Get cited by Claude.

The systems behind Claude's answers.

Claude.ai is a stack of model, retrieval, and tool layers — each playing a different role in whether a domain gets cited.

Claude models (Opus / Sonnet / Haiku). Anthropic's frontier model family, with Opus, Sonnet, and Haiku as quality/speed tiers. Closed-book recall is rooted here — the model only "knows" what it learned during training. Anthropic publishes more about training-data curation than most labs and has been increasingly explicit about the role of high-trust sources in training.

Web Search (Brave-powered). When Claude needs current information, it queries Brave Search and grounds the answer in Brave's results. The reranker prefers primary sources, official documentation, and authoritative publishers — the same biases visible in Anthropic's published research on factuality.

Tool use & agentic actions. Claude can invoke tools — including a fetch tool that retrieves a specific URL and reads its content. This is the surface where Claude-User agents live. Sites that block Claude-User break every agentic workflow involving live page reads.

Claude Code & developer surfaces. A separate, fast-growing Claude product running at $2.5B in annualized revenue alone. For technical content — documentation, API references, developer tooling — being cited by Claude Code is its own GEO outcome distinct from claude.ai citations.

Constitutional AI. Anthropic's training methodology biases the model toward acknowledging uncertainty and citing primary sources. The downstream behavior is observable: Claude is more likely to refuse than fabricate, and more likely to attribute than synthesize without attribution.

Why Claude prefers the source.

Claude consistently cites the most authoritative primary source for a claim, even when secondary sources rank higher in standard search. Original research, official documentation, government and academic publications, and manufacturer-direct product pages get cited disproportionately to their backlink-graph weight. This is the single biggest behavioral difference between Claude and Bing-derived AI assistants.

Implication for brands. The version of your content that's most likely to be cited is the version on your own primary domain — not the version syndicated to a publisher, not the version paraphrased on a roundup site. For Claude GEO, owned-content investment compounds harder than off-site citation acquisition does.

Implication for content. Claude rewards content that reads like primary research — original data, firsthand experience, technical depth — over content that summarizes or rewords what already exists. If you can be the canonical source for a category-defining claim, Claude will cite you over the ten secondary sources that summarize you.

Implication for citation density. Pages that cite primary sources clearly and inline are themselves rated as more credible by Claude's reranker. The model has learned to use citation patterns as a proxy for trustworthiness — both inbound and outbound.

Letting Claude actually read your site.

Claude has multiple user-agent identities, each with its own role in the GEO pipeline. The technical layer matters less than for ChatGPT (Brave handles the search-side crawling) but the live-fetch pipeline is where most silent failures happen.

ClaudeBot (anthropic-ai). Anthropic's training-data crawler. Allowing it in robots.txt feeds the next training cycle. Like GPTBot, it's commonly blocked by default in Cloudflare's bot-management rules — and the block is silent.

Claude-User. The live user-agent for in-conversation fetches and tool use. Blocking Claude-User breaks every "summarize this page" or agent-driven retrieval workflow involving your URLs. Increasingly important as agentic use cases scale.

Claude-SearchBot. A more recent crawler used for Anthropic's emerging search-and-summary surfaces. Treat it the same as the others — explicit allow in robots.txt, no aggressive WAF blocks.

Brave Search hygiene. Because Claude's web search uses Brave, every Brave-side technical concern compounds into Claude visibility: index coverage on Brave, no over-aggressive bot blocking of Brave's crawler, valid sitemap. Brave SEO and Claude GEO are tightly coupled.

llms.txt + schema. Same baseline as the other GEO platforms — llms.txt deployed, JSON-LD schema for Article / FAQ / Organization / Product / BreadcrumbList. Claude's reranker uses schema as a structural signal, not just a rich-results gateway.