Claude is the engine most people optimise for last and understand least, and it rewards a different kind of attention. Its web search runs through Brave, not Bing or Google, so your usual rankings do not set the candidate pool. Anthropic gives publishers three separate crawlers to control, more granular than anyone else. And unlike Google, Claude actually reads llms.txt. The shared GEO fundamentals still apply, but the index you court and the bots you allow are particular to Claude. This is the per-engine playbook, sitting under the broader pillar on generative engine optimization.
Key takeaways
- What Claude rewards — Brave ranking first, then clarity and verifiability: a clean answer, primary-source citations, and content a careful reader could fact-check. llms.txt is read here too.
- What it cares less about — Bing and Google ranking specifically — Claude retrieves through Brave, so the usual index leaders are not the entry gate. Keyword density and heavy formatting do little.
- The crawlers — Three of them: ClaudeBot (training), Claude-User (real-time fetches during an answer), Claude-SearchBot (search index). Allow Claude-User and Claude-SearchBot to be eligible for citations.
- How long it takes — Tied to Brave indexing and the search-index refresh. Newer and less documented than ChatGPT, so measure rather than assume a timeline.
How Claude decides what to cite
Claude's web search is a tool the model invokes when it judges that a question needs current or external information. The retrieval loop is short, and its backend is the part that surprises people.
- Invoke the web search tool when the query calls for external evidence.
- Query Brave Search, whose independent index returns the candidate results.
- Fetch pages as needed with Claude-User, and draw on the Claude-SearchBot index.
- Evaluate the retrieved passages for clarity and factual grounding.
- Compose the answer with citations to the sources it actually used.
Two implications follow. First, Brave is the gate, not Google or Bing — a structurally different candidate pool from every other engine in this cluster, and the reason Claude visibility has to be optimised and measured on its own terms. Second, because Anthropic exposes three independently-controllable crawlers, a misconfigured robots.txt is a more common cause of invisibility here than anywhere else: teams that block "AI bots" wholesale often knock out Claude-User and Claude-SearchBot without meaning to, and quietly disappear from Claude's answers.
The seven-step playbook
Tactics ordered by leverage, calibrated for Claude. Two of the seven — the Brave focus and the bot configuration — are specific to Claude and have no equivalent on the other engines.
- Configure the three Claude bots deliberately. Allow Claude-User and Claude-SearchBot, which together make you eligible for live answers and the search index. Decide on ClaudeBot, the training crawler, as a content-licensing choice. Audit your robots.txt for blanket AI-bot blocks that catch the wrong ones — this is the most common reason a site is absent from Claude.
- Earn Brave Search visibility. Because Brave is the retrieval backend, Brave ranking is the entry gate, and Brave runs its own independent index rather than reselling Google's or Bing's. Make sure Brave can crawl you, and treat Brave as a distinct surface worth a look in its own webmaster tooling. This is the lever no other engine in the cluster shares.
- Lead with a clear, verifiable answer. Claude appears to favour passages that are unambiguous and checkable. A self-contained answer near the top, written so a careful reader could verify it, aligns with how the model evaluates candidates. Vague or promotional phrasing is the opposite of what gets pulled.
- Cite primary sources with care. Factual verifiability seems to weigh heavily, and a page that grounds its claims in datable primary sources reads as more trustworthy to the model. The sourced-statistic and authoritative-quotation interventions from the GEO literature apply here, with verifiability rather than recency as the reason they work.
- Publish an llms.txt. Claude-SearchBot parses it, and CTAIO Labs measured Claude among the engines that responded to an llms.txt rollout. It is low-effort and reversible. This is the clearest case where per-engine calibration matters: the same file Google ignores can earn you citations in Claude.
- Render content server-side. Put the answer, headings, and structured data in the initial HTML rather than behind hydration, so Claude-User and Claude-SearchBot see the content without depending on client-side execution.
- Build topical depth and authority. Claude's evaluation step rewards pages that read as authoritative on their topic. Comprehensive coverage, internal linking that demonstrates depth, and a recognisable author and organisation all support the model's read of you as a trustworthy source — the slow lever, as on every engine, and the durable one.
What's different from ChatGPT, Perplexity, and Google
Claude's divergence is structural: a different retrieval backend changes the candidate pool before any on-page tactic comes into play. CTAIO Labs measured per-engine citation deltas, Claude included, in the framework test.
- ChatGPT grounds in Bing; Claude grounds in Brave. A strong ChatGPT page is not automatically a strong Claude page, because the indexes differ. Both read llms.txt, which is a point of agreement. The ChatGPT playbook is at how to rank in ChatGPT.
- Perplexity runs its own crawler and prizes freshness; Claude leans on Brave and prizes verifiability. The Perplexity playbook is at how to rank in Perplexity.
- Google AI Overviews use Google's index and ignore llms.txt; Claude uses Brave and reads it. These two are near-opposites on the llms.txt question, which is the cleanest argument against a one-size checklist. The Google playbook is at how to rank in Google AI Overviews.
- The bot granularity is Claude's own: three independently-controllable crawlers where other engines give you one or two, which makes robots.txt hygiene a bigger factor for Claude than for any other engine.
Measurement
Claude is the hardest engine to measure directly, so the loop leans more on trackers and indirect signals. Build it in three layers:
- Citation tracker. Profound, Peec AI, AthenaHQ, Otterly, or one of the others, choosing one whose Claude coverage is mature — it varies more than for ChatGPT. The Radar's scored shortlist is at 6 GEO Tools the Radar Actually Recommends; CTAIO Labs tested ten head-to-head in the visibility tools test.
- Server-log analysis for the Claude bots. Confirm Claude-User and Claude-SearchBot are actually reaching your priority pages. Because misconfiguration is the most common failure here, the log check is more valuable for Claude than for engines with a single crawler.
- Branded-search lift in GSC. With clean referral data scarce, the cleanest downstream signal that Claude citations are building brand equity is users later searching your name directly. The lag is months; the signal is unambiguous.
Field evidence
Related reads
Frequently asked questions
How does Claude decide what to cite?
When Claude uses its web search tool, it issues queries to Brave Search, retrieves the candidate results Brave returns, and evaluates the retrieved passages before composing an answer with citations to the sources it used. Brave's ranking therefore sets the candidate pool, and Claude's own assessment of clarity and factual grounding decides which of those candidates make it into the answer. Anthropic has not published the exact selection weights, so the on-page tactics here are informed inference from how the system behaves.
Which crawler should I allow for Claude?
Anthropic runs three. ClaudeBot collects training data. Claude-User fetches a specific page in real time when a user's Claude session needs it. Claude-SearchBot builds the index for Claude's search features. To be eligible for citations you want Claude-User and Claude-SearchBot allowed; blocking ClaudeBot alone is a training opt-out that does not affect citations. This three-way split, clarified by Anthropic in early 2026, is the most granular publisher control any major engine offers.
Does Claude use Bing or Google like other engines?
Neither. Claude's web search retrieves through Brave Search, which maintains its own independent index. This is the single most important fact for optimising for Claude: your Google and Bing rankings do not directly determine your Claude candidate pool, your Brave visibility does. Brave's index is smaller and has its own ranking, so a page strong on Google can be weak on Brave and therefore weak in Claude, and vice versa.
Does llms.txt help with Claude?
It appears to. Unlike Google, which states its systems do not reference llms.txt, Anthropic has not disavowed the file, and Claude-SearchBot parses it. CTAIO Labs measured a positive early citation delta from rolling llms.txt out on test sites, with Claude among the engines that responded. It is low-effort and reversible, so for Claude specifically it is worth publishing — the opposite of the recommendation for Google.
How is getting cited by Claude different from ChatGPT or Perplexity?
The retrieval backend is the headline difference: ChatGPT grounds in Bing, Perplexity in its own crawler, and Claude in Brave. That means the three draw from different candidate pools for the same query. Claude also gives publishers more granular bot control than either, and it reads llms.txt where Google does not. The shared advice — clear answers, primary sources, server-rendered content — holds across all three; the index you optimise for is what changes.
Does updating content frequently help with Claude?
Freshness helps less than it does on Perplexity and is harder to characterise precisely, because Anthropic has not documented a freshness weight and Brave's index refresh governs how quickly updates appear. Keeping pages accurate and current is good practice and supports the verifiability Claude seems to favour, but there is no evidence that a Perplexity-style fortnightly cadence is necessary. Prioritise correctness over recency churn.
Does Claude pass referral traffic to my analytics?
Rarely in a clean, attributable way. Many Claude sessions happen in the app or via the API, where a referrer is not passed, so claude.ai referrals understate true influence even more than other engines. Treat Claude as a brand-influence channel measured indirectly — through citation trackers and branded-search lift — rather than one you can read directly in a GA4 referral report.
What tools measure my Claude citation rate?
Several LLM-visibility trackers — Profound, Peec AI, AthenaHQ, Otterly, and others — include Claude among the engines they monitor, though coverage of Claude is generally less mature than coverage of ChatGPT and Perplexity. CTAIO Labs scored ten trackers head-to-head and noted which handled Claude well; the Radar's shortlist of the six that earned a recommendation is at /en/radar/geo-tools/.
How does Claude decide what to cite?
When Claude uses its web search tool, it issues queries to Brave Search, retrieves the candidate results Brave returns, and evaluates the retrieved passages before composing an answer with citations to the sources it used. Brave's ranking therefore sets the candidate pool, and Claude's own assessment of clarity and factual grounding decides which of those candidates make it into the answer. Anthropic has not published the exact selection weights, so the on-page tactics here are informed inference from how the system behaves.
Which crawler should I allow for Claude?
Anthropic runs three. ClaudeBot collects training data. Claude-User fetches a specific page in real time when a user's Claude session needs it. Claude-SearchBot builds the index for Claude's search features. To be eligible for citations you want Claude-User and Claude-SearchBot allowed; blocking ClaudeBot alone is a training opt-out that does not affect citations. This three-way split, clarified by Anthropic in early 2026, is the most granular publisher control any major engine offers.
Does Claude use Bing or Google like other engines?
Neither. Claude's web search retrieves through Brave Search, which maintains its own independent index. This is the single most important fact for optimising for Claude: your Google and Bing rankings do not directly determine your Claude candidate pool, your Brave visibility does. Brave's index is smaller and has its own ranking, so a page strong on Google can be weak on Brave and therefore weak in Claude, and vice versa.
Does llms.txt help with Claude?
It appears to. Unlike Google, which states its systems do not reference llms.txt, Anthropic has not disavowed the file, and Claude-SearchBot parses it. CTAIO Labs measured a positive early citation delta from rolling llms.txt out on test sites, with Claude among the engines that responded. It is low-effort and reversible, so for Claude specifically it is worth publishing — the opposite of the recommendation for Google.
How is getting cited by Claude different from ChatGPT or Perplexity?
The retrieval backend is the headline difference: ChatGPT grounds in Bing, Perplexity in its own crawler, and Claude in Brave. That means the three draw from different candidate pools for the same query. Claude also gives publishers more granular bot control than either, and it reads llms.txt where Google does not. The shared advice — clear answers, primary sources, server-rendered content — holds across all three; the index you optimise for is what changes.
Does updating content frequently help with Claude?
Freshness helps less than it does on Perplexity and is harder to characterise precisely, because Anthropic has not documented a freshness weight and Brave's index refresh governs how quickly updates appear. Keeping pages accurate and current is good practice and supports the verifiability Claude seems to favour, but there is no evidence that a Perplexity-style fortnightly cadence is necessary. Prioritise correctness over recency churn.
Does Claude pass referral traffic to my analytics?
Rarely in a clean, attributable way. Many Claude sessions happen in the app or via the API, where a referrer is not passed, so claude.ai referrals understate true influence even more than other engines. Treat Claude as a brand-influence channel measured indirectly — through citation trackers and branded-search lift — rather than one you can read directly in a GA4 referral report.
What tools measure my Claude citation rate?
Several LLM-visibility trackers — Profound, Peec AI, AthenaHQ, Otterly, and others — include Claude among the engines they monitor, though coverage of Claude is generally less mature than coverage of ChatGPT and Perplexity. CTAIO Labs scored ten trackers head-to-head and noted which handled Claude well; the Radar's shortlist of the six that earned a recommendation is at /en/radar/geo-tools/.
Ready to Find the Right AI Tools?
Browse our data-driven rankings to find the best AI tools for your team.