How to rank in Claude

• 13 of 38

  shut out Claude-SearchBot, the token Claude’s search depends on

• 22 of 38

  shut out ClaudeBot, the training crawler

• 9

  block training and still let Claude’s search in

• 15

  still name anthropic-ai or Claude-Web, tokens missing from Anthropic’s current list

Shut out: the rules that apply to that token forbid an ordinary page. Open: they allow it, though some sections may still be off limits. Twelve sites could not be read: ten answered with a bot check, an error or nothing, and two sent us to an international edition. We fetched each file with an ordinary browser user agent; a site can serve crawlers a different one, and a firewall can block what robots.txt allows.

Scroll sideways to see every column.

In all: 16 searches, 154 results back and 60 pages cited, about 39 in 100 of what the searches returned. One run on one model is a snapshot: claude.ai, another model or another day can answer the same question differently.

Seven things the answers show

• It searched for 11 of 12. The exception was the how-to question about web hosting. Claude treated it as stable knowledge, the case in which its API documentation says it answers without searching.

• It added the year on its own. 9 of its 16 searches ended in “2026”, though no question mentioned a year. The searches were short rewrites of the question, such as “best CRM for small sales team 2026”.

• Second searches went after prices. 5 answers ran a second search. 3 of those asked for pricing, and two named the leaders outright: “Gusto vs Rippling vs ADP pricing comparison” and “Klaviyo vs Mailchimp pricing small ecommerce”.

• Lists and comparisons won. 59 of the 60 cited pages had a list or comparison title (best, top, a number, vs, alternatives or compared), and 47 carried 2026 in the title or the address.

• Vendors were cited for their own categories. In 7 of the 11 answers, at least one cited page was a “best of” list published by a company that makes a product in that category: Pipeline CRM and SuperOffice on CRM, Drip and Sender on email marketing, Joist on invoicing.

• Few household names. The 60 citations came from 58 different sites and no site appeared in two answers. None was a .edu or .gov page, and only five of the sites were among the 38 big ones in the table above.

• Blocking training did not block citations. Of the 57 cited sites whose robots.txt we could read, 5 shut out ClaudeBot and were cited anyway, Forbes and Tripadvisor among them. None shut out Claude-SearchBot or Claude-User, and none of the 154 results came from the 13 big sites that do.

1. 1

   Open the two doors Claude uses

   Allow Claude-SearchBot and Claude-User in robots.txt, on every subdomain you want found. Then look past robots.txt: a firewall or bot filter can turn Anthropic away before it reads the file, and Anthropic publishes the IP ranges of its crawlers so you can let them through. If you want out of model training, block ClaudeBot; it is a separate token. The free AI crawler checker lists, token by token, what your robots.txt allows for Claude.

2. 2

   Rank for the searches Claude runs

   Claude turns a buyer’s question into a short web search, often with the current year added, and sometimes into a second one that compares the leaders on price. Write those searches down for your main questions and make sure that a page of yours, or a page that lists you, ranks for them. Through the API every query comes back in the response; on claude.ai you see when Claude is searching.

3. 3

   Publish the comparison Claude is looking for

   Nearly every page Claude cited in our sample was a ranked list or an “X vs Y” page. If you sell in a category, publish an honest “best X for Y” or “X vs Y” page on your own site: name the rivals, give real prices and say who each option suits, yourself included. Claude cited pages like that from vendors in most of our answers.

4. 4

   Write facts Claude can quote

   A Claude citation carries up to 150 characters of the passage it leans on. Put each fact you want repeated (the price, who it is for, a key spec, the city you serve) in one short sentence under a heading that matches the question. Keep it in the HTML itself: Anthropic does not say whether its crawler runs scripts, so text that only appears after JavaScript is a risk you do not need.

5. 5

   Keep prices and dates current

   Each search result reaches Claude with its page age, and its follow-up searches ask for prices. Update prices and plan names when they change, show the date of the last real update, and take down pages that still quote old figures.

6. 6

   Get onto the pages Claude already cites

   Ask Claude the questions your buyers ask and list the third-party pages it cites. Check that each one lets Claude-SearchBot in, then earn a place on the open ones: an accurate listing, a product for the reviewer, data the author can use. A famous site that shuts Anthropic out is a weaker target than a niche one that lets it read. Doing this for dozens of questions is what AI citation tracking is for.

7. 7

   Ask the same questions every week

   Claude writes each answer fresh and searches again each time, so a single check proves little. Keep a fixed list of questions, ask them on a schedule, and count how often you are named or cited over several weeks, not in one reply. Before you trust a rise or a fall, read how often to measure and why answers move.

On hidden text, see Google’s spam policies (opens in a new tab). On training cutoffs, see Anthropic’s note on how up to date each model is (opens in a new tab).

By hand. Open a new chat on claude.ai and ask your buyers’ questions word for word, without your brand name. For each answer, note whether Claude searched, whether it names you and in what place, which pages it cites and whether one of them is yours. Ask again on other days, because the answer moves.

With Slibur

• Weekly on paid plans. Slibur puts your prompts to Claude’s model once a week with web search turned on. ChatGPT and Google AI Overviews are the two it reads every day*.

• Once during the trial. In the 7-day trial, Claude answers up to 10 prompts on the day you sign up.

• The full reply and its sources. Slibur keeps what Claude wrote and the URLs it cited, labelled by owner: your domain, a tracked rival or someone else. Filter by Claude to see only its answers.

• A warning when Claude lags. When Claude is your weakest AI platform and the share of its answers naming you sits 25 points or more below your strongest one, the weekly improvements raise “Claude barely mentions you” and point you to its sources.

• Alerts with enough history. Alerts for Claude start once it has eight weekly runs behind it.

• Two limits. Slibur reads Claude’s model with a web search, not the claude.ai app, so a person there can get a different answer drawn from other pages. And one weekly answer per prompt is a sample: trends need several weeks.

On Starter (20 prompts, $59 a month) that is about 80 Claude answers a month, out of roughly 1,440 across the five AI platforms. How each one is read is on the methodology page.

* On paid plans. ChatGPT is read on chatgpt.com in every country Slibur tracks except Venezuela and Puerto Rico, where it is asked weekly through its API.

• AI crawler checkerType your domain and get a verdict for Claude-SearchBot, Claude-User and ClaudeBot, with the exact rule behind each one.Use it free
• Free GEO audit tool18 tests on your site, from robots.txt and the sitemap to the text a crawler reads when scripts do not run (step 4).Use it free
• GEO audit checklistA wider audit for every AI platform, scored out of 100, with the source behind each check.Use it free

• How to rank in ChatGPTWhat makes ChatGPT recommend a business, and how to check whether it recommends yours.Read the guide
• How to rank on PerplexityHow Perplexity chooses the pages it cites, and how to check whether it cites yours.Read the guide
• How to rank in AI OverviewsHow Google picks the sources for an AI Overview, and how to check whether yours are cited.Read the guide
• AI crawler checkerFree check of whether GPTBot, ClaudeBot, PerplexityBot and other AI crawlers can read your site.Read the guide
• LLM visibility toolsHow LLM visibility tools read live answers, and which language models each one covers.Read the guide
• AI citation trackingEvery URL the five AIs cite when they answer, and which of them send the credit to your rivals.Read the guide
• AI competitor analysisSee which competitors AI platforms recommend instead of you, and which sources they rely on.Read the guide

What this guide says about Claude comes from Anthropic’s articles on its crawlers (opens in a new tab), on its hosted browsers (opens in a new tab), on web search (opens in a new tab) and on training data (opens in a new tab), from its web search tool documentation (opens in a new tab) and from its crawler IP list (opens in a new tab). The ChatGPT and Perplexity columns come from OpenAI’s (opens in a new tab) and Perplexity’s (opens in a new tab) crawler pages, and the matching rules from RFC 9309 (opens in a new tab). All were read on November 9, 2026, the same day as the robots.txt files and the 12 answers. Anthropic does not publish how Claude ranks one source above another, so the steps on how to rank in Claude are practical advice built on what it does publish.