Claude SEO: how to get your brand into Claude’s answers

Buyers do not only ask ChatGPT and Perplexity for recommendations. Some ask Claude, Anthropic's assistant, to compare tools, summarise options and sanity-check a shortlist. Claude SEO is the work of making sure your site can be reached and quoted when that happens.

This guide covers what you can verify and control: the three crawlers Anthropic documents, the content that gives Claude something to quote, and a way to measure whether your brand actually appears. It complements our broader guide to AI SEO, which covers the concept across assistants.

What Claude SEO means

Claude SEO shifts the workflow focus from positions on a results page to presence in an assistant's answers. Two separate things matter in any answer: whether your brand is mentioned at all, and whether your pages are cited as sources. Record both, because an assistant can name a brand without citing its site, and cite a page without recommending the brand.

The crawler facts below come from Anthropic's own help centre, labelled with a link and date. The content and measurement suggestions are editorial recommendations to test, not documented engine behaviour.

One more difference from classic search is the shape of the visit. A buying conversation can be modelled as more than one question: options, comparison, objections about price or trust, then a decision. Presence in the first answer is a starting point, not the outcome, so it is worth testing your brand across those stages rather than checking a single opening answer.

The three Anthropic crawlers

Anthropic documents three robots in its crawler help article (as of November 2026), and they split along the same lines as other providers: training, search indexing, and user-triggered fetches.

ClaudeBot collects web content that could contribute to training Anthropic's models. Anthropic states that when a site restricts ClaudeBot, it signals that the site's future materials should be excluded from training datasets.

Claude-SearchBot indexes content for search. Anthropic states that disabling it prevents the system from indexing your content for search, which may reduce your site's visibility and accuracy in user search results.

Claude-User handles fetches triggered when a person asks Claude a question. Anthropic states that site owners can control which sites Claude-User can access, and that disabling it prevents retrieval in response to user queries, which may reduce visibility for user-directed web search.

Deciding what to allow

Because the three robots are documented separately, you can state separate preferences. Allowing Claude-SearchBot and Claude-User while disallowing ClaudeBot is a coherent, documented position: it keeps your pages available for search indexing and user-triggered retrieval while signalling that your content should not be used for training. Whether that trade-off suits your organisation is a policy decision.

Two cautions apply. First, allowing a crawler makes your pages available to the corresponding feature; it does not compel citation or recommendation. Second, a robots.txt rule is a stated preference rather than an absolute access control, so check what your firewall or WAF does with unfamiliar agents as well.

Before changing anything, record your starting point: the current robots.txt entries for ClaudeBot, Claude-SearchBot and Claude-User, and any firewall rules that mention them. That way, if you do adjust the rules, you can compare answers before and after instead of guessing whether anything changed. If you are unsure what your site currently allows, a crawler checker will show you.

The content work on top

Access decides whether your pages can be read at all. Content decides what there is to quote once they are. These are practical suggestions to test, in an order that works well in practice: access first, then the facts buyers ask about.

Put the deciding facts in plain text. Pricing, comparisons, warranty and returns terms, integration lists and security details are the facts buyers tend to ask assistants about. Pages that state them plainly give any reader, human or machine, a quotable version.

Keep the story consistent. If your pricing page, your listings and your public documentation disagree, every reader has to decide which to believe, and an assistant quoting the wrong one passes the confusion on. Aligning them is a hygiene recommendation, not a verified engine behaviour.

Structure pages around questions. A heading that matches a real buyer question, followed by a direct answer in the first sentence or two, makes the quotable text easy to locate. Structured files such as llms.txt are a related idea; we cover the evidence in llms.txt: what it is and whether it helps.

Keep third-party sources healthy. Independent reviews and comparisons give assistants material beyond your own site to draw on when a buyer asks for an opinion. Treat this as building a broader source footprint over time, not a switch to flip.

A fictional illustration: two invented brands, Demo Cycles and Rival Bikes, sell at similar prices. Rival's pricing, comparisons and returns terms sit on plain pages with question-style headings; Demo's sit in a long PDF catalogue where they are harder for a reader to find quickly. The comparison is about content layout, not predicted engine outcomes: Rival's facts are easy to locate, Demo's take more digging, and findability is the part a team can fix.

Measuring whether it works

A workable routine, framed as a recommendation: fix a set of buying-conversation prompts in your category, run them on a schedule that includes Claude among the assistants you check, and record the answers with the date and engine each time. For each run, note separately whether your brand was mentioned and whether your pages were cited, and track both over time. One check is a single sample: limited evidence, however it lands. Use repeated checks over several weeks rather than drawing conclusions from one good or bad result, and score two or three named competitors on the same prompts so you can tell a change in your visibility from a category-wide shift. When you do see a change, treat it as a signal to investigate rather than a verdict. If your brand starts disappearing at the trust stage, read those answers, see which sources are quoted instead of yours, and check whether the facts buyers ask about are published plainly anywhere before deciding what to change. Our guide to AI search visibility covers the measurement framework in more detail.

How Vouch helps

Vouch measures how brands appear in AI assistant answers, including Claude. It tracks seven engines on every plan, as of November 2026: ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews, Google AI Mode and Copilot, with single prompts running daily. For deeper analysis, Vouch runs Threads, six-question buyer conversations across the stages Discovery, Shortlist, Objection, Trust, Value and Decision, on ChatGPT, Claude, Gemini and Perplexity three times a week. Threads are modelled test conversations: they record where a brand drops out and which rival replaces it, and the headline metric is the Vouch Rate, a score from 0 to 100 built from how many stages a brand survives and whether it is chosen at the end.

Vouch costs $39 to $599 a month depending on the plan. Access is by invitation while we grow, each invitation opens with a 14-day free trial, and you can ask for one on the waitlist. You can also check what AI crawlers can currently reach on your site with the free AI crawler checker.