Does Claude cite your website?

Updated 2026-06-13 · maintained by Unsourced

The usual instinct with Claude is to block ClaudeBot to keep your content out of it. That misreads how Claude works. ClaudeBot is the training crawler. Claude's live, cited answers come from a web search, which is a different pathway entirely. Block the training crawler and you change what future models learn, not whether Claude can find and cite you today. Here's the real picture.

What most guides get wrong

• Blocking ClaudeBot stops Claude from using my site. It stops your content from training future Claude models. When Claude answers with live web search, it draws on a search index rather than ClaudeBot, so blocking the training crawler doesn't remove you from those answers.
• Claude cites everything it reads. By default Claude is conservative about citing. It cites most clearly when it has searched the web for an answer, which is also when it can link back to you. Whether that happened for a given answer is exactly what Origin Signal disentangles.
• The visitor said Claude, so it was Claude. Anthropic publishes its crawler details, but a user-agent is still only a claim until it checks out against published ranges and reverse DNS.

How Claude surfaces sources

With web search switched on, Claude looks the answer up on the live web (its search is powered by Brave's index) and can cite the sources it used, with links. Without search, it answers from training alone. Anthropic runs ClaudeBot for training and separate user-triggered fetchers for live requests, and like other AI crawlers they read server-rendered HTML rather than running your JavaScript.

The crawlers behind Claude

• ClaudeBot collects training data for future Claude models. Blocking it affects training, not Claude's live web-search answers.
• Claude-User / Claude-SearchBot Anthropic's user-triggered and search fetchers handle live requests. These, not ClaudeBot, are tied to whether Claude can read and cite you in the moment.

Because Claude's search runs on Brave's index, your presence there matters as much as how you handle ClaudeBot.

How to check whether Claude cites you

You can switch on web search, ask Claude what your customers ask, and look for your name. It's a fine gut-check, but Claude's answers move between sessions, so it isn't proof. Unsourced works unlike tools that send the same generic prompts to everyone. We draw the test questions from your own signals: the queries you already rank for in Google and Bing Search Console, the subjects in your published content, and the pages real visitors land on, framed as natural questions. We run them on a schedule, note every time Claude cites your domain and which competitors it reaches for instead, and use Origin Signal to tell a live web-search citation from one drawn out of training. That's measurement on your real demand, not a canned list.

How to improve your chances

• Don't expect blocking ClaudeBot to hide you. It's a training control, separate from Claude's live answers.
• Be findable in Brave. Claude's web search leans on Brave's index, so being indexed there feeds its cited answers.
• Structure for depth. Claude tends to reward clear, well-organised, substantive pages.
• Keep the substance in HTML. The fetchers read server-rendered content, not script-built pages.
• Verify, then measure. Confirm Anthropic's crawls are genuine and track whether Claude actually cites you over time.

No one can guarantee a Claude citation, and anyone who promises one is guessing. What you can do is keep the training and search controls straight, publish answers worth citing, and measure honestly what comes back, and that last part is what Unsourced is built for.

Common questions about Claude

Other AI assistants