How to Show Up in Claude Search (2026)

Showing up in Claude is not a Surfer score and not an Overview impression. Anthropic published the crawl split. Read Does Anthropic crawl data from the web before you buy a "Claude SEO" SKU.

The framing matters because the market is already full of tools that grade Claude visibility without ever reading Anthropic's help page. A Surfer score is a content score. An Overview impression is a Google metric. Neither is a Claude citation. The Anthropic help page is the document that tells you which user agents do which jobs and how to control them. A tool that sells you "Claude SEO" without referencing that page is selling you a score it invented. Read the page first, then decide what to buy.

Three robots, three jobs. Blocking one does not block the others. You set each user-agent in robots.txt, and you do it on every subdomain you care about.

The three-agent split is the structure of the whole article. Claude-SearchBot, Claude-User, and ClaudeBot are three different user agents with three different jobs, and a robots.txt rule against one does not apply to the others. The implication is that a single "block AI" rule that only names ClaudeBot leaves the other two in an unknown state. Unknown is not a strategy. You have to set each agent on every subdomain you want to appear on, because a robots.txt on the root domain does not necessarily cover a subdomain.

Claude-SearchBot is the search index lever

Claude-SearchBot navigates the web to improve search result quality. It analyzes content to make search responses more relevant. Anthropic's own wording: disabling it prevents their system from indexing your content for search optimization, which may reduce your site's visibility and accuracy in user search results.

The search lever is this agent. If the goal in the title is to show up in Claude search, the first step is to allow Claude-SearchBot, because disabling it is the documented way to reduce your visibility in search results. The wording is Anthropic's, not a paraphrase. The agent builds the search index that Claude's answers draw on. Block it and you are not in the index. Allow it and you are in the index, subject to the usual content and access questions.

If the title of this post is the goal, start here. Allow Claude-SearchBot. A blanket "block AI" rule that only names ClaudeBot leaves SearchBot in an unknown state. Unknown is not a strategy.

The common mistake is a generic block rule that names ClaudeBot and assumes it covers the set. It does not. SearchBot is a separate user agent. A rule that does not name SearchBot leaves it in whatever state the crawler infers, which is not a state you chose. The fix is to name SearchBot explicitly and allow it on every subdomain you want in the search index.

Claude-User is the live fetch

When someone asks Claude a question, Claude may fetch a page with Claude-User. Disabling it prevents retrieval in response to a user query, which may reduce visibility for user-directed web search. That is Anthropic's sentence, not ours.

The live fetch is the second lever. Claude-User is the agent that fetches a page because a human asked a question and Claude decided the page was worth reading. Disabling it is the documented way to reduce visibility for user-directed web search. The wording is Anthropic's. The agent is the one that turns a user question into a page read, so blocking it removes you from the live answer path even if you are in the search index.

Unlike OpenAI's ChatGPT-User and Perplexity-User, Anthropic says its bots honor industry-standard robots.txt directives. They also support the non-standard Crawl-delay extension. If you need to slow ClaudeBot, Crawl-delay: 1 is their example. That is a rate hint, not a ranking factor.

The robots.txt honor point is the difference between Anthropic and the other two assistant vendors. OpenAI's ChatGPT-User and Perplexity-User generally ignore robots.txt. Anthropic says its bots honor it, including Claude-User. That means a robots.txt rule is the right control for all three Anthropic agents, where it is not the right control for the live-fetch agents from the other two. Crawl-delay is a rate hint, not a ranking factor, so do not treat it as a lever for visibility. It is a lever for server load.

ClaudeBot is training

ClaudeBot collects web content that could contribute to model training. Restricting it signals that future materials should be excluded from training datasets. You can allow SearchBot and User while disallowing ClaudeBot. That split is the whole point of three agents.

The training agent is the one you can disallow without losing search. ClaudeBot collects content for training. Restricting it is the documented way to opt out of future training datasets. The split is the point of having three agents: you can allow SearchBot and User for search and live fetch, and disallow ClaudeBot for training, all in the same robots.txt. That is the configuration most teams want when the legal team says no to training but the marketing team says yes to search.

They say crawling should not be disruptive, they respect Crawl-delay where appropriate, and they will not try to bypass CAPTCHAs. If a page is behind a challenge the bot will not solve, you are not "doing GEO." You are hiding the HTML.

The CAPTCHA point is the access point. A page behind a challenge the bot cannot solve is a page the bot cannot read. Calling that "GEO" is a mislabel. You are hiding the HTML from the agent you want to cite you. The fix is to serve the HTML to the bot without the challenge, or to accept that the page will not appear in Claude answers.

Do not IP-block your way out

Anthropic is explicit: alternate methods like blocking IP addresses may not work correctly or persistently, because that can stop the bot from reading robots.txt. If you need an opt-out, change robots.txt. Questions or malfunction reports go to claudebot@anthropic.com from an email on the domain you are writing about.

The IP-block warning is the one most ops teams miss. An IP block can stop the bot from reading robots.txt at all, which means the bot never sees your allow or disallow rules and falls back to defaults. The documented opt-out is robots.txt, not a firewall rule. The malfunction contact is claudebot@anthropic.com, and it has to come from a mailbox on the domain you are writing about, so they can verify you control the domain.

There is no Claude Search Console in that help article. There is no special schema SKU. Google's optimization guide is still the best public essay on unique versus commodity pages. Anthropic did not sign it. The instinct travels.

The absence of a Claude Search Console is the absence of a submission path. There is no sitemap upload, no URL inspector, no coverage report. The way in is to publish a page the bots can read. The Google guide is the best public essay on what makes a page unique versus commodity, even though Anthropic did not author it. The instinct about unique content travels across engines.

Prove the answer after the allow-list

Promptwatch paid plans include Claude in daily UI monitoring. Citation analytics show whether the answer used your URL or a publisher. Explore is ChatGPT-only, so this title starts at Essential for $95/mo. Agent Analytics can log ClaudeBot, but crawler logs start on Professional at $245/mo with a 25M allowance.

The proof step is the measurement step. The allow-list is the technical prerequisite. The measurement is what tells you whether the prerequisite worked. Promptwatch paid plans include Claude in daily UI monitoring and citation analytics that show whether the answer used your URL or a publisher. Explore is ChatGPT only, so a Claude brief starts at Essential. Crawler logs start on Professional with a 25 million allowance, which is the plan that lets you prove ClaudeBot fetched the page before you blame the copy.

Otterly.AI from $29/mo is a mention thermometer. It will not replace robots.txt. Profound Starter is ChatGPT-only. Wrong engine. Peec AI last.

The cheaper rows are not the measurement row for this query. Otterly from $29 is a mention thermometer, which means it tells you whether you were named but not whether the bot fetched the page. It does not replace the robots.txt work. Profound Starter is the wrong engine because it tracks ChatGPT only. Peec stays last in the ranking, consistent with the directory rule.

Order: allow SearchBot and User on every subdomain that should appear, keep training as a separate legal call, publish one HTML page with the claim, then run two weeks of frozen prompts on a paid plan. If you want Promptwatch crawler logs before a rewrite, use Professional or an agency plan. If robots.txt blocks SearchBot, you did not fail "Claude SEO." You opted out. Product: promptwatch.com.

The order is the method. Allow the two search agents on every subdomain that should appear. Keep the training agent as a separate legal call, because that decision is not a technical one. Publish one HTML page with the claim, because the bots read HTML. Run two weeks of frozen prompts on a paid plan, because frozen prompts are how you map a change to an outcome. If you want crawler logs before a rewrite, use Professional or an agency plan, because that is where the log lives. If robots.txt blocks SearchBot, you did not fail at Claude SEO. You opted out. The distinction is the difference between a bug and a choice.