GA4 AI Assistant Traffic vs Search Console AI Visibility

“How much traffic did AI search send us?” sounds like one question. In practice, two different measurements are involved: whether Google displayed your page in an AI feature, and whether someone clicked from an AI assistant to your site. Combining them into one “GEO score” hides the useful part of the story.

Google now offers a native AI Assistant channel in GA4 for recognized assistant referrals. Search Console separately reports impressions for supported generative AI features in Google Search. This guide shows how to read both, where to find actual visits and outcomes, and what remains unmeasured.

First, decide which question you are answering

• Was our page shown in Google's AI Search features? Use Search Console's generative AI performance report. It covers supported features such as AI Overviews and AI Mode and reports impressions by page, country, device, and date.
• Did a recognized AI assistant send a session to our site? Use GA4's AI Assistant channel and inspect its underlying session sources.
• Did those visitors do something valuable? Use GA4 engagement and key-event data for that traffic, with the same caveats you would apply to any small segment.

These are related observations, but they are not sequential steps in a measured funnel. A Search Console AI impression is an appearance on Google. A GA4 AI Assistant session is a visit attributed to a recognized referring assistant. One cannot be divided by the other to calculate an AI-search CTR.

GA4: start with the native AI Assistant channel

On 13 May 2026, Google announced a default AI Assistant channel for traffic from recognized assistants. When a referrer matches one it recognizes, GA4 can assign the medium ai-assistant, the campaign (ai-assistant), and the AI Assistant channel. This is already part of GA4's default channel grouping; many older guides that begin by asking you to build a custom channel are now out of date. Google's release notes describe the change.

1. Open Reports → Acquisition → Traffic acquisition in your GA4 property. If the report is missing, an Editor can add it to the navigation.
2. Use Session default channel group as the primary dimension and look for AI Assistant. Keep the date range visible and compare it with an equivalent preceding period only when the segment has enough data.
3. Switch or add Session source / medium to see which sources are behind the row. This is more informative than treating every assistant visit as one platform.
4. Review sessions, engaged sessions, engagement rate, and key events for that segment. A key event has to be configured before its count is useful. Google's Traffic acquisition guide defines these dimensions and metrics.

To learn which content received the visits, add a page-path dimension or filter to the report. Google documents Page path and screen class as a secondary dimension or report filter. Keep the channel and page scope together: a sitewide AI Assistant total cannot tell you which article earned a visit.

Search Console: use its AI report for Google visibility

In the verified Search Console property, open Performance → Generative AI features for Search. Choose the same reporting window you used in GA4, then inspect Pages first. You can also group by country, device, and date. Google's report documentation says its core metric is impressions: links to your site shown in supported generative AI features on Google Search. It is available worldwide, although a property with too few impressions may have little to show.

Do not read that impression count as clicks, citations across every assistant, or conversions. The report does not reveal the complete prompt or query fan-out path that led to an appearance. Google Discover has a separate generative AI report. The AI feature data is also included in Search Console's ordinary Web performance data, so adding the two impression totals together would double count. Google's AI features guidance explains how these experiences fit into Search.

If Search Console shows an important page with AI impressions, ask whether its content gives readers a reason to go beyond the AI answer: original examples, a useful workflow, a comparison, or a tool. If no page shows impressions, first confirm indexability and whether the topic tends to trigger AI features. Do not assume the page needs a special file or schema. My full Search Console AI visibility guide covers those limits and content decisions in more depth.

Build one monthly worksheet, with separate columns

You do not need a new score. A simple sheet or dashboard can record these fields for each important URL and reporting period:

• Date range and canonical URL
• Search Console generative AI impressions
• Search Console ordinary Web impressions and clicks
• GA4 AI Assistant sessions for a selected page path
• GA4 engaged sessions and configured key events for that segment
• Content change made, date, and reason
• Known limitations or unusual SERP observations

Keep the sources labeled. Search Console uses its own reporting rules and time zone; GA4 has different session and attribution rules. An apparent mismatch between the products is a question to investigate, not proof that one is broken. If you need a starting point for ordinary Search Console views, see my first-five-checks guide.

Three comparisons that are useful—and three that are not

Useful: Which URLs have AI impressions this month, and did that set change after we updated the content? Which landing pages receive visits from recognized assistants? Do those visitors engage or trigger a key event? These observations help choose where to investigate next.

Not useful: Dividing GA4 AI Assistant sessions by Google AI impressions to invent a click-through rate; summing both as “AI traffic”; or claiming that a content edit caused a change from a handful of impressions. The surfaces, units, and attribution rules differ.

It is entirely possible to see Google AI impressions with no GA4 AI Assistant sessions: Google Search is not the same source as a third-party assistant referral. The reverse can also happen when visitors arrive from an assistant even though your page has no Google AI feature impressions. Small counts are especially noisy.

What remains invisible

GA4's channel only classifies visits for which it has a recognizable source. A person can see your brand in an AI answer, never click, and visit later by another route. Some visits arrive without a clear referrer and may be classified as (direct) / (none). Google lists missing traffic information, redirects, and blockers among the reasons source attribution can be incomplete in its direct-traffic guidance. Do not relabel all direct traffic as AI traffic.

Search Console's report covers Google's supported AI Search features, not ChatGPT, Claude, Perplexity, or every other assistant. GA4 can report recognized assistant visits but cannot tell you every answer that showed or cited your page. Neither tool gives a complete prompt-to-conversion trail.

When a custom GA4 channel is worth building

Google also documents an AI assistants custom channel example. You may need one if you want to group additional observed sources, maintain a different definition, or compare a curated subset. Creating it requires Editor access. Put the new channel above Referral, and above other channels if your rules require it, because the first matching channel wins.

Start by checking the actual Session source values in your property. Write narrow rules for sources you can verify and test them against unwanted matches. Google's example uses a broad regex and explicitly says to update it as assistant URLs change; copying a wide pattern without auditing it can overcount. Custom channel groups can be applied retrospectively, but they are not a substitute for the native default channel or a complete AI attribution system.

A 20-minute baseline you can repeat

1. Five minutes: capture the Search Console AI impression trend and its top canonical pages for a completed period.
2. Five minutes: open GA4 Traffic acquisition and inspect the AI Assistant channel and its Session source / medium values.
3. Five minutes: inspect relevant page paths, engaged sessions, and key events for that GA4 segment.
4. Five minutes: record one question worth answering next, such as whether a page needs a better example, a clearer title, or a more useful next step. Add a review date before making a change.

The point: measure Google AI visibility, assistant-referred visits, and business outcomes as separate signals. You can compare their direction over time, but you cannot collapse them into one trustworthy “GEO score.”