Claude SEO: How to Get Cited and Recommended by Claude

Claude SEO means getting Claude to name your brand and link your pages when someone asks a buying question. It is a form of answer engine optimization (AEO), also called generative engine optimization. You cannot rank in Claude the way you rank on Google: when it searches the web it cites the sources it used, so you win by being reachable, being the clearest source and being confirmed by other sites.

Does Claude cite sources?

Yes, whenever it searches the web. Anthropic's help center says every response that uses web search includes citations, and the API documentation says citations are always enabled for the search tool. An answer written from training alone shows no links.

What that means in practice:

• Claude decides when to search. The API docs say it searches for current or changing information, including organizations and products that might have changed, and answers directly for stable knowledge. In the Claude app, the help center says new-experience accounts have no search toggle: Claude searches when it helps.
• A citation is specific. Each one carries the page URL, its title and up to 150 characters of the cited text.
• You can't see your buyers' answers. The app shows a search indicator only to the person asking. Plan for both: crawlable pages for searched answers, independent coverage for answers from memory.
• Mention and citation are separate. Claude can name you without linking you, or link your page while recommending a rival. Track both.

Any source named without a visible search is unverified; open it before you trust it.

How does Claude find and choose its sources?

Anthropic documents the loop, not a ranking formula. Claude decides to search, the search tool returns results, and that can repeat several times before Claude writes a cited answer. Anyone selling "Claude ranking factors" is guessing.

1. Claude writes one or more queries. Per the API docs, simple factual questions typically use 1 to 3 searches, while comparative or multi-entity research can use 10 or more. "X vs Y" and "best X for Y" questions fan out, so you need pages for the narrow sub-questions (pricing, alternatives, who it suits), not one page for the head term.
2. The tool returns results. Each carries a URL, a title and a page_age field the docs describe as when the site was last updated.
3. Claude reads and cites the pages it drew on.

Which search index sits behind this? The pages we read do not say. TechCrunch reported in March 2025 that Anthropic had added Brave Search to its subprocessor list and that a developer found a Brave parameter in Claude's search function; Anthropic did not confirm. Our reading: treat Brave as a likely feed, not a certain one, and check Brave Search as well as Google.

Whose pages get cited? Often not yours. A September 2025 arXiv preprint ran 1,000 consumer-ranking prompts through search-enabled engines, Claude with Anthropic's web search tool among them. In its U.S. software test, AI answers drew 72.7% of sources from earned media (independent publishers, review sites) and 26.7% from brand sites, against 45.4% and 43.7% for Google.

Which Anthropic crawlers matter, and which can you block safely?

Anthropic runs three bots with three jobs, each with its own robots.txt line. Allow Claude-SearchBot and Claude-User if you want to be found. ClaudeBot is the training crawler and a separate decision. This table follows Anthropic's crawler page (dated April 2026):

Bot	Job, per Anthropic	If you block it	Blocked by (our study, n = 384)
ClaudeBot	Collects content that may contribute to training	Future material is excluded from training datasets	44 (11.5%)
Claude-SearchBot	Improves the quality of search results	Content is not indexed for search; may reduce visibility and accuracy	24 (6.2%)
Claude-User	Fetches pages when a Claude user asks	Content is not retrieved for user queries; may reduce visibility	26 (6.8%)

In AEOeye's crawler access study (October 2026, 384 well-known sites), 0 of 193 SaaS sites and 2 of 90 e-commerce and consumer brands blocked Claude-SearchBot, against 19 of 38 news publishers. None of the 43 AI-visibility vendors blocked any Claude bot. Of the 44 sites blocking ClaudeBot, 20 still allowed Claude-SearchBot. See the per-site data and method.

Limits: one fetch per site on one day, main host only, convenience sample. Under RFC 9309, robots.txt rules are not access authorization.

Our position: a brand should allow all three. A publisher with licensed content can block ClaudeBot alone; the search bot has its own rule. To allow everything explicitly:

User-agent: ClaudeBot
User-agent: Claude-SearchBot
User-agent: Claude-User
Allow: /

Two traps. Anthropic says blocking by IP address may not work, because it stops the bots reading your robots.txt, and rules apply per subdomain. Firewalls: 24 of 88 e-commerce and consumer homepages (27.3%) blocked or challenged our self-identified crawler, so look for Claude-SearchBot and Claude-User in your CDN logs. Our ClaudeBot explainer covers verification.

How do you get recommended by Claude?

Be reachable, be the clearest answer, then be confirmed by others. Each step fails silently if the one before it is broken, so work in this order.

1. Allow the three bots in robots.txt and your firewall, as above.
2. Put the answer in the raw HTML. Vercel's December 2024 analysis found Anthropic's crawler fetched JavaScript files in 23.84% of requests but did not execute them. That covers the training crawler, and Anthropic has not published how its search bot renders pages, so check view-source for your key claims.
3. Confirm you are indexed. Search your brand and money pages on Google and Brave Search.
4. Write one page per buyer sub-question. Open with the answer, then add facts a model can quote: numbers, named sources, direct quotations. In the GEO benchmark, citing sources, quotations and statistics lifted visibility by 30 to 40% on one metric, while keyword stuffing gave little to none. It used a GPT-3.5-based engine and Perplexity, not Claude, so read it as a direction.
5. Earn mentions where Claude looks. Ask your buyer questions with search on, open the citations and list the domains: roundups, review sites, comparison pages. Earn real coverage there; our guide to getting recommended by AI covers the wider playbook.
6. Keep facts current and identical. Show an updated date and use the same name and prices everywhere.

None of the Anthropic pages we read offers a submission form or paid placement. We leave llms.txt off the list too: 173 of the 330 sites we could check publish one, yet Anthropic's crawler and search documentation never mention it.

How do you check whether Claude mentions you?

Ask Claude your buyers' questions with search on, repeatedly, and log how often it names and links you. One answer proves nothing. SparkToro's study (600 volunteers, 2,961 runs across ChatGPT, Claude and Google AI, published January 2026) found under a 1-in-100 chance that ChatGPT or Google's AI repeats the same brand list. Claude was slightly more consistent on lists, still unlikely to repeat an order.

1. Write 10 to 20 brand-free buyer questions: "best X for Y", "X vs Y", "alternatives to X".
2. Run each at least five times in fresh chats (a quick-check floor; SparkToro suggests 60 to 100 runs for a stable average), adding "search the web" to the prompt, which Anthropic says makes Claude search.
3. Log four things: named, linked, which competitors appeared, which domains were cited.
4. Report percentages, not ranks. SparkToro calls visibility percentage reasonable and rank positions unreliable.

Developers can also test through the API's search tool ($10 per 1,000 searches plus tokens, as of October 2026), but that differs from the app your buyers use, so treat it as a supplement.

AEOeye's free audit asks ChatGPT three unbranded buyer questions; the $29 full report adds Claude, Gemini, Google AI and Perplexity across 10 questions, each engine at the default tier a free user gets. In our audit records (127 audits, June to September 2026), Claude named the audited brand in 143 of 509 answers (28.1%): a small, self-selected sample, not a market rate and not comparable with other engines. For hundreds of prompts on a schedule, use a dedicated tracker.

Can Claude help with your own SEO work?

Yes, for structure and speed, not for facts. It handles long inputs and code well; it does not know your analytics.

• Long documents. Paste a content inventory or crawl export and ask it to group pages with duplicate intent.
• Code. Redirect maps, JSON-LD, log-parsing regex. Claude Code reads a codebase, edits files and runs commands for batch fixes.

Never ask it to estimate volume or traffic. Anthropic's own help page says to cross-reference cited sources, and Google's guidance says to fact-check AI output and warns that mass-produced pages without added value can break its scaled content abuse policy. The prompts in our ChatGPT SEO guide transfer to Claude with little change.