AI traffic from GA4

AI visibility tells you whether ChatGPT, Gemini and Perplexity mention a brand. AI traffic closes the loop: it shows how many people those assistants actually send to the website, and what they do once they arrive. Once a Google Analytics 4 property is connected, CrunchJunkie can isolate the sessions, engaged sessions, key events and revenue that originate from AI assistants — turning "we're visible in AI" into a number you can put in front of a client.

GA4 records a source and a medium for every session. CrunchJunkie counts a session as AI traffic in two cases. The first is a source that belongs to an AI assistant. A visit that arrives with the assistant's referrer has the assistant's host as its source: chatgpt.com, chat.openai.com, gemini.google.com, perplexity.ai, copilot.microsoft.com, claude.ai, meta.ai, grok.com and a few others. A visit from a link the assistant has tagged has the tag instead, because GA4 takes a utm_source over the referrer: links ChatGPT cites have been seen tagged utm_source=openai, and Gemini has been seen tagging links utm_source=gemini (Plausible, September 2026). Both forms are counted, whatever their upper or lower case. The second is GA4's own verdict. GA4 gives sessions from the assistants on Google's list the medium "ai-assistant" (its AI Assistant channel), and every such session counts too. A session that meets both conditions is counted once. Clicks tagged with a paid medium (cpc, ppc, paid and the like), a display medium (display, banner and the like), an email medium or affiliate are not counted, even when the source is an assistant's name: a ChatGPT Ads click tagged utm_source=chatgpt with a cpc medium is bought traffic and belongs to the ChatGPT Ads channel. An ad click that carries no tag arrives as the assistant's host, and nothing in GA4 tells it apart from an earned one. openai.com and x.ai are company sites, and bing.com can be Copilot in Bing or anywhere else on Bing, so all three are counted under their own names rather than as an assistant. There is nothing to install; the match runs on the GA4 data you already have, and the list is updated as assistants change how they link. Reports ask GA4 when they are opened, so a change to this list applies to every period, earlier ones included. When the tagged links were added in October 2026, AI traffic for past months could rise the next time a report was opened, and fall only where clicks tagged with an assistant's host and a paid, display, email or affiliate medium (utm_source=chatgpt.com with cpc, say) had been counted. Saved AI summaries and PDFs that were already sent keep the figures they were written with. One surface is not on that list and cannot be: Google's own AI answers. A click from an AI Overview or from AI Mode arrives in GA4 as google / organic, indistinguishable from a click on a classic blue link, because Google sends no separate referrer. Search Console shows impressions for these surfaces in its generative-AI report, but only in the interface, without clicks, and since 7 September 2026 with AI Overviews and AI Mode combined; its API exposes none of it. So AI traffic here counts the assistants that identify themselves; Google's AI surfaces are measured on the visibility side (AI Overviews and AI Mode scans), not on the traffic side. If Google ever adds an AI value to the Search Console API, the Search Appearance dimension in a Search Console widget will show it without a change on our part.

The fastest route is the "GA4 — AI Assistant Traffic" template: in the report builder, choose Google Analytics as the source and pick that template. It comes pre-built with a per-assistant breakdown table and two donut charts (sessions and key events by assistant), plus an AI-written summary. To add it to an existing report, drop in a Table or Breakdown (donut) widget on a GA4 data source and set its dimension to "AI Source". You can pair any GA4 metrics with it — Sessions, Users, Engaged Sessions, Engagement Rate, Key Events or Revenue — to show both the volume and the quality of AI-driven traffic. Each assistant gets one row, whatever GA4 recorded it under: gemini.google.com and the gemini tag both read as Gemini, chatgpt.com and the openai tag both read as ChatGPT. GA4 calculates that row over all of the assistant's sessions, so Users and rates such as Engagement Rate are GA4's own figures, never added up or averaged by us. A source counted only because GA4 marked it ai-assistant keeps the name GA4 gave it.

For a shop, the "GA4 — AI Assistant Traffic (Shop)" template shows what AI assistants bring in sales. Its table has one row per assistant with Sessions, Engaged Sessions, Purchases, Purchase Revenue, Purchase Conversion Rate and Purchase Revenue / Session, each against the report's comparison period; a row of KPIs gives the same figures for all AI assistants together, and two donuts split sessions and purchase revenue by assistant. It ends with a note on payment sites (below). Purchases are GA4's ecommerce purchases (its purchase events; in-app purchases and subscriptions are not counted), and Purchase Revenue is GA4's purchase revenue, net of refunds. Both come from the purchase events the shop sends to GA4; Shopify's Google & YouTube app sends them once GA4 is connected to it. Revenue is in the currency the GA4 property reports in, which the report names under the table. GA4 converts purchases made in other currencies at the previous day's rate; the report itself converts nothing and never adds amounts in different currencies. The two rates are worked out per row from that row's own totals: Purchase Conversion Rate is purchases ÷ sessions and Purchase Revenue / Session is purchase revenue ÷ sessions, one division each, never an average of other rates. The sessions they divide by are in the Sessions column, and a KPI tile says "Based on N sessions" under the rate. A session with two purchases counts twice in the conversion rate. If the GA4 property recorded no purchase at all in the 13 months up to the end of the period, the purchase figures say "No purchase data in this GA4 property" instead of 0: a 0 would read as "AI assistants sold nothing" when the shop sends GA4 no purchases at all. A shop that does record purchases shows a real 0 when nothing was sold through an assistant in the period, even in a week without any sale, with the sessions the 0 is counted over. The comparison period is judged the same way and is not compared when it has no purchase data. GA4 is asked about the property's purchase days once every few hours, not for every widget. Payment sites can hide the AI source. When a shopper pays through PayPal, Klarna, Stripe or Shop Pay, or finishes on Shopify's checkout, and then returns to the shop, GA4 can count the purchase as a visit from that payment site instead of from the assistant that sent the shopper. Adding paypal.com, klarna.com, stripe.com, shop.app, pay.shopify.com and checkout.shopify.com in GA4 under Admin → Data streams → Web → your web stream → Configure tag settings → Show all → List unwanted referrals (one condition each, match type "Referral domain contains") stops the return from the payment site from counting as a new visit from it. The setting applies from the moment it is saved; earlier data stays as it was.

Look at three things together. Volume (Sessions, Users) shows how much traffic AI is sending. Engagement (Engaged Sessions, Engagement Rate) shows how much of that traffic engages. Outcomes (Key Events, Revenue) show whether it converts. Comparing the same report period over period shows whether AI-referred traffic and conversions grew; set next to AI visibility, it shows whether the two moved together, not that one caused the other.

The AI → Conversions tab inside a client's AI Visibility project puts the brand's AI-search visibility trend right next to the sessions, conversions and revenue that AI assistants sent to the site (from GA4), on the same timeline. You get headline totals (AI-driven sessions, conversions, revenue over the last 90 days), the AI-traffic conversion rate benchmarked against your site-wide rate — so you can see whether AI referrals convert better or worse than your average visitor, with a ± band and a small-sample flag so a thin sample never reads as a hard number — a dual-axis chart overlaying visibility against the metric you pick, and a correlation coefficient. The coefficient appears from 20 paired days (days with both a scan and GA4 data) with its 95% interval; below that the tab says how many paired days there are, that it needs 20, and what the project's scan schedule needs to get there: a weekly schedule fits about 13 scan days into the 90-day window, so the correlation needs scans every 4 days or more often (daily scans reach 20 in about three weeks). A strength (weak, moderate, strong) is named only when the interval excludes 0, and the interval allows for both series carrying over from one day to the next, so two slow trends that happen to coincide don't read as a firm link. This is deliberately honest. The AI-driven figures count only GA4 sessions from a known AI assistant (ChatGPT, Perplexity, Gemini, Copilot and more, identified as described above), which is a lower bound — GA4 attributes some AI referrals to Direct, so the true number is higher. Visibility carries forward between scans so the line stays readable, but the correlation is computed only on real scan days, and we say plainly what it is: it shows whether the two moved together, not why. It needs GA4 connected for the client and some scan history to populate. It puts AI visibility and what AI assistants send to the site on one timeline — something pure AI-visibility tools can't show, because they have no analytics. It's a report widget too, not just a dashboard. Add the "AI → Conversions" source to any report for a KPI band (AI visibility, AI-driven sessions, conversions, revenue, and the correlation) and the visibility-vs-conversions overlay chart — both render in the live report, the shared link and the PDF. Or start from the "AI Visibility — Complete Report" template, which stitches the whole picture together: brand performance, competitors, per-engine breakdown, topics and cited sources, source health, GEO readiness, shopping visibility, and the AI → Conversions section — so a client sees, in one report, both how visible they are in AI answers and the revenue AI assistants sent to the site.