains Claude to weigh source transparency and epistemic credibility when generating responses. The practical citation effect: Claude avoids self-promotional content without third-party corroboration and favors sources that present falsifiable claims with visible methodology. A brand's official product documentation, a company's own research with a clear methodology, or a named practitioner's signed analysis all satisfy Claude's credibility threshold in ways that marketing blog posts and social posts don't.
This creates an opportunity for brands that understand it. Your official documentation, your product pages written with specific claims and named entity attribution, and your signed research are what Claude's retrieval system prioritizes. Most brands spend GEO budget on social and PR and then wonder why Claude doesn't cite them.
Third-party corroboration is the entry requirement.Oltre.ai's technical breakdown of Claude citations found that Claude won't cite a brand's own description of itself if it can't find third-party corroboration. Wikipedia presence and mentions on authoritative external platforms are the most important entity-establishment step before investing in content structure.
Claude visitor conversion rate reaches 16.8%, the highest of any AI engine per Omnibound's GEO statistics compilation, despite Claude sending lower referral volume than ChatGPT or Perplexity. Fewer citations from Claude reach buyers who are already researching and deciding.
How does Claude's citation logic differ from ChatGPT, Perplexity, and Gemini?#
Claude surfaces roughly 7.92 sources per response, while Perplexity surfaces 21.87. The gap reflects fundamentally different retrieval philosophies, not content quality differences.
Yext's 2026 cross-engine analysis found that only 11% of domains cited by one AI platform appear across multiple platforms. The same content strategy won't produce equal citation results across all AI engines. Platform-specific optimization isn't optional.
Claude vs. Perplexity. Perplexity cites on nearly every response and weights named author bylines with LinkedIn sameAs markup as its single highest citation signal. Claude doesn't share this author-identity weighting. Claude weights entity coherence and passage self-containment over byline visibility. An article with a named author but pronoun-heavy writing will outperform in Perplexity and underperform in Claude.
Claude vs. ChatGPT. ChatGPT's web search runs through Bing's index. FAQPage schema lifts ChatGPT citation probability by 40% per Conbersa's FAQ schema study. Claude doesn't use Bing's index and doesn't weight FAQPage schema in the same way. The same FAQ schema investment that drives ChatGPT citations transfers minimally to Claude citation, though it improves content structure in ways that benefit all engines. For a full breakdown of FAQ schema for ChatGPT, see BrandCited's guide to FAQ schema for AI citations.
Claude vs. Gemini. Gemini grounds responses against Google's Knowledge Graph before retrieving web content. Brands with Wikidata entities, Crunchbase listings, and G2 profiles appear in Gemini responses at higher rates regardless of Brave Search ranking. Claude doesn't use Google's Knowledge Graph. The entity-graph signals that drive Gemini citations don't transfer to Claude. See BrandCited's guide to winning Gemini citations for the Gemini-specific signals.
Claude's social citation rate of 3.99% is the lowest of any major AI engine, nearly four times lower than Google AI Overviews (15.3%) and three times lower than Perplexity (11.3%). Building an AI visibility strategy around social and PR coverage will produce results in every engine except Claude.
Do you need to rank on Brave Search to get cited by Claude?#
Brave Search ranking is a prerequisite for Claude citation eligibility in practice. A page absent from Brave's index is invisible to Claude's web search tool regardless of its Google ranking.
Anthropic confirmed in July 2025 that Claude's web search tool routes queries through Brave Search as its primary index. GEO consultant Raouf Douihech's analysis of a Profound comparative study found an 86.7% overlap between Claude's cited pages and Brave Search's top organic results, with a p-value below 0.0001. The overlap is not chance.
Brave runs an independent crawler (BraveBot/1.0) that doesn't license Google or Bing data. A page that ranks at position 1 on Google but has poor Brave index coverage can be outranked in Claude citations by a page at position 8 on Google with strong Brave coverage. This is the most overlooked signal in GEO optimization for Claude.
Track your AI visibility for free
See how ChatGPT, Claude, Gemini, and 4 other AI platforms mention your brand.
Brave crossed 1.6 billion monthly queries in 2026 and 100 million monthly active users, per Scale Xpert's Brave Search analysis. Ranking in Brave's index now targets two engines at once: Brave's own AI Answers feature and Claude's web search tool.
Brave down-ranks pages with heavy tracker footprints. Brave's ranking model applies a measurable penalty for pages with three or more third-party tracking scripts. Blog pages with retargeting pixels, third-party ad scripts, and multiple analytics tools are down-ranked in Brave's index, making them less likely to appear in Claude's citation candidates.
Submit your site to Brave Webmaster Tools to verify ownership, submit your XML sitemap, and monitor BraveBot crawl status.
How do you structure content for Claude's embedding retrieval system?#
Write every H2 section as a 150 to 200-word self-contained block that answers one discrete question without requiring context from adjacent sections. Claude's retrieval system evaluates passages at the chunk level, and chunks that depend on prior paragraphs for meaning are not retrieved.
Four structural rules that match Claude's citation signals:
Use one entity noun throughout the article. Pick one term for your brand and one for the main concept, then keep them across every paragraph. "BrandCited monitors citation rates" outperforms "BrandCited / the platform / this tool monitors citation rates" in embedding similarity scores. Pronoun resolution — using "it" or "this" to refer to an entity named in a prior sentence — degrades the semantic coherence of the chunk.
Open every section with the direct answer. The first sentence of each H2 section must answer the question in the heading with no setup. This is the sentence Claude's retrieval system extracts as the citation candidate. A section that opens with context or framing before reaching the answer gets the framing extracted instead.
Keep paragraphs to three to five sentences. Longer paragraphs mix multiple entity claims, which reduces the embedding similarity of the chunk. Three to five sentences per paragraph, one primary entity or claim per paragraph.
Link to official documentation at the end of every section. Claude's retrieval system weights passages that reference official documentation, named studies, or authoritative sources over passages with no external grounding. Princeton's foundational GEO study demonstrated that adding citations, statistics, and quotations increases AI citation visibility by 30 to 40%.
Anthropic: Claude web search is now globally available on all plans as of August 2026. The Brave Search pipeline that powers Claude citations is a standard capability on every account tier. (Anthropic)
OpenAI: GPT-5.6 Sol retuned for factual reliability, with 68% fewer factual errors than GPT-5.5 Instant. GPT-5.6 Luna is now the default model for all free and Go users with no text-chat limits, putting ChatGPT citation volume into roughly 1 billion weekly active users. (TechCrunch)
Perplexity: Comet landed on iOS with multimodal Deep Research. The Agent API now accepts Claude Opus 4.7, GPT-5.5, and Grok 4.20 Reasoning as available models. Perplexity holds 100 million MAUs with an estimated 138 million monthly visits. (Perplexity changelog)
Google: AI Overviews now appear on approximately 48% of all tracked queries, up 58% year-over-year. Cited brands in Google AI Overviews earn 120% more organic clicks per impression than uncited brands on the same SERP. (Heroic Rankings)
GEO market: Only 30% of brands remain visible in consecutive AI responses for the same query. Pages with well-organized headings are 2.8x more likely to earn AI citations across all platforms. (Superlines 2026)
How does BrandCited audit Claude citation signals?#
BrandCited's audit engine checks four structural signals that predict Claude citation eligibility: Brave Search index status, entity noun use across your highest-traffic pages, passage self-containment scores for H2 sections, and third-party corroboration coverage.
BrandCited catches two Claude-specific gaps that standard SEO tools miss: pages absent from Brave's index despite strong Google rankings, which generates zero Claude citation exposure; and entity noun alternation patterns where a brand switches between its name and generic descriptors mid-article, which reduces embedding coherence.
Run a free BrandCited scan at brandcited.ai to see your Claude citation rate alongside all 8 other AI platforms, with each gap ranked by citation impact.
1Submit your site to Brave Webmaster Tools today. Go to brave.com/leo/webmaster, verify ownership, and submit your XML sitemap. Check BraveBot's crawl log. If BraveBot hasn't visited your key pages in the last 30 days, your Claude citation exposure is zero regardless of content quality.
1Audit your five highest-traffic pages for entity noun use. Open each page and count how many different terms you use for your brand. If you use more than two variants, rewrite to use the brand name throughout. This single change improves embedding similarity for Claude's chunk-level retrieval.
1Rewrite your H2 section openings. The first sentence of every H2 section must answer the question in the heading with no setup. Open a page, read the first sentence of each H2, and ask: does this sentence, read alone, answer the H2 question? If not, rewrite it.
1Close your third-party corroboration gaps. Search for your brand name on Wikipedia, Wikidata, Crunchbase, and G2. If your brand doesn't appear in any of these, Claude's retrieval system can't verify it as a real, authoritative entity. A Crunchbase listing is free. A G2 listing is free. Create them before investing further in content optimization.
1Remove ad pixels and retargeting scripts from blog pages. Brave's algorithm penalizes pages with three or more third-party tracking scripts. Every pixel on your blog page reduces Brave index ranking and, by extension, Claude citation eligibility. Remove retargeting and ad pixels from blog routes.
1Add one external link to official documentation per H2 section. Claude's retrieval system weights passages that reference official documentation, named studies, or authoritative sources. One authoritative outbound link per section.
</section>
Run a free AI visibility audit on your brand at brandcited.ai. You'll see your Claude citation rate alongside 8 other AI platforms in 30 seconds, with every structural gap ranked by citation impact.
Claude cites content that meets three conditions at once: the claim attributes to a named, verifiable entity; the passage is self-contained within roughly 200 tokens without requiring context from adjacent paragraphs; and the source appears in Brave Search's index. OtterlyAI's June 2026 analysis of 379,321 Claude citations found that company and product domains account for 64% of all citations. Social media earns 3.99%, the lowest of any major AI engine.
What is the difference between Claude, ChatGPT, and Perplexity citations?
Claude surfaces roughly 7.92 sources per response versus Perplexity's 21.87. Claude's social citation rate is 3.99%, compared to Perplexity's 11.3% and Google AI Overviews' 15.3%. A cross-platform study found that only 11% of domains cited by one AI platform appear across multiple platforms. FAQPage schema drives ChatGPT citations but doesn't transfer to Claude; named author bylines drive Perplexity citations but don't drive Claude citations.
Do you need to rank on Brave Search to get cited by Claude?
Brave Search ranking is a prerequisite for Claude citation eligibility in practice. Anthropic confirmed in July 2025 that Claude's web search tool routes queries through Brave's index. An independent analysis found an 86.7% overlap between Claude's cited pages and Brave's top organic results, with a p-value below 0.0001. Pages absent from Brave's index are invisible to Claude's search tool regardless of their Google ranking.
Does Constitutional AI affect what Claude will cite?
Anthropic's Constitutional AI training builds a prior against promotional, vague, or unverifiable content. Claude is less likely to cite self-promotional brand content without third-party corroboration and favors sources with visible methodology and falsifiable claims. The practical implication: third-party corroboration on Wikipedia, G2, and Crunchbase is a prerequisite for brand citation eligibility.
Which content types does Claude cite most often?
OtterlyAI's June 2026 study of 379,321 Claude citations found company and product domains account for 64% of citations, followed by News and Media at 14.9% and Educational and Academic at 9.1%. Social media accounts for 0.9%, and Reddit was cited zero times in their sample. Claude prioritizes primary brand content, official documentation, and practitioner explainers with specific claims over social content and aggregators.
Sources: [OtterlyAI — Claude AI Citation Study June 2026](https://otterly.ai/blog/claude-ai-citation-study/) | [OtterlyAI — AI Citations Report 2026](https://otterly.ai/blog/the-ai-citations-report-2026/) | [Raouf Douihech — How to Get Cited by Claude](https://raoufdouihech.com/en/ai-seo/get-cited-by-claude) | [Seoteric — Claude Visibility and Brave Search](https://www.seoteric.com/claude-visibility-hinges-on-brave-search-rankings-what-seos-should-do/) | [Yext — How AI Engines Decide What to Cite](https://www.yext.com/blog/how-chatgpt-perplexity-gemini-claude-decide-what-to-cite) | [Oltre.ai — Technical Breakdown of Claude Citations](https://www.oltre.ai/blog/how-claude-picks-sources-technical-breakdown-claude-citations/) | [Anthropic — Constitutional AI Research](https://www.anthropic.com/research/constitutional-ai-harmlessness-from-ai-feedback) | [Anthropic — Claude Web Search](https://www.anthropic.com/news/web-search) | [Princeton / Georgia Tech — GEO Study](https://arxiv.org/pdf/2311.09735) | [Superlines — AI Search Statistics 2026](https://www.superlines.io/articles/ai-search-statistics/) | [Heroic Rankings — Google AI Overview Statistics 2026](https://heroicrankings.com/seo/managed/google-ai-overview-statistics-2026/) | [Omnibound — GEO Statistics](https://www.omnibound.ai/blog/generative-engine-optimization-statistics) | [Conbersa — FAQ Schema AI Citations](https://www.conbersa.ai/learn/faq-schema-for-ai-citations) | [Scale Xpert — Brave Search](https://scale-xpert.com/what-is-brave-search-submit-url-geo-seo/) | [Discoveredlabs — AI Citation Patterns](https://discoveredlabs.com/blog/ai-citation-patterns-how-chatgpt-claude-and-perplexity-choose-sources/)
