How To Get Traffic From Claude

• Measurement comes before optimisation, because a channel you cannot isolate cannot be tuned.
• The Claude desktop and app frequently strip the referrer.
• A custom channel group lets you label AI sessions in one place rather than rely on default groupings.
• GA4 added a native AI Assistant channel to its Default Channel Group in May 2026.
• Reconciliation means checking GA4 against a second source before trusting a Claude figure.

Measurement comes before optimisation, because a channel you cannot isolate cannot be tuned. Claude referrals in GA4 appear with claude.ai as the session source and referral as the medium.

To view them, open Reports, go to Acquisition, then Traffic acquisition, set the primary dimension to Session source/medium, and search for claude.ai, as described by Perkfuel and Hawk Academy. Claude traffic is growing quickly, with Perkfuel reporting roughly 4x referral growth within a year and an over-index among developers, analysts, writers and founders (Perkfuel). Those audiences shape which pages earn visits. Expect the referral row to be the smaller part of the picture, because the rest arrives under Direct.

The Claude desktop and app frequently strip the referrer. The browser then reports no source, so GA4 files the session under Direct.

• Hawk Academy states that the Claude desktop app does not pass a referrer.
• Those sessions therefore default to Direct (Hawk Academy).
• Perkfuel and Loamly describe the same pattern (Perkfuel, Loamly).

However, the exact share cannot be recovered from GA4 alone. Specifically, a Direct spike with no campaign behind it can be partly Claude. Treat any Claude count as a lower bound.

A custom channel group lets you label AI sessions in one place rather than rely on default groupings. Beket.ai recommends an AI Search channel group with a source regex covering claude, anthropic, chatgpt, perplexity, gemini and similar sources, placed above the Referral and AI Assistant channels (Beket.ai). Placement matters, because channel rules are evaluated in order.

Saving the group also reclassifies historical data, which lets you compare periods on the same basis. According to Loamly, a regex on source names may catch only roughly 30 to 50% of actual Claude traffic when referrers are present, so read the group as a partial view (Loamly).

GA4 added a native AI Assistant channel to its Default Channel Group in May 2026. It tags sessions from recognised AI tools with the medium ai-assistant, as reported by Hedgehog Marketing. At launch, Google named ChatGPT, Gemini and Claude as the recognised sources (Hedgehog Marketing). Two limits matter.

The channel processes data only forward from its launch, so earlier AI visits remain in Referral or Direct with no back history. It also inherits the same referrer gap, so sessions without a referrer still fall outside it. Beket.ai adds that GA4 counts only visits that click through with JavaScript running, and that AI crawlers such as GPTBot never reach GA4 at all (Beket.ai).

Reconciliation means checking GA4 against a second source before trusting a Claude figure. Loamly describes a server log method: search raw HTTP logs for ClaudeBot or anthropic-ai user agents, then correlate crawler hits with human visits to the same URL within about five minutes (Loamly). This is correlation, not definitive attribution, so it indicates likely Claude visits rather than proving them.

Hawk Academy recommends reconciling GA4 with Microsoft Clarity, Bing data and server logs before trusting Claude figures (Hawk Academy). A useful routine is to compare GA4 referral counts, the Direct trend and log correlations for the same pages over the same weeks. Where the three agree on direction, the trend is credible even if the absolute number is not.

Claude visitors skew toward developers, analysts, writers and founders, and Perkfuel reports that Claude over-indexes on those groups. Those visitors also tend to stay longer. Semrush data cited by Loamly puts Claude visitors at an average of 9 minutes and 29 seconds per session, which is higher engagement than many other AI platforms (Loamly). That profile favours pages with depth: specific methods, documented steps and clear data.

Specifically, pages that answer one precise question with verifiable detail are more likely to be cited and visited than thin summaries. Keep each page self-contained, so a reader arriving from an AI answer gets the full answer without needing navigation. Conversion depends on matching that intent with a relevant next step, not a generic call to action.