Tracking AI search traffic: how AI Overviews and ChatGPT show up in GA4 (and how they don't)

TL;DR

AI search — Google's AI Overviews and assistants like ChatGPT, Perplexity, and Gemini — is changing how people find answers, and it breaks measurement in two ways agencies need to handle now. (1) Zero-click: AI Overviews answer the query on the SERP, so impressions can hold while clicks fall — visible in Search Console as a rising-impressions / falling-CTR pattern, not as an obvious "AI" line. (2) A new referral source: assistants that do link out send referral traffic that lands in GA4 under hosts like chatgpt.com, perplexity.ai, gemini.google.com — small today, growing, and easy to lose in "Unassigned" or generic referral. The honest state: AEO measurement is immature — you can track AI referrals and watch CTR erosion, but "are we cited in the AI answer?" is still largely a manual/qualitative check. Below: what to track, how, and how to set client expectations without overpromising.

---

Clients are asking "are we showing up in AI search?" and "is AI taking our traffic?" — and the agencies that can give a measured answer (even a partial one) will stand out, because most are either ignoring it or hand-waving. The trick is to measure what's actually measurable today and be honest about what isn't, rather than inventing precision that doesn't exist.

What's breaking: the two effects

1. Zero-click answers (AI Overviews). When Google answers a query directly with an AI Overview, the user often doesn't click through. For the client that means the page can still rank and get impressions while clicks and CTR drop — the traffic is being consumed on the SERP. This is the bigger, harder-to-see effect, because it shows up as erosion, not as a labelled channel.

2. A new referral stream. Assistants that cite and link (ChatGPT with browsing, Perplexity, Gemini, Copilot) send users to the site. That traffic arrives in GA4 as referral from the assistant's domain — a genuinely new acquisition source that's small now but compounding, and that frequently gets mislabelled or lost in Unassigned if you're not looking for it.

What you CAN measure today

AI referral traffic in GA4

Build a segment / report that isolates traffic from AI hosts — chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and similar. An exploration filtered to those sources (or a custom channel group that buckets them as "AI") turns an invisible trickle into a trend you can report. Track its volume, conversions, and growth over time. It's small — report it as emerging, not as a primary channel — but it's real and it's the part you can actually quantify.

CTR erosion in Search Console

The AI-Overview effect shows up in Search Console as queries/pages where impressions hold or rise but clicks and CTR fall. That divergence is your best available proxy for "AI is answering this query before the user reaches us." Watch it at the query and page level for the client's important terms — a widening impressions-vs-clicks gap on informational queries is the AEO signal.

Branded-search and direct lift

If AEO is working — the client is being cited in AI answers even without a click — one downstream signal is a lift in branded search and direct traffic (people see the brand in an AI answer, then search it). It's indirect, but it's measurable, and it's consistent with how AI visibility actually converts.

What you mostly CAN'T measure (yet)

Be straight with clients here:

• "Are we cited in the AI answer?" is largely a manual/qualitative check today — running the client's key queries through the assistants and AI Overviews and seeing whether they're referenced. Some emerging tools attempt AI-visibility tracking, but it's early and inconsistent; don't sell it as precise.
• Attribution of an AI-influenced conversion is murky — someone who read an AI answer, formed intent, and converted days later via direct looks like "direct" in every model. Like upper-funnel media, AEO's influence is real and partly unmeasurable with today's tooling.

The professional move is to say so — report the measurable parts (AI referrals, CTR erosion, branded lift) and frame the rest as a known limitation, not a number you've invented.

Set the client expectation

The conversation to have: AI search is changing how people find you. We can track the new AI referral traffic and watch where AI Overviews are eating clicks, and we'll optimise to be the cited answer — but some of AI's influence won't show up cleanly in analytics yet, so we'll report what's measurable honestly rather than fabricate precision. That candor is itself a differentiator in a space full of hype. Put the AI-referral channel and the CTR-erosion watch into the measurement plan so it's a tracked, recurring thing rather than a one-off curiosity.

Where this fits

AI search adds a new source to capture and a new erosion to watch — both of which quietly hide in GA4's "referral" and "Unassigned" buckets and in Search Console's blended numbers unless someone deliberately surfaces them. Phloz keeps each client's channels and search data legible — including an "AI" view of the emerging referral traffic and the CTR-erosion signal — as part of the tracking-infrastructure map, so "is AI search sending this client anything, and where is it taking clicks?" is a view you can act on instead of a question you wave away. The CRM for SEO agencies and pricing pages cover the workflow — but the takeaway is to measure AI search like the early, partly-invisible channel it is: track the referrals, watch the CTR erosion, and be honest about the rest.