How to Rank on Claude in AI Answers: Web Search, Citations, Brave Index

Quick answer

Canlah AI’s answer in this guide is that you rank on Claude in three moves: allow Claude-SearchBot and Claude-User in robots.txt, earn a Brave Search top-ten position for the queries Claude writes and state each fact in one liftable sentence on your own domain. Brave matters because Claude’s web search runs on the Brave Search index, and a public analysis of roughly 35,000 URLs across 400 queries found that 79.2 percent of the URLs Claude cited also appeared in Brave’s top ten. Your own domain matters because brand and vendor sites took 64.0 percent of 379,321 Claude citations in an Otterly.ai study of June 2026.

Canlah AI runs this kind of work as a managed service and publishes the manual version here so any site owner can test a site without buying a tool. Canlah AI’s standard tracking covers ChatGPT, Gemini, Google AI Overviews and Google AI Mode rather than Claude, so every Claude-specific number below is attributed to a named third-party study rather than to Canlah AI’s own archive.

Treat any single Claude answer as directional. Claude decides for itself whether to search at all, so one run cannot prove a stable position, a share of voice or causation. Freeze a panel of eight to twelve buyer questions and repeat it by engine and language before deciding what to change.

Disclosure: Canlah AI publishes this guide and sells the managed measurement service described near the end. Crawler and search-tool behaviour comes from Anthropic’s documentation and trust portal, the Brave findings from Josh Blyskal’s July 22, 2026 research page and a Profound analysis, the citation mix from an Otterly.ai study of June 2026 and the tool rows from each vendor’s public pricing or product page. All were read on September 23, 2026 and none was tested through a paid account.

What does ranking on Claude mean?

Ranking on Claude is six separate outcomes rather than one position, and a page reaches any of them only in the answers where Claude chooses to search. Anthropic’s documentation states that Claude searches when a request depends on information that is “current, changing, or outside its training data”, and answers from stable knowledge otherwise. Everything a site owner influences sits downstream of that decision.

• Search trigger: whether the question makes Claude search at all or answer from memory with no sources attached.
• Retrieval: whether the page appears in the Brave results Claude reads for the search queries it writes.
• Mention: whether the answer names the brand in prose, with or without a link.
• Citation: whether the answer attaches a web_search_result_location to that sentence pointing at your URL rather than at a directory or a competitor.
• Position: whether the brand is the main recommendation, one name in a shortlist or an aside near the end.
• Accuracy: whether the sentence Claude cites describes the category, market, price or capability correctly.

Branded questions measure retrieval. Unbranded category questions measure the part that produces new buyers.

How Claude decides what to cite

Claude’s citation choice is mostly a Brave ranking outcome: 79.2 percent of Claude’s cited URLs, across roughly 35,000 URLs and 400 queries, also appeared in Brave’s top ten, per Josh Blyskal’s state-of-AEO research of July 22, 2026. Anthropic added Brave Search to its subprocessor list on March 19, 2025, and the entry still names Brave Software for “Web Search” across all products when read on September 23, 2026.

A separate Profound analysis put the overlap between Claude’s cited results and Brave’s top non-sponsored results at 86.7 percent, or 13 of 15 results. Neither study is an Anthropic statement, and they measure different units, but they agree on the direction: the index Claude reads is Brave’s, not Google’s and not Bing’s.

The second decision is the fan-out. Claude writes its own search queries from the buyer question, and Anthropic’s documentation states the process “can repeat multiple times throughout a single request”, with simple factual queries typically using one to three searches and comparative research ten or more. A page that matches the buyer’s literal wording but none of the queries Claude writes from it never enters the pool.

The third decision is the sentence. Citations are always enabled for web search on the Claude API, and each one carries the URL, the title and up to 150 characters of cited_text lifted from the source. That window is the practical unit of Claude SEO: if the answer is spread across a paragraph, there is nothing clean to lift.

Which Anthropic crawler controls which Claude surface

Anthropic documents three separate user agents, and blocking the wrong one removes a site from Claude’s answers while leaving training access untouched. A User-agent: * disallow written as a training opt-out blocks all three at once, which is the easiest way to disappear from Claude search by accident.

Anthropic’s documented user agents, read on September 23, 2026.

User agent	What Anthropic says it does	Where to check	What blocking it costs
ClaudeBot	Collects web content “that could potentially contribute to their training”	robots.txt, then CDN bot rules	Future training use only; Anthropic lists no search effect
Claude-SearchBot	“Navigates the web to improve search result quality for users”	robots.txt, then CDN bot rules	Index coverage for Claude’s web search; Anthropic warns of reduced “visibility and accuracy in user search results”
Claude-User	Fetches a page live “when individuals ask questions to Claude”	robots.txt, CDN and any WAF rate limit	Live retrieval when a user asks Claude about your page or brand

Source: Anthropic’s crawler support article, read September 23, 2026. Anthropic states its bots honour robots.txt and supports the non-standard Crawl-delay directive. It publishes no propagation window, so re-test after a change rather than assuming it is live.

Read your robots.txt as a machine would, then check the CDN separately. Bot rules at Cloudflare, Akamai or Fastly run independently and return a 403 that robots.txt never reveals.

What 379,321 Claude citations show about its source mix

Claude cites brand-owned domains more than any other source type: 64.0 percent of 379,321 citations tracked across a month of SaaS and technology prompts, against 14.9 percent for news and media and 0.9 percent for social platforms. The figures come from an Otterly.ai study covering June 2026 across 16,406 unique domains.

Claude citation mix by source category, Otterly.ai, June 2026, n = 379,321 citations.

Source category	Share of Claude citations	What it means for a brand
Brand and vendor sites	64.0%	Your own domain is the biggest lever, unusually so
News and media	14.9%	Earned coverage matters, but it is a distant second
Blogs and personal sites	5.4%	Independent reviews and practitioner posts get cited
Other	3.7%	Tooling and aggregators
Education	2.7%	University and course pages
Government and NGO	2.4%	Regulator pages, on compliance questions
Unclassified	2.3%	Domains the study left uncategorised
Encyclopedia	2.1%	Wikipedia-class entity pages
Community and forum	1.3%	Forum threads rarely carry a Claude citation
Social media	0.9%	LinkedIn took 43.1% of it; Reddit zero times
Video	0.3%	The smallest category in the study

Source: Otterly.ai Claude citation study, June 2026, read September 23, 2026. The prompt set covers SaaS and technology topics, so the mix is a sector sample rather than a market-wide one.

Three numbers change how a Claude programme is scoped. The top ten domains take only 9.5 percent of all citations and 32.9 percent of cited domains appear exactly once, so Claude spreads citations across a long tail rather than a fixed authority set. Reddit was cited zero times in the whole sample, while LinkedIn took 43.1 percent of the small social share. And only 13 percent of cited domains overlapped between Claude and ChatGPT, falling to 4.2 percent at the URL level.

How to rank on Claude, step by step

Ranking on Claude takes seven steps in a fixed order: crawler access, Brave indexing, fan-out coverage, liftable sentences, brand fact pages, earned coverage and a repeated panel. Skipping the first two makes the other five invisible to Claude.

1. Allow Claude-SearchBot and Claude-User explicitly in robots.txt, and decide ClaudeBot separately on training grounds. Then request the page with each of the three user agent strings and confirm a 200 rather than a 403 from the CDN.
2. Check whether the page is in Brave’s index at all. Search Brave for site:yourdomain.com plus the buyer phrase. Brave publishes a submit-URL form at search.brave.com/submit-url but no webmaster console and no index-coverage report, so this check stays manual.
3. Write the fan-out queries yourself. Take one buyer question, list the five to eight sub-questions a researcher would search to answer it, and check which of your pages ranks in Brave for each. The gaps are the pages to write.
4. Put a liftable answer in the first 150 characters of the relevant section, because that is the size of the cited_text window Claude attaches to a citation. Canlah AI’s study of 526 citations across 60 buyer queries in ChatGPT and Google AI Mode found the copied sentence is almost always a paragraph’s first, and that about 73 percent of copied facts sit in the first 30 percent of the page.
5. Make the brand page the best source on its own facts. With 64.0 percent of Claude citations going to brand domains, a plain, current, crawlable page of pricing, scope and capabilities does more than a blog series does.
6. Earn coverage in news, trade media and practitioner blogs for the claims a vendor cannot self-verify. Those categories carry 20.3 percent of Claude citations combined.
7. Re-run the same panel on a fixed cadence and save the full answer, the cited URLs and the date. Claude’s search trigger is probabilistic, so an answer with no sources is a result to record, not a failed test.

Seven Claude visibility checks at a glance

A Claude visibility panel needs seven checks, from crawler access to repeat stability, and each one supports a single decision rather than feeding one blended score. Canlah AI uses the same seven-row shape for the engines it does probe and keeps the raw answer alongside every row.

Seven checks for a manual Claude visibility panel, Canlah AI method, September 2026.

Check	What to record	Decision it supports
Crawler access	HTTP status returned to Claude-SearchBot and Claude-User, at origin and at CDN	Whether the problem is access or content
Brave presence	Brave rank for each fan-out query, or absent	Whether to fix indexing or to fix ranking
Search trigger rate	How many panel answers ran a search at all	Whether the question set is commercial enough
Mention rate	Answers naming the brand, over answers returned, per question	Whether the brand is known to the category at all
Citation source	Which domain Claude cited for your category	Whether to fix your page or earn a third-party one
Accuracy	Whether the cited sentence describes you correctly	Whether the priority is correction or exposure
Repeat stability	Variation across three runs of one question in one day	How much of the movement is noise

Source: Canlah AI panel method, September 23, 2026. Record each check per question, per engine and per language, because a blended score across the seven hides which decision the data supports.

How to interpret five common results

Five results cover most outcomes in a Claude panel, and each points to a different fix: no sources, a directory citation, a ChatGPT-only presence, an unlinked mention and a wrong description.

No sources attached: Claude answered from stable knowledge and never searched, so this is a question-design result rather than a visibility result. Rewrite the question so it depends on current, changing or organisation-specific information, then re-run it.

Cited through a directory: Claude found the fact on a third-party page because your own page did not state it plainly. Put the same fact on your own domain in a single sentence, with the number, and re-test in two to four weeks once Brave has re-crawled.

Visible in ChatGPT, absent in Claude: Treat this as an index difference rather than a fault. Only about 13 percent of cited domains overlap between the two engines, so the fix is Brave coverage and page-level answerability, not another round of ChatGPT work. Canlah AI sees the same split between ChatGPT and Google AI Overviews, where 15 domains of 165 appeared in both source pools on September 7, 2026.

Named without a link: The brand entered the answer through prior knowledge or an unlinked passage, so there is no traceable evidence path. Record it, re-run it and do not attribute it to the last website change.

Cited but described wrongly: Prioritise correction over exposure. Align the site, the pricing page and the most-cited third-party pages on the same approved wording, then monitor the exact question that produced the error.

What ranking on Claude cannot prove

A Claude ranking check cannot prove four things: a stable share of voice, causation, a position inferred from Brave rank alone or a result that holds across Claude surfaces.

• It cannot prove a stable share of voice, because Claude decides per request whether to search and which queries to write.
• It cannot prove that a content change caused an answer to change without a controlled timeline and repeated observation.
• It cannot be verified from Brave rank alone. Brave’s top ten supplied 79.2 percent of cited URLs in one study, not 100 percent, and its author describes the overlap as correlation rather than proof of retrieval dependency.
• It cannot be generalised across surfaces, because Anthropic’s API alone ships three web search tool versions (web_search_20250305, web_search_20260209 and web_search_20260318) with different result handling.

Tools that report Claude visibility

Three of the five options reviewed here name Claude on their public pages: Ahrefs Brand Radar lists Claude on its $199 per month AI Visibility Index, Otterly.ai sells Claude as an add-on from $29 per month and Profound names Anthropic Claude on its custom-priced Enterprise plan. Canlah AI is not one of them. Pricing and coverage were read on September 23, 2026 and were not verified through paid accounts.

Five options for Claude visibility reporting, public pages reviewed September 23, 2026.

Tool	Best for	Claude coverage as publicly stated	Starting price	Main point to verify
Otterly.ai	Small teams that want Claude in a low-cost panel	Claude sold as an add-on, as are Google AI Mode and Gemini; ChatGPT, Google AI Overviews, Perplexity and Copilot are included	$29 per month Lite plan with 15 prompts, plus $29 per month for the Claude add-on	Whether 15 prompts cover a useful Claude panel
Ahrefs Brand Radar	SEO teams already inside Ahrefs	“All platforms (Claude available)” on the AI Visibility Index; “Claude consumes 8 checks per update”	$199 per month for the AI Visibility Index	Whether the eight-check multiplier keeps a Claude-heavy panel affordable
Profound	Enterprise answer-engine programmes	Anthropic Claude named on the Enterprise plan with ChatGPT, Perplexity, Google AI Mode, Gemini, Microsoft Copilot, DeepSeek, Google AI Overviews and Exa Search	Free trial without Claude; Enterprise price is custom	Which Claude surface is sampled, API or consumer
Peec AI	Agencies tracking many client brands	Not found on reviewed page; the pricing page names ChatGPT, Perplexity and Gemini	Not published on the pricing page	Current Claude status, confirmed with the vendor
Canlah AI	Buyers who want measurement plus the execution that follows	Not part of standard tracking; ChatGPT on Starter, Growth adds Google AI Overviews, Full adds Google AI Mode and Gemini	GEO service from $1,500 a month per AI platform; free AI visibility report within 48 hours	Claude is not part of standard tracking, so cover it separately

“Not found on reviewed page” means the vendor’s public materials did not name Claude when reviewed on September 23, 2026. It is not proof that the capability is absent, and buyers should confirm current scope with the vendor directly.

When to move from a manual Claude panel to ongoing measurement

A manual Claude panel of eight to twelve questions is enough while nothing commercial depends on the result. It stops being enough when Claude answers feed a real buying decision, when several teams publish facts that have to stay consistent or when an inaccurate answer creates regulatory risk.

Canlah AI is a Singapore-based SEO + GEO agency that measures and improves brand visibility inside ChatGPT, Gemini, Google AI Overviews and Google AI Mode, reporting AI visibility as re-verifiable ranges with timestamped evidence. Canlah AI locks the client’s buyer intents (4, 8 or 16 by tier) into the contract for 90 days, samples each at a depth calibrated per metric and reports citation share as a range rather than a single number.

The limitation is specific to this guide. Claude is not part of Canlah AI’s standard tracking: the September 7, 2026 self-audit of canlah.ai sampled OpenAI and Gemini at the API layer plus a Google AI Overviews leg, and recorded 100.0 percent brand mentions on 39 branded questions against 4.0 percent on 177 open questions, same site, same day. That branded-versus-open gap generalises to Claude; the engine coverage does not. A buyer who needs a Claude leg should pair the retainer with one of the tools above.

Do not choose a tool or a service from this page alone. Run your own eight-question Claude panel first, save which answers searched and which URLs were cited, then ask each vendor above to show a dated Claude answer for one of your own questions before signing.

Frequently asked questions

Does Claude use Google or Brave for web search?

Brave. Anthropic added Brave Search to its subprocessor list on March 19, 2025, and the entry still names Brave Software for web search across all products when read on September 23, 2026. Two studies put the overlap between Claude’s citations and Brave’s top results at 79.2 percent and 86.7 percent.

How do I rank on Claude?

Rank on Claude by allowing Claude-SearchBot and Claude-User in robots.txt, earning a Brave top-ten position for the sub-queries Claude writes and stating each fact in one liftable sentence on your own domain. When Claude cites a competitor instead, the usual cause is Brave rank on those sub-queries rather than on the buyer question itself. Brand domains took 64.0 percent of Claude citations in Otterly.ai’s June 2026 study, so a missing plain-language fact page is a frequent gap.

Can I submit my site to Claude so it gets indexed?

No. Anthropic operates no submission form or webmaster console for Claude. The nearest equivalent is Brave’s submit-URL page, which has no verification step and no index-coverage report, so verify indexing by searching Brave rather than by reading a dashboard.

Will blocking ClaudeBot stop Claude from citing my site?

No. Blocking ClaudeBot restricts future training use only, because Claude’s search index is fed by Claude-SearchBot and live retrieval runs through Claude-User. Anthropic documents all three as separate agents with separate robots.txt rules. A User-agent: * disallow written as a training opt-out blocks all three, and that version does remove a site from Claude search.

Is Claude SEO different from ChatGPT SEO?

Yes, more than the shared advice suggests. Only about 13 percent of cited domains and 4.2 percent of cited URLs overlap between the two engines in the June 2026 Otterly.ai comparison, Claude cited Reddit zero times there, and the index is Brave rather than OpenAI’s own retrieval stack. The page-level craft transfers; the source strategy does not.

Does Canlah AI track Claude?

No. Claude is not part of Canlah AI’s standard tracking, which covers ChatGPT, Gemini, Google AI Overviews and Google AI Mode by tier (Starter covers ChatGPT, Growth adds Google AI Overviews, Full adds Google AI Mode and Gemini) on a locked pool of buyer intents, so the Claude figures in this guide are third-party. Buyers who need a Claude leg can use a tool add-on such as Otterly.ai’s Claude add-on, which starts at $29 per month on the Lite plan.

Method and sources

This guide was written on September 23, 2026. Crawler behaviour and search-tool mechanics were taken from Anthropic’s documentation and trust portal. The Brave-overlap figures are third-party findings rather than confirmed Anthropic behaviour, and the citation-mix figures come from one vendor study of a SaaS and technology prompt set, not a market-wide sample. The five tool rows come from each vendor’s public pricing or product page. No paid accounts were used, and no Claude probe was run for this article.

• Anthropic crawler documentation. ClaudeBot, Claude-User and Claude-SearchBot roles, robots.txt handling and Crawl-delay support.
• Anthropic web search tool documentation. When Claude searches, repeated searches within one request, the three tool versions, always-on citations and the 150-character cited_text window.
• Anthropic subprocessors. Brave Search listed for web search across all products.
• Josh Blyskal, The state of AEO in 2026, July 22, 2026. Roughly 35,000 URLs across 400 queries; 79.2 percent of Claude’s cited URLs also in Brave’s top ten.
• Profound, Claude web search explained. The 86.7 percent result overlap with Brave.
• Otterly.ai Claude citation study, June 2026. 379,321 citations, 16,406 domains, the category mix, the 9.5 percent top-ten share, zero Reddit citations and the ChatGPT overlap.
• Otterly.ai pricing. Lite plan at $29 per month with 15 prompts and the Claude add-on price.
• Ahrefs Brand Radar. AI Visibility Index at $199 per month, Claude availability and the eight-check multiplier.
• Profound pricing. Trial and Enterprise engine lists, including Anthropic Claude on Enterprise.
• Peec AI pricing. Engines named on the pricing page; Claude not found when reviewed.
• canlah.ai/ai-info. Locked query pool, sampling calibrated per metric and what Canlah AI declines to guarantee.
• Canlah AI self-audit evidence archive, anonymised, September 7, 2026. Branded and open-question mention counts on canlah.ai; OpenAI and Gemini API layers plus a Google AI Overviews leg, with no Claude leg.
• Canlah AI citation study, September 2026. Sixty buyer queries in ChatGPT and Google AI Mode, 526 citations and 92 cited pages; source of the first-sentence finding.

See where your brand stands in AI answers. Run the free AI visibility audit →

Related articles

• How to rank on ChatGPT in search results
• How to get cited by AI search engines
• How to rank in AI search engines
• What is llms.txt, and why it isn’t a citation lever