← All posts

Claude citation tracking: the practical guide for 2026.

[REPLACE WITH REAL AUTHOR NAME]6 min read

Claude is the quietest of the major assistants in most analytics dashboards, and the one teams most often assume sends nothing. Usually it does send something — it just arrives without a label. Here is how to identify it, what to expect, and what to build.

ClaudeBot signature

Anthropic documents its crawler behaviour in its site-owner article. Three identities are worth separating in your reporting:

  • ClaudeBot — the general web crawler.
  • Claude-User — a fetch performed because a person asked Claude something that required loading your page.
  • Claude-SearchBot — fetches connected to search functionality.

Each is a distinct user-agent token, so a server-side classifier can split them cleanly without heuristics. Citealytics tracks all three against the registry on the crawlers page, with a last-verified date per signature so you can see when a token was last confirmed against live traffic.

Where Claude referrals come from

Claude referrals reach you from the web application and from the desktop and mobile clients when a user clicks a linked source. Because a large share of Claude usage happens in contexts where no link is clicked at all — code assistance, drafting, analysis of pasted material — Claude tends to produce fewer referred sessions than crawler fetches. That ratio is a property of how the product is used, not a measurement failure, and it is the reason you should report crawl activity and referrals side by side rather than collapsing them into one AI number. We are not publishing a referral share figure here because we do not have a source for one that would survive scrutiny; measure it on your own property instead.

Build a citation dashboard for Claude

  • Fetches by identity. ClaudeBot vs Claude-User vs Claude-SearchBot, weekly. A rise in Claude-User is the signal that people are asking about your topic.
  • Most-fetched URLs. The pages Claude reaches for. If your best commercial page is absent, that is a content-structure problem you can fix.
  • Referred sessions and landing pages. The human half of the picture.
  • Engine share over time. Claude as a percentage of all AI activity, so you can spot engine-specific shifts.

One caveat worth keeping in view: an assistant that answers completely and sends nobody through is still shaping demand. Crawl coverage is the only observable proxy you have for that influence, which makes it a first-class metric rather than a technical detail.

Practical setup

Do not block ClaudeBot unless you have decided you do not want to appear in Claude answers. Do keep your robots directives explicit rather than accidental — many sites block these crawlers through a copied ruleset and then wonder why they are invisible. Then install server-side collection so the fetches are actually recorded: a client-side tag will never see a crawler, because crawlers do not execute JavaScript.

Frequently asked questions

What user-agents does Claude use?

Anthropic documents ClaudeBot for general crawling, Claude-User for fetches triggered by a person's request, and Claude-SearchBot for search-related fetches.

Why do I see Claude crawler hits but no Claude referrals?

Much Claude usage never involves clicking a source link. Crawl coverage is the observable proxy for that influence, so track fetches and referrals separately.

Does blocking ClaudeBot protect my content?

It removes your pages from Claude's crawl, which also removes any chance of being cited or referred. It is a trade-off, not a pure win.

Can I track Claude without cookies?

Yes. Both crawler identification and referral classification use the request's user-agent and referrer, which require no cookie and no device storage.

Measure the AI traffic your analytics is hiding.

One line of script, no cookies, no consent banner. Citealytics classifies every visit against a registry of answer-engine signatures and shows which pages get cited.

Start free — 10k events/mo →

Related posts