How to track AI referral traffic in GA4.

The slow way: hand-picking sources from the referral table

The manual version is a monthly scroll through Reports → Acquisition → Traffic acquisition, eyeballing the source list for chatgpt.com and perplexity.ai, and copying the numbers into a spreadsheet. It sort of works, and it fails in four quiet ways. It misses the utm-tagged sessions that arrive with no referrer. It can't trend without the spreadsheet ritual repeating every month. It says nothing about Google AI Overviews, whose clicks route through google.com and blend into organic. And it badly undercounts ChatGPT, because in-app browsers and agent fetches often strip the referrer entirely — those sessions land in Direct, indistinguishable from a bookmark.

The undercount is the dangerous part. A hand-built number that reads "0.4% of traffic" gets quoted in a planning meeting as proof that AI search doesn't matter yet, when the measurable referral slice was never the whole of AI influence — the shopper who reads an answer and then searches the brand by name shows up as branded search, not referral. Built honestly, the number is a floor. Built carelessly, it is an argument for ignoring the channel where buying decisions are increasingly made.

The setup, step by step

1. Build the channel group. GA4 → Admin → Data display → Channel groups. Add an "AI Referrals" channel matching session source against the known engine domains: chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai — current as of mid-2026; engines change referrer behavior without notice. Google's own reference for how channel conditions evaluate is the GA4 channel-group documentation.
2. Catch the utm-tagged arrivals. ChatGPT appends utm_source=chatgpt.com to many outbound links as of mid-2026, and the tag survives even when the referrer is stripped. Include the utm source in the same channel condition, or those sessions split off from the rest of the ChatGPT traffic.
3. Report at landing-page grain. Build an exploration: sessions, engaged sessions, conversions, and revenue by landing page, filtered to the AI Referrals channel. Landing page is the closest GA4 gets to the SKU, and SKU grain is what makes the number actionable — a channel total tells you AI traffic exists; a per-PDP series tells you which products earn it.
4. Footnote the undercount, every time. Stripped referrers land in Direct; AI Overviews clicks blend into Google organic; engines influence purchases without sending a click. State in the report that the figure is a floor on AI influence, not a ceiling. The cautious framing is also the credible one.
5. Join the series to citation data. This is the half GA4 cannot do alone. Prompt runs timestamp every citation per SKU, designed to sit next to the GA4 series — when a PDP's citations rose and its AI-referred sessions rose in the same window, the lift is attributable, and that joined number is what survives a budget review. The Shopify side of the revenue join is covered in the Shopify integration docs; the boardroom packaging is its own job: prove AI search ROI to leadership.

What "good" looks like

One pattern worth watching once the channel exists: where referrers are measurable, AI-referred visitors often convert better than search-referred ones, because the answer did part of the education before the click — a pattern eCommerce Insights has observed across tracked D2C catalogs to date, worth validating on your own data rather than assuming. The session series is half the attribution join; per-SKU citation tracking via prompt tracking is the other half.

Related jobs