AI search, SEO & online discovery
Practical coverage from Dubai

Saved
Measurement

How to Track AI Referral Traffic in GA4

Summarize with AI
Data analysis on a computer; illustrative photograph.
Photo by Lukas Blazek / PexelsIllustrative photography
In this article 6 sections

Start in GA4's Traffic acquisition report. Look for the AI Assistant row in the session default channel group, then inspect the source and medium behind the sessions. If you need to include verified older referral values or build a reporting series under your own rules, keep that series separate and date its definition. That gives your team two usable views. One follows Google's channel classification. The other explains the sources your team can consistently identify in its own property.

The result is a count of tracked sessions with an identifiable acquisition signal. A customer can see an AI answer, search for the business later and arrive through another channel. The answer could also provide a phone number and produce no website visit. A GA4 traffic report addresses the part of that journey that produced a recorded session.

Start with the session report and Google's current channel

Open Reports, then Acquisition, then Traffic acquisition where the standard report is available. Set the reporting period and confirm the property and stream scope. Select Session default channel group to find the AI Assistant row. Then view Session source / medium, using it as a secondary dimension or inspecting the relevant rows in a separate view. Keep the date range and property filters consistent when reconciling the counts. Google Analytics Traffic acquisition report

Google announced the AI Assistant channel in May 2026. Its current documentation says GA4 assigns the medium ai-assistant and campaign (ai-assistant) when the referrer matches its recognized assistant list. Google excludes AI Overviews and AI Mode from the AI Assistant channel. The presence of the new label is therefore a useful starting point for assistant traffic reporting, while clicks from Google's search experiences need their own evidence. An empty row or a change in the trend needs to be interpreted against the property, period and data available. GA4 release notes · GA4 default channel definitions

Stay at session scope for this question. First user source / medium describes acquisition of the user, which may predate the session you are reviewing. Event-level channel attribution can answer a different question about credit for a key event. If a returning visitor was first acquired through organic search and later visits from a recognized assistant, a session report and a first-user report can put the activity under different sources. Label the scope in any exported table.

A product announcement also has a narrower use than an acquisition signal. The official Google Maps Instagram post below introduces a conversational Maps feature. Use it to identify the feature being discussed. If a site visit follows an interaction with that product, a property-specific acquisition record is still needed to support a GA4 classification.

Make a dated inventory of the source values you can see

Export or record session source, session medium, session default channel group and sessions for the reporting window. Include landing page when it helps explain which parts of the site receive traffic. Keep a total at the report's original scope. A landing-page breakdown and a source summary can have different row structures, so reconcile at the intended reporting level before using them together.

A source value is a value GA4 recorded. It may reflect referral information or campaign parameters, depending on how the visit arrived and how the data was collected. Keep the original values in the inventory. For each source you want to add to a custom grouping, record the exact source and medium combination, first and last observed dates, which rule it matches, and why it is included. A generic organic source or direct session does not identify a specific assistant. Keep those rows visible in the total traffic figures and outside the identifiable assistant count.

Distinguish a recognized referral from a tag you control. If your team uses an assistant-like name as a campaign tag in a newsletter, it can appear as a source string even though the sending channel is email. A broad match on text such as “chat” can similarly pick up unrelated sites or labels. Review the medium and campaign when they help distinguish traffic types. Record exceptions with enough detail to reproduce why a value was excluded.

Use evidence from your own property before adding source variants. A link from a documented placement or a controlled tagging check can help establish what reaches your site. A controlled test is a separate test observation. If you use one, keep its date and test label so it does not inflate customer traffic. Do not infer that every device, browser, app or redirect path sends the same signal because one path did.

Build a custom group only if you need a second definition

A custom channel group is useful when you want a stable, documented operational view that includes specific older referral rows your team has verified. Google's Analytics Help describes building channel rules from traffic-source fields, ordering channels and selecting the custom group in supported reports. Keep the standard channel available as your reference so you can reconcile the result after the change. GA4 custom channel groups

For a simple custom definition, include sessions whose medium exactly matches ai-assistant, plus reviewed exact sources where the medium is referral. The second condition is optional and depends on your records. Write the Boolean rule clearly. A match on the first condition qualifies on its own. A match on the second requires both an allowed source and the referral medium. The classifier in this article is illustrative. Check the resulting rows in your property before applying it to a dashboard.

Position the custom channel above broader categories that would otherwise capture the intended rows, particularly Referral for a historical referral rule. Review competing rules before changing the order. If a custom group's paid or email channels are important to your organization, check that the assistant rule leaves those rows where the report owner expects. Compare the output with the source and medium inventory, and save a rule version, date, approver and sample of included and excluded rows.

Apply the same definition to both months when a trend is meant to be comparable. A custom group can be applied to historical reporting data in supported reports, but the grouping can work only with fields that were collected. It cannot restore a missing signal. A useful change note states whether you recomputed past periods under the new rule or left the previously published definition in place. Label the dates and definitions so a chart does not present a classification change as a traffic increase.

Before sharing a new channel, save a fixed date range and compare its membership before and after the rule change. Review any shift in the broader Referral, Organic Search or Email rows as well as the assistant total. A change can be arithmetically correct while moving a source the team intended to leave alone. Test a known inclusion and a known exclusion before refreshing the recurring dashboard.

Reconcile the standard and custom counts

The following table uses constructed source and medium rows. It shows how a verified historical referral rule could differ from the default channel. In the fixture, assistant.example stands for an exact, reviewed value. It is a reserved illustrative domain. It is not a source to enter in GA4.

Session source / medium Sessions Default AI Assistant in fixture Custom identifiable assistant
chatgpt.com / ai-assistant 18 18 18
assistant.example / referral 7 0 7
assistant.example / email 4 0 0
unrelated-assistant.example / referral 6 0 0
google / organic 61 0 0
(direct) / (none) 104 0 0
Total 200 18 25

The standard row reports 18 sessions under the fixture assumptions. The custom view reports 25 sessions, which includes those same 18. The seven-session difference comes from one verified referral row. Adding the standard and custom totals together would count some visits twice. An exact comparison also keeps the email row and the lookalike referral out of the custom view.

If you publish a share, give the reader the total population. In the example, the custom share is 25 of 200 sessions, or 12.5%, for the same property and reporting filters. It is possible to build a narrower view for a site section or country. Keep the numerator and denominator under the same filters and date range. When data is missing or unavailable for a segment, state the coverage limit before reading the share as an audience comparison.

Separate traffic counts from contact outcomes

A session with a click on a WhatsApp link can show interest, while the click itself does not confirm a message was sent. A submitted form can be a stronger signal if the event fires only after a successful submission. Give each event a name that reflects what it measures and inspect duplicate or misfiring events before making a performance comparison. If you want a session rate, use sessions with the event as the numerator and eligible sessions as the denominator. A count of events can be higher because a visitor may trigger the event several times in one session.

For example, if the 25 constructed assistant sessions generated six contact-click events across four sessions, report six events and four of 25 sessions with a click. Those are different units. A later tally of confirmed enquiries comes from the customer contact process and may not match those numbers. The existing AI Visibility referral-to-enquiry guide covers that handoff for forms, phone and WhatsApp. AI Visibility referral-to-enquiry guide

Keep customer-reported discovery separate from measured sessions when no reliable connection exists. Someone who says they found the business through an assistant can be recorded as self-reported even if their site session is classified differently. Likewise, GA4 may record an assistant session with no subsequent contact. In a monthly report, a session metric, a customer-reported count and a confirmed enquiry count each have different evidence. The report can place them together and name their scope. AI Visibility monthly reporting guide

Save a reporting definition another analyst can reproduce

Record the property, report dates, timezone, filters, dimension scope, rule version and snapshot date. Include the standard channel total, the custom total if used, and the difference explained by source and medium. Keep the approved exact-match list and any exclusion examples with the export. If new source values appear, review them against evidence before bringing them into the custom group. Where a source no longer appears, preserve its historical mapping if you still rely on that definition for earlier months.

GA4 assistant traffic record

Report the number of identifiable assistant sessions and the number of sessions in scope. Specify the channel definition and date, include a separate line for each measured contact event, and record any change to tagging, grouping or filters. Unknown acquisition signals should remain unknown in the source record. This gives the next analyst a concrete way to check whether the underlying traffic changed or the definition changed.

AI Visibility is published by Lunasol, which offers GEO and AEO services. The table and contact figures are constructed examples. About AI Visibility

You’ve reached the end of this article. Read as plain text

Search the blog

Summarize with AI

Choose your assistant. Copy the article and summary request, then paste it into a new chat.

The full article is included, so your assistant can read it without opening this website. Nothing is submitted automatically.

View or copy the summary requestOpen assistant

Summaries are generated by your chosen service and may miss details. Its account requirements and terms apply.