AI Traffic Analytics: Measure AI Assistant Visits and Outcomes in GA4

Start with GA4’s AI Assistant channel

Google’s May 2026 Analytics release introduced a native AI Assistant channel. For recognised assistant referrers, Google documents the ai-assistant medium and (ai-assistant) campaign value. Check the values returned by your property before building a separate classification.

Request sessionDefaultChannelGroup with sessions and engaged sessions over a complete period. Inspect the returned channel names, then filter that dimension to equal AI Assistant for a focused report. Keep the property, dates and timezone in the saved output.

Use session-scoped acquisition dimensions for questions about visits starting during the period. First-user dimensions answer a different question about how users were originally acquired. Mixing first-user source with a session-channel comparison changes what the rows mean.

Google’s default-channel definitions place Google AI Overviews and AI Mode within Organic Search rather than the AI Assistant channel. Keep those categories separate when describing your coverage.

Check sources without forcing every visit into Referral

Break down the selected channel by sessionSource and sessionMedium. Retain the original values. A report limited to medium equal to referral would miss assistant traffic classified under Google’s newer medium.

For historical comparisons or an unexpected source, inspect source rows outside the channel as a separate follow-up. Keep a reviewed list of the assistant domains actually observed in your property. Avoid treating every source containing the letters “ai” as an assistant.

The connector supports one case-insensitive dimension filter per request, using contains, equals or begins_with. It does not accept arbitrary regex expressions or an OR list in this report. Use an exact channel filter for the main breakdown, then make separate exact-source requests when needed.

If you manually classify historical source rows, state the rule and coverage. Do not add that source-based total to the channel total without checking overlap. A single visit can already be represented by both reports.

The direct-traffic investigation covers missing attribution. A visit with no usable assistant referrer cannot be reassigned to AI simply because it landed on a page that assistants sometimes cite.

Measure entry pages and documented outcomes

With the AI Assistant channel filter applied, request landingPage, sessionSource and sessions. Start with a small set of dimensions, inspect response warnings and narrow follow-ups when the 1,000-row limit could omit relevant pages.

Use landingPage for where a visit began. pagePath describes pages viewed during visits, so adding sessions across those page rows can count a session more than once. Retain query parameters until you establish which ones identify distinct content or campaigns.

For engagement, use engaged sessions and engagement rate. The API returns engagement rate as a fraction, so 0.6 displays as 60%. Your GA4 engagement and bounce-rate guide explains the calculation and how to compare pages with the same definitions.

The connector’s conversions metric maps to aggregate GA4 key events. Document which events the property marks as key events before treating the total as a business outcome. Twelve key events across 200 sessions is six events per 100 sessions, and does not prove that twelve sessions or twelve customers converted.

Calculate traffic share and engagement separately

This hypothetical property records 5,000 sessions in a complete 28-day period. Its AI Assistant channel contains 200 sessions. The source rows below are an illustrative breakdown after confirming the property’s actual source labels.

AI Assistant traffic represents 200 ÷ 5,000, or 4% of all sessions. The combined engagement rate is 120 ÷ 200, or 60%. Averaging the three displayed rates would produce the wrong combined result because the rows contain different numbers of sessions.

If the previous comparable period recorded 100 AI Assistant sessions, the channel grew 100%, adding 100 sessions. Keep both figures visible. A large percentage on a small base is useful context for a small content experiment, without establishing a dependable growth rate.

Compare the entry pages behind engaged visits and key events. Check whether a page answers its reported topic clearly, provides current evidence and offers an appropriate next step. A small source row should not drive a complete site rewrite.

Ask AI for a report that preserves the measurement limits

Analyse AI Assistant traffic for [GA4 property] over [two complete periods]. Make no account, website or file changes. First retrieve sessionDefaultChannelGroup and inspect the returned values. For AI Assistant, report sessions, engaged sessions and engagementRate by sessionSource, sessionMedium and landingPage using compatible small reports. Use the connector’s supported single dimension filter and preserve row limits, timezone and warnings. Show absolute and percentage changes, channel share of all sessions, and weighted engagement. If conversions is requested, label it aggregate key events and document the property definition. Keep historical source reclassification separate and prevent double counting. Do not estimate citations, missing referrers, prompt text or query-level revenue.

Ask the assistant to return the original source labels beside any presentation labels. Keep a short evidence appendix containing the requested periods, exact filter and unavailable fields so another person can reproduce the analysis.

Compare the report with the GA4 interface for the same dates and dimensions if a surprising result would change your content budget. A missing row can result from a scope or response limit, so investigate the request before calling it zero traffic.

Separate visits, citations and crawler activity

A citation is an appearance in an AI answer, a visit requires a click and a measurable arrival, and crawler activity belongs in server or CDN logs. These are different observations with different denominators. Your report should state which one it measures.

The current GA4 and Search Console operations in SourcetoMCP do not provide a cross-engine citation monitor, the original assistant prompts or crawler access logs. A channel report cannot reconstruct those missing observations. Supply independently collected evidence if you need to compare them.

Google’s AI search guidance explains that its AI features are included in overall Search Console web traffic. The connector’s web-search report should not be relabelled as a complete AI-only report.

Choose one observed landing page for a content experiment. Record its current answer, the factual improvement and the measurement period, then compare subsequent tracked visits and outcomes under the same reporting definitions.