how to track Claude referral traffic

to track Claude referral traffic, count visits with a referrer hostname of claude.ai, report them as their own channel rather than one row in a referrers list, and add ?utm_source=claude to links you place yourself. most Claude clicks come from the desktop app or from URLs copied out of an answer, both of which arrive with no referrer and get bucketed as direct, so your measured number will always be lower than reality.

why Claude traffic is harder to see than ChatGPT traffic

it's the same underlying problem, with the mix tilted worse.

Claude's usage skews toward the desktop app, toward long research-style conversations, and toward developers who read an answer, copy the link, and open it later next to their editor. every one of those paths strips the referrer. what you're left with in your referrals report is the browser-tab slice of an audience that spends a lot of its time outside a browser tab.

then there's a category that barely exists for other engines: agents fetching your pages on someone's behalf. a developer runs Claude Code, points it at your docs, and your server gets a request that is neither a crawler indexing you nor a human reading you. it's a person consuming your content through a tool. that request has no referrer, no session, and no chance of appearing in client-side analytics at all, because no JavaScript runs.

so the honest framing for Claude is: you can measure the referred browser clicks, you can measure the crawler and agent fetches from your server logs, and there's a gap in the middle you infer rather than count. treat referrer data as evidence, not as a ledger. the same caveats apply to every engine, and they're laid out in how to see how much traffic ChatGPT, Claude, and Perplexity send you.

what each Claude surface actually sends

the bottom four rows are the ones people misread. they are real signal, they just aren't visitors, and mixing them into a traffic number will make you believe something that isn't true.

how to actually measure it

1. add claude.ai to your AI channel rule. if you already built a channel group for chatgpt.com and perplexity.ai, this is one more hostname. keep Claude visible as its own line inside that channel, because the volume is usually smaller and it disappears if you only look at the aggregate.
2. look at the channel weekly, not the referrers list. a hostname with 40 visits will never surface in a top-referrers view. a channel with a trend line will.
3. tag every link you place. docs URLs you paste into conversations, links in your README, links in support replies. ?utm_source=claude&utm_medium=ai recovers a slice that the referrer never would.
4. pull crawler and agent hits from server logs separately. grep for ClaudeBot, Claude-User, Claude-SearchBot in your access logs. report them as "pages fetched for AI answers," never as visitors. this is the metric that moves first when your content starts getting used.
5. watch which paths the agents hit. if Claude-User and Claude Code requests concentrate on three docs pages, those are the pages doing the work. that's a content roadmap you got for free.
6. compare the Claude cohort's behaviour to search. small channels can still be your best channel. measure signup rate, not just sessions, before you decide it's noise.
7. watch direct in parallel. if Claude referrals and unexplained direct traffic rise together, that's the invisible half confirming itself. it's not proof, it's the best correlation available.

the honest part: what this number can and can't tell you

referrer headers are a courtesy, not a guarantee. the sending page decides whether to send one, browsers trim them by policy, and native apps skip them entirely. that was already true for messaging apps and email clients, and AI surfaces are worse because so much of the interaction happens outside a browser.

what that means in practice:

• trend is trustworthy. if your Claude channel triples over two months, something real changed, even if the absolute number is understated.
• landing pages are trustworthy. whatever pages show up are genuinely being cited or linked. the sample is partial but it isn't biased toward particular content in a way that would mislead you.
• share of total traffic is not trustworthy. you cannot say "Claude is 4% of our traffic" with a straight face. you can say "at least 4%."
• crawler volume is not traffic. a thousand ClaudeBot hits is a thousand fetches, not a thousand readers.

I'd rather publish a number with a stated floor than a confident number that's wrong. that's the same reason attribution is worth reading before you argue with anyone about channel credit: last-touch numbers are a model, and the model has gaps in exactly the places AI traffic lives.

what Claude traffic is usually worth

the pattern most people see is small volume, high intent. someone asked a question, got a considered answer that named you with a reason, and clicked through already half-convinced. that's a different visitor from someone who scanned ten blue links.

so measure it as a cohort, not as a rounding error:

• conversion rate versus organic search
• pages per session, since research-mode readers tend to go deeper
• which entry pages produce signups, not just visits

if your docs are what gets cited, that tells you docs are marketing, which changes how you staff them. if a single comparison page does all the work, write two more like it. the fuller checklist is in what to measure once AI is sending you traffic.

FAQ

does claude.ai send a referrer?

sometimes. links clicked from claude.ai in a normal browser tab usually pass claude.ai as the referrer. clicks from the desktop and mobile apps, and URLs copied out of a conversation, generally arrive with no referrer and get counted as direct.

how do I tell Claude's crawler from a real person?

by user agent, in your server logs. ClaudeBot, Claude-User and Claude-SearchBot identify themselves. none of them run JavaScript, so they will not appear in client-side analytics at all unless you're also counting server-side requests. keep them in a separate report from visitors.

why is my Claude referral traffic so low?

partly because the channel is genuinely smaller than search, and partly because a larger fraction of it is unmeasurable. the app-heavy usage pattern means more referrer-less clicks. read your number as a floor and watch the trend rather than the absolute.

is it worth tracking a channel this small?

usually yes, because the quality tends to be high and the trend tells you whether your content is getting cited at all. a channel that's 1% of traffic and 8% of signups is not small in the way that matters.

---

smolanalytics tracks claude.ai, chatgpt.com and perplexity.ai as their own channel with nothing to configure, and you can ask it questions in plain english straight from your editor over MCP. see what's in it, pricing, or the setup for apps built with AI.