How to Get Cited by Claude: Rank in Claude AI Answers (2026)

Claude now searches the web on every plan, and when it does, it shows the pages it relied on. That makes Claude citations a real channel for B2B and SaaS brands, especially for the comparison and pricing questions buyers ask late in their research. The catch is that Claude works differently from Google, and Anthropic documents only part of how it picks sources.

This guide separates what is documented from what is inferred, then gives you a concrete process: open the right crawler access, get into the index Claude searches, write passages Claude can quote, and measure the results with a repeatable test.

How does Claude decide which sources to cite?#

Claude cites a web page only when it decides to run a web search. If it answers from training data, there is no citation at all, even if your brand is mentioned.

Anthropic's web search tool documentation describes the loop. Claude decides whether to search based on the prompt, the search runs, results come back, and the process can repeat several times before Claude writes a final answer with cited sources. The same page lists what tends to trigger a search: recent events, current prices and statistics, explicit requests to look something up, and "information about specific organizations, people, or products that might have changed." Stable knowledge, such as math or established facts, is usually answered without searching.

For brands, that last trigger is the important one. Questions like "best SOC 2 tool for a 20-person startup", "X vs Y pricing", or "does X integrate with HubSpot" involve products that change, so they are the questions most likely to produce citations.

The docs also explain how much searching happens. Simple factual queries typically use one to three searches, while comparative or multi-entity research can use ten or more. A comparison prompt gives Claude more chances to find your page, but also more competing pages.

What Anthropic does not publish is how Claude ranks or chooses among the results it retrieves. Any claim that Claude "weights" freshness, authority, or schema a certain way is an inference, not documentation. The practical approach is to control what you can document and measure the rest.

Two ways your brand appears in Claude

Which Anthropic crawlers should you allow?#

Allow Claude-SearchBot and Claude-User if you want to be cited. ClaudeBot is a separate decision about model training.

Anthropic's help article on how it crawls the web describes three bots, all of which honor robots.txt:

Three details from the same article matter in practice:

1. Rules are per subdomain. You need a robots.txt rule on each subdomain you want to control, so check docs., blog., and help. subdomains, not just your root domain.
2. Crawl-delay is supported. Anthropic supports the non-standard Crawl-delay directive if crawl load is a concern. Use it instead of blocking.
3. Do not block by IP. Anthropic says IP blocking may not work reliably and can stop its bots from reading your robots.txt at all.

A common, consistent setup for a brand that wants citations but not training use looks like this:

User-agent: ClaudeBot
Disallow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

Also check your CDN or bot-protection settings. Many firewall products have one-click "block AI bots" rules that can block Claude-User and Claude-SearchBot along with training crawlers. Your robots.txt can say "allow" while your firewall says no.

Does Claude use Brave Search?#

Probably, at least in part, but Anthropic has not documented its current search provider in its public product docs.

Here is what is verifiable. In March 2025, Anthropic added Brave Search to its subprocessor list, as reported by TechCrunch. Developer Simon Willison then ran a test and found that the ten citations Claude returned matched Brave's results for the same query exactly, and that Claude's web search function definition contained a parameter named BraveSearchParams. Anthropic did not comment at the time.

That evidence is from 2025. Search providers can change, and Anthropic is not obliged to announce it. So use Brave as a working proxy, and keep testing:

• Search Brave (search.brave.com) for your core category queries and see whether your pages appear.
• Run site:yourdomain.com on Brave to see roughly how much of your site it has indexed.
• Compare: when Claude cites a page for one of your test prompts, check whether that page also ranks on Brave for a similar query. If the overlap holds, the proxy still works for your category.

If your key pages are missing from Brave but present in Google, look at the basics first: server-rendered HTML for main content, no accidental noindex, working internal links, and an XML sitemap.

What makes a page easy for Claude to cite?#

Claude cites passages, not whole pages, so each section should contain a short, self-contained answer that still makes sense when quoted alone.

The API documentation shows exactly what a citation contains: the URL, the page title, and up to 150 characters of cited_text. That is roughly one sentence. If the sentence that answers the question is buried in paragraph four, or only makes sense with the paragraph before it, it is a weak candidate.

Practical rules that follow from that:

• Lead each section with the answer. Put a one or two sentence answer under every H2, then the detail. "Plan X costs $49 per seat per month, billed annually, as of October 2026" is quotable. "Pricing depends on many factors" is not.
• Name entities in full. Write "Acme CRM integrates with HubSpot and Salesforce" rather than "it integrates with both." A quoted sentence loses its surrounding context.
• Date-stamp anything that changes. The search results Claude receives include a page_age field, so the date your page was updated is visible to the system. Keep a visible "updated" date and actually update the content when facts change.
• Use tables for comparisons. Prices, limits, and feature support in a table are easier to extract accurately than the same facts spread through prose.
• Publish pages for the questions buyers ask. Comparison pages, "alternatives to" pages, integration pages, and pricing pages map directly to the queries most likely to trigger a search.

None of these rules is specific to Claude, which is the point. Pages built this way perform across engines. Our answer engine optimization guide covers the full framework, and the GEO checklist turns it into an audit.

How does Reddit fit into Claude visibility?#

Reddit matters to Claude in the same way any widely discussed third-party source matters: when Claude searches, Reddit threads often appear among the results for product questions.

Two facts keep this in proportion. First, there is no documented content partnership between Reddit and Anthropic. Second, Reddit sued Anthropic in June 2025, alleging that Anthropic's bots kept accessing Reddit after being asked to stop and that it trained Claude on Reddit data without permission. Anthropic said it disagreed and would defend itself. Whatever the outcome, it means you should not assume Reddit is either heavily or lightly represented in Claude's training data. Test your own prompts.

What you can act on is the live-search side. If Claude's sources for "best tool for X" include a Reddit thread, that thread is shaping the answer. The ethical way to be part of it:

1. Find the threads that show up as sources for your category prompts.
2. Read each subreddit's rules before you post. Many restrict vendor replies or require flair.
3. Reply from your real account, disclose your role ("I work at X"), and answer the question, including when a competitor is the better fit.
4. Never use multiple accounts, bought accounts, or vote manipulation. These violate the Reddit Rules and get accounts and domains banned.

Our guide to Reddit self-promotion rules covers what subreddits typically allow, and Reddit AEO explains how threads end up in AI answers.

How do you check whether Claude cites you?#

Build a fixed set of prompts, run them in Claude with web search on, and log every source. Repeat on a schedule, because answers vary between runs.

Step 1: Write a prompt panel

Write 20 to 40 prompts that a buyer in your category would actually type. Mix four types:

• Category: "What are the best tools for [job] for a [company size]?"
• Comparison: "[Your brand] vs [competitor] for [use case]"
• Problem: "How do I [solve the problem your product solves]?"
• Brand: "Is [your brand] good for [use case]? What are the downsides?"

Step 2: Run the prompts in the Claude app

Turn on web search (Anthropic's help article explains where the toggle is; on Team and Enterprise plans an admin must enable it first). For each prompt, record:

• Whether a search ran at all
• Whether your brand is mentioned, and in what position
• Which URLs were cited, and whether any are yours
• Which competitors and which third-party pages (reviews, Reddit, directories) were cited

Step 3: Automate with the API for scale

The web search tool returns structured citations, so a short script can run your panel and save each cited URL along with the exact cited_text. That tells you not just that you were cited, but which sentence Claude used. As of October 2026, Anthropic prices web search at $10 per 1,000 searches plus normal token costs. API results come from the same model family but may differ from what the consumer app shows, so spot-check against the app.

Step 4: Check your server logs

Filter your access logs for the Claude-User and Claude-SearchBot user agents. Claude-User hits on a specific URL are a direct sign that someone's conversation needed that page. If you see no Claude-SearchBot activity at all, check your firewall.

Step 5: Track share of voice over time

A single run is noise. Run the panel weekly, and track the percentage of prompts where you are mentioned and cited, compared with your top competitors. Our guide to AI share of voice explains how to calculate it.

This is the work Sleeng automates: it generates buyer-intent questions from your URL, runs them on Claude along with ChatGPT, Perplexity, Gemini, Google AI Overviews, and AI Mode, logs every mention and citation, and shows your share of voice against competitors. It also flags the Reddit threads those engines cite, so you know which conversations are worth a disclosed reply.

What mistakes keep brands out of Claude answers?#

Most brands that never appear in Claude's citations have a fixable access or format problem, not a reputation problem.

• Blocking all Anthropic bots by default. A blanket block, often added in 2023 or 2024 to stop training, also blocks Claude-SearchBot and Claude-User. Review robots.txt and firewall rules together.
• Assuming Google rankings carry over. Claude's search provider is not Google. A page ranking third on Google may be missing from Brave.
• Hiding facts in scripts or images. Prices rendered by client-side JavaScript, or comparison charts saved as images, may not be readable by crawlers.
• Stale pages. A pricing page that still shows last year's plans can be quoted accurately and still hurt you.
• Measuring once. Claude's answers change between runs and between model versions. Track trends, not snapshots.

A quick Claude visibility checklist#

• robots.txt allows Claude-SearchBot and Claude-User on every subdomain
• CDN and firewall rules do not block those user agents
• Key pages appear in Brave for your main category queries
• Main content is server-rendered HTML with no accidental noindex
• Each H2 opens with a one or two sentence answer that names the entity
• Prices, limits, and features are in tables with an "as of" date
• Comparison, alternatives, integration, and pricing pages exist
• A prompt panel of 20 to 40 prompts runs weekly with logged citations
• Reddit threads cited for your category are tracked, and any replies are disclosed

If you want this running without a spreadsheet, Sleeng tracks your Claude mentions and citations alongside five other AI engines, and shows which Reddit threads they cite. Start a 14-day free trial and see your first share-of-voice report from your own URL.

Sources#

• Anthropic, Web search tool documentation
• Anthropic, Does Anthropic crawl data from the web, and how can site owners block the crawler?
• Anthropic, Enabling and using web search
• Anthropic, Claude can now search the web (March 20, 2025, updated May 27, 2025)
• TechCrunch, Anthropic appears to be using Brave to power web searches for its Claude chatbot (March 21, 2025)
• Simon Willison, Anthropic use Brave (March 21, 2025)
• ABC7 News (AP), Reddit sues AI company Anthropic for allegedly scraping user comments (June 4, 2025)
• Reddit, Reddit Rules
• Brave, Brave Search