AI search traffic is measurable today with the analytics you already run. It arrives as ordinary referral visits, and the work is in segmenting it before it drowns in the rest, plus knowing which caveats make the number an undercount.
The referrers to segment
Build one segment (in GA4, a custom channel group) matching these referrer domains: chatgpt.com and chat.openai.com for ChatGPT, perplexity.ai for Perplexity, copilot.microsoft.com for Copilot, and gemini.google.com for Gemini. Traffic from Google's AI Overviews is the awkward one: it arrives as ordinary google.com organic traffic and cannot be separated in analytics; Search Console impressions are the closest proxy Google offers.
URL parameters help
ChatGPT appends utm_source=chatgpt.com to many outbound links, which survives even when the referrer is stripped, so match on that parameter too. When you see both the parameter and the referrer, dedupe in that order: parameter first, referrer as fallback.
Why the number is an undercount
- Answers satisfy many readers without any click, and no analytics tool sees a citation that was read but not clicked.
- Some in-app browsers and privacy settings strip referrers, landing those visits in direct traffic.
- AI Overviews traffic hides inside organic, as above.
Pair it with citation sampling
Referrals measure clicks; they say nothing about how often you are cited. The complement is sampling: a fixed list of the questions you care about, asked in each engine monthly, with citations recorded in a sheet. Together the two numbers tell you whether visibility or click-through is your bottleneck. For tool support, see Best AI SEO tools in 2026; for raising the citation count itself, start with the AEO guide.