To get mentioned in Claude, you need to work on two layers at once. When Claude answers without searching, it relies on what the model learned during training, so your brand needs a clear, consistent presence across the public web over time. When Claude searches the web, it reads live pages and cites them, so your pages need to be reachable by Anthropic’s search and user crawlers and written so that a specific passage can support a specific claim. Most brands that show up well in Claude have done both.
This guide explains how Claude answers with and without web search, which crawlers matter, what Claude tends to cite, and the steps that make your brand easier to mention and cite. If you are new to the field, our guide to LLMO explains how language models form a picture of a brand.
Key takeaways
- Two layers: Claude’s trained knowledge changes only with new models, while its web search results can change as soon as your pages are crawled and indexed.
- Three crawlers: ClaudeBot collects training data, Claude-SearchBot indexes pages for search, and Claude-User fetches pages when a person’s question needs them. Each is controlled separately in robots.txt.
- Citations come with search: when Claude searches, Anthropic says every response includes citations, so you can see exactly which pages it used.
- Quotable beats promotional: Claude often supports its claims with short quotes, so pages with specific, self-contained statements are easier to use than marketing copy.
- Mentions are not only citations: Claude can name your brand from its own knowledge with no link at all, which is why consistency across third-party sources matters.
How Claude answers a question
Claude is Anthropic’s AI assistant, available on the web, on desktop and mobile apps, and through an API that many other products build on. How it answers depends on whether it uses web search for that question.
Without web search: trained knowledge
When Claude answers from its own knowledge, it draws on what the model learned during training. Anthropic’s model documentation describes training data that includes public web content along with other sources, and each model has a training cutoff. Anything that happened after the cutoff, such as a new product, a price change or a rebrand, is unknown to the model unless it searches.
Trained knowledge has three traits that matter for brands. It is shaped by repetition, so a brand described the same way across many independent sources is represented more clearly than one mentioned rarely or inconsistently. It can be out of date, so Claude may describe an old version of your product. And it changes only when Anthropic releases a new model, which you cannot schedule or influence directly.
With web search: live pages and citations
Claude can also search the web. Web search is available on Claude plans worldwide, and in current versions of the apps Claude decides for itself when a search would help, for example when a question is about recent events, prices or specific products. Users can also ask it to search, or ask it not to. On Team and Enterprise plans, an administrator has to enable web search for the organization first.
When Claude does search, Anthropic says every response includes citations, with links to the sources and relevant quotes where appropriate. Claude can also fetch a specific page when a user pastes a URL into the conversation. Image results, when shown, carry their own source links.
Anthropic has not published how Claude ranks or selects the pages it reads and cites. What you can see is the output: which pages were cited for your prompts, how often and next to which competitors.
Research and connected tools
On paid plans, Claude also offers Research, a mode that runs several searches that build on each other and works through a question from different angles before answering. Research can combine web results with connected sources such as Gmail, Google Calendar, Google Docs and other connectors the user has turned on. For complex buying questions, this means Claude may read many more pages than in a quick answer, which gives well-structured, specific pages more chances to be used.
Anthropic’s crawlers and what they control
Anthropic documents three robots that site owners can manage in robots.txt, and each one affects a different part of how Claude can know about you:
- ClaudeBot collects public web content that may be used to train future models. Blocking it signals that your future content should be left out of training data.
- Claude-SearchBot indexes content to improve the quality of Claude’s search results. Anthropic warns that blocking it may reduce your visibility and accuracy in search answers.
- Claude-User fetches pages when a person asks Claude a question that needs them. Blocking it stops Claude from retrieving your pages for those users, which can reduce your visibility in user-directed search.
Anthropic says these robots honor robots.txt, and it supports the non-standard Crawl-delay rule if you need to slow crawling down. It advises against blocking its bots by IP address, because that may not work reliably and can stop the bots from reading your robots.txt at all. It also publishes the IP addresses its crawlers use, so you can check that traffic claiming to be Anthropic is genuine.
Separately, browsers that Anthropic hosts to act on a user’s behalf identify themselves with their own token, Claude-Agent, which you can manage independently of the crawlers. And from late October 2026, Anthropic says it also recognizes a machine-readable “no AI training” preference based on a draft standard, with rules that name ClaudeBot directly taking priority.
If you want to stay out of future training data but remain available for search and user requests, a robots.txt like this expresses that choice:
User-agent: ClaudeBot
Disallow: /
User-agent: Claude-SearchBot
Allow: /
User-agent: Claude-User
Allow: /
Think about the trade-off before you copy it. Blocking ClaudeBot does not affect search, but over time it means future models learn about you only from what other sites say. If you want Claude to know your brand when it answers without searching, allowing ClaudeBot is the more direct route. Our guide to which AI crawlers to allow walks through the same decision for every major engine.
Tip: Check your CDN and firewall as well as robots.txt. Some security services offer one-click AI bot blocking that catches Claude-SearchBot and Claude-User along with training crawlers.
What Claude tends to cite
Anthropic has not published a list of preferred sources, and patterns differ by topic, so the most reliable picture comes from your own prompts. Still, how Claude works tells you a lot about what makes a page useful to it.
Pages that support a specific claim
Claude often backs a statement with a short quote from the page it cites. A page full of general claims gives it little to quote. A page that says plainly what a product does, what it costs, who it suits and what its limits are gives Claude a passage it can use to support a sentence in its answer.
Primary and reference sources
For factual questions, you will usually see official documentation, product pages, standards bodies, reference sites and established publications among the citations. If the question is about your product, your own pricing page, documentation and changelog are the primary sources. When those pages are thin, outdated or hard to read, Claude has to rely on someone else’s description of you.
Third-party opinion and discussion
For “best” and comparison questions, Claude tends to read what others say: reviews, comparison articles, forums and community discussions. A brand that only describes itself on its own site is easy to leave out of a recommendation. One that is discussed accurately by independent sources is easier to include.
Balanced answers
Anthropic publishes the principles it trains Claude to follow, and they put weight on honesty and calibrated, balanced answers. In practice, Claude often lays out trade-offs instead of naming a single winner. Pages that are honest about who your product is for, and who it is not for, fit that style better than pages that claim to be best at everything.
How to earn mentions in Claude’s answers
The steps below cover both layers. The first three help Claude’s search find and cite you. The rest build the consistent presence that shapes trained knowledge and recommendations.
Step 1: Open access to search and user crawlers
Open your live robots.txt and check how it treats Claude-SearchBot and Claude-User. If no group names them, your User-agent: * rules apply, and a broad disallow there blocks them too. Then check your CDN, firewall and security plugins for AI bot blocking, and confirm your important pages return their full content in the HTML rather than only through client-side scripts. The free robots.txt tester shows how a given bot is treated, and the AI bot user agent checker helps you identify Anthropic’s bots in your logs.
Step 2: Make your brand a clear, consistent entity
Use one brand name, one plain description of what you do and who it is for, and the same core facts everywhere: your site, your social profiles, directories, review sites and partner pages. Publish an about page that states those facts clearly, and add Organization structured data that links your site to your official profiles. The free organization schema generator builds that markup in a few minutes.
Consistency matters most for trained knowledge. If half the web describes you as a project management tool and the other half as a CRM, a model has no clear picture to draw on.
Step 3: Publish pages that answer one question well
When Claude searches, it looks for the passage that answers the question in front of it. Give it one.
- Answer first. Put a direct two or three sentence answer under each key heading, then add detail and exceptions.
- Name the subject. Write “Acme Notes syncs with Gmail and Outlook” rather than “it syncs with your email”, so a quoted sentence makes sense on its own.
- Be specific. Use numbers, prices, limits, dates and conditions, and keep them current.
- State the trade-offs. Say who the product suits and where it falls short. Honest pages are easier to cite in a balanced answer.
- Show your sources. Link to primary sources for claims you did not originate, and show who wrote the page.
Run drafts through the free answer readiness checker to catch buried answers and vague openings before you publish.
Step 4: Keep your primary sources current
Claude may be asked about your pricing, features, integrations or policies, and those questions often trigger a search. Make sure the pages that hold those facts are easy to find, consistent with each other and dated honestly. Remove or redirect pages for discontinued products and old pricing, because a search can surface them just as easily as the current version.
Step 5: Earn accurate coverage on third-party sources
Look at the domains Claude cites for your most important prompts. If they are review sites, comparison articles, forums or industry publications, that is where you need to be described accurately. Claim and complete your profiles on review and directory sites, offer expertise and data to publications that cover your category, and take part honestly in the communities where your buyers ask questions. Our guide to Reddit and forums in AI answers covers how to do that without spamming.
Step 6: Correct wrong claims at the source
If Claude describes your brand incorrectly, find out where the claim comes from. When Claude searched, the citations show you the page. When it answered from its own knowledge, look for the same claim on prominent pages about you, such as old reviews, outdated directory listings or your own archived content. Fix what you control, ask the owners of other pages for corrections, and publish a clear, current page that states the right fact. The fix wrong AI claims playbook lays out the full process.
How to measure your visibility in Claude
Claude’s answers vary between runs, and they differ depending on whether it searched. Measure with that in mind.
- Mentions: ask your prompt list on a schedule and record whether Claude names your brand, in what position and how it describes you. Our guide to brand mentions vs citations explains why you should track the two separately.
- Citations: when Claude searches, record which of your pages are cited, which competitor pages appear and which third-party domains keep coming up.
- Crawls: check your server logs for ClaudeBot, Claude-SearchBot and Claude-User, and verify them against Anthropic’s published IP addresses. A page that is crawled often but never cited needs a content review. A page that is never crawled needs an access check.
- Visits: people who click a citation in the Claude apps often arrive with a claude.ai referrer, though some apps and settings strip it. Segment those visits in analytics to see which pages they land on.
Judge trends across weeks and many prompts, not single answers. A model release can shift trained knowledge overnight, so note release dates next to your data.
Common mistakes
- Blocking every AI bot at the CDN, then wondering why Claude never cites your site.
- Blocking ClaudeBot without realizing it affects how future models learn about you, or allowing it and assuming it helps search. The two are separate.
- Describing your brand differently on every profile, so models never form a clear picture.
- Writing only promotional copy, which gives Claude nothing specific to quote in a balanced answer.
- Leaving old pricing and discontinued features online, where a search can find and repeat them.
- Measuring from one answer instead of repeated runs across many prompts.
Frequently asked questions
Does Claude search the web for every question?
No. Claude decides when a search would help, typically for current, specific or fast-changing topics, and answers simpler questions from its own knowledge. Users can ask it to search or not to search, and organizations on Team and Enterprise plans must enable web search first.
Will blocking ClaudeBot remove me from Claude’s answers?
Not from search answers. ClaudeBot relates to training data, while Claude-SearchBot and Claude-User handle search indexing and user requests. Blocking ClaudeBot means future models may learn about you mainly from other sites, which can affect answers given without a search.
How long does it take for Claude to learn about a new product?
For search answers, it can be soon after your pages are crawled and indexed. For answers from trained knowledge, it depends on when Anthropic trains and releases new models, and on how widely the product has been written about before that. There is no way to request inclusion.
Can I pay to be mentioned by Claude?
There is no public way to pay for a mention or citation in Claude’s answers. Mentions come from what the model knows and from the pages it finds when it searches.
Why does Claude mention my competitor and not me?
Compare how often and how consistently each brand is described on the sources Claude cites for that prompt. Common reasons are stronger third-party coverage, clearer positioning, more specific pages, or more recent content. Fix access first, then close the gap on substance.
How aeoora helps
aeoora asks your tracked prompts in Claude every day, along with ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode, and records whether you are mentioned, your position, how you are described and which pages are cited, for you and for your competitors. Mentions come with sentiment and perception terms, so you can see what Claude associates with your brand. On the Signal and Scale plans, AI crawler analytics verifies visits from Anthropic’s crawlers in your server logs, so you can see which pages they fetch, and a cookie-free snippet measures the visits Claude sends to your site.
Claude is included on every plan: Shine at $29 a month, Signal at $69 and Scale at $299, with 25% off yearly and a 2-day free trial.
Next steps
- Get the free AI visibility report for a quick outside-in look at how AI engines describe your brand.
- Read the mentions and sentiment docs and the AI crawler analytics docs to see how aeoora tracks Claude.
- Check what each plan includes on the pricing page, then start your 2-day free trial and add your first prompts with the quickstart.


