Measure AI Traffic in GA4: What You Can See in 2026

A new row suddenly appears in GA4: “AI Assistant.” At the same time, your company seems to be mentioned in ChatGPT more often than the small number in the report suggests. Is the tracking wrong?

Not necessarily. Since May 2026, Google Analytics has automatically measured some visits from AI assistants. The new channel is a useful starting point, but it is not a complete measure of your AI visibility. This guide shows where to find the data, which visits are missing, and which numbers belong in a credible report.

What the New AI Assistant Channel Measures

Google introduced the “AI Assistant” default channel on May 13, 2026. When a visit to your website carries a recognized referrer, GA4 automatically sets the medium to ai-assistant, the campaign to (ai-assistant), and assigns the session to the new channel. The channel definition also applies to traffic whose medium has been set manually to ai-assistant. Automatic recognition requires no additional tracking script or custom regex.

In its current Default channel group definition, Google names ChatGPT, Gemini, DeepSeek, Copilot, and Grok among its examples. Google maintains the list of recognized sources. That list can change as new providers appear or existing platforms start using different domains.

A session in this channel shows a specific GA4 classification: the visit either carried a recognized AI referrer or used the ai-assistant medium. You can then analyze it like any other session, including its landing page, engagement, and triggered key events.

The number does not tell you how often your company appeared in an answer. A recommendation without a click leaves no session behind. The channel measures traffic, not overall visibility across ChatGPT and other AI systems.

Where to Find AI Traffic in GA4

The simplest route in GA4 is Reports → Acquisition → Traffic acquisition. Use Session default channel group as the primary dimension. “AI Assistant” should appear as a separate row once GA4 has recognized at least one matching session.

Four steps are enough for an initial analysis:

1. Start with a date range beginning on or after May 13, 2026.
2. Filter the channel group for “AI Assistant.”
3. Add Session source as a secondary dimension to distinguish individual platforms.
4. Compare sessions, engaged sessions, and key events. Add Landing page + query string as a secondary dimension.

The start date matters. Google introduced the channel on May 13 and does not clearly state whether older sessions are classified completely under the new definition. A comparison between April and June is therefore not automatically reliable. Mark the launch in your reporting and build a new baseline from that date.

On June 11, 2026, Google also introduced the Source Group dimension. It standardizes different variants of source names and includes groupings for platforms such as ChatGPT and Perplexity. According to the release note, Google populates this dimension retroactively, so it can also group historical data. It is not the same thing as the session-based “AI Assistant” default channel. Do not combine the two values in one time series without explaining the difference.

If there is no AI Assistant data in your property, that does not automatically mean nobody arrived through an AI answer. The visit may simply have lacked a recognized referrer. That is where the most important limitation begins.

What GA4 Still Cannot See

The name “AI Assistant” sounds more comprehensive than the measurement really is. Four gaps should be stated clearly in every analysis.

Google AI Overviews and AI Mode

Clicks from Google AI Overviews and AI Mode do not appear in the “AI Assistant” channel. Google continues to classify them as Organic Search. In GA4, you therefore cannot reliably tell whether an organic session came from a traditional search result, an AI Overview, or AI Mode.

This matters when you want to assess the business impact of your complete AI visibility. Some AI-mediated visits are hidden inside another channel.

Visits Without a Recognized Referrer

The automatic classification relies on the referrer URL. If an app, browser, redirect, or privacy feature does not transmit this information and no campaign parameters are present, GA4 cannot identify the origin reliably. The visit may then appear as Direct or under a different source.

Sources That Are Not Yet Classified

Google's list is not a definitive directory of every AI product. New providers, changed domains, and different link paths may cause relevant sessions to appear under Referral at first. You should not assume that today's coverage will remain unchanged next year.

Mentions Without a Click

An AI system can recommend your company, quote a price, or summarize a statement from your article without the user opening your website. That impact is invisible in every traffic report. Even ten accurate brand mentions produce zero GA4 sessions if nobody clicks.

The key sentence for internal reporting is therefore: The GA4 number is a conservative but imperfect measurement, not a complete reach metric. I explain how to complement it by testing AI visibility with a fixed set of questions in my GEO guide.

How to Build Credible AI Traffic Reporting

A useful report does not need to show every available GA4 column. For small and mid-sized businesses, three levels are enough when they are measured consistently.

1. Reach

Show recognized AI sessions by source and landing page. This answers two practical questions: which assistants actually send visitors, and which content brings them into the site?

Landing pages often tell you more than the total. If a detailed guide regularly attracts users from ChatGPT, that is a concrete signal about the type of content being cited and clicked. A general increase with no identifiable destination provides much less direction for the next content decision.

2. Quality

Review engaged sessions, returning users, and relevant next pages. Does someone read one paragraph and leave? Or do they continue to a service page, case study, or contact form?

AI-referred traffic is often small at first, so daily figures can fluctuate heavily. A monthly comparison is usually more useful than a daily dashboard. Always show absolute numbers as well: 200 percent growth may simply mean an increase from one session to three.

3. Business Value

Meaningful key events are what matter: enquiries, bookings, qualified downloads, or other actions with genuine business value. Before interpreting the results, verify that these events are configured correctly. Ten contact-form views do not make an AI channel valuable if nobody successfully submits the form.

For links you control, such as those in your own GPT, newsletter, or partner campaign, use UTM parameters consistently. UTMs do not reconstruct the HTTP referrer, but they can transmit an explicit source, medium, and campaign even when it is missing. They cannot retroactively tag past clicks or links in third-party ChatGPT answers.

That is why GA4 data should be paired with repeatable visibility checks: the same questions, the same systems, a documented date, and recorded sources. This allows you to monitor whether your company appears at all as well as whether users click. It is not a perfect reach measurement, but it creates a traceable trend instead of an occasional spot check.

When a Custom Channel Group Makes Sense

A custom channel group is useful when you repeatedly find relevant AI referrals in your session sources that appear under Referral or another channel. It can also help when your internal reporting deliberately covers more providers than Google's standard classification.

GA4 supports custom channel groups for this purpose. You copy an existing group, add a rule-based channel, and set its priority in the evaluation order. That order matters: a session is assigned to the first channel whose rules it matches. Your custom AI channel therefore needs to be evaluated before a general referral channel. These rules are also applied retroactively to existing data in reports, explorations, and segments.

Do not paste a supposedly complete regex from an old blog post into your property and forget about it. Providers, domains, and Google's own classifications change. Inspect your current Session source values first, document the rules, and review them regularly.

A custom group only improves the classification of data that already exists. It cannot reconstruct a missing referrer, measure mentions without clicks, or separate Google AI Overviews cleanly from Organic Search.

Which Number Should You Show Management?

A single ROI figure for AI visibility would be convenient, but it would imply more certainty than the data can support. A compact monthly snapshot with four elements is more useful:

• recognized AI sessions, with absolute numbers and month-over-month changes
• the leading AI landing pages
• key events generated by those sessions
• documented brand and source mentions for a fixed set of test questions

Evaluate the trend over at least 90 days. At low volumes, a single link can double that month's session count and disappear again in the next. Only when traffic or documented visibility rises repeatedly, and the landing pages matter to the business, can you turn the signal into a priority for further content or technical optimization.

The report should support a decision, not produce the largest possible number. Which topics attract qualified visitors? Which pages contribute to enquiries? Where does the brand appear but fail to earn a click? These questions are more valuable than an isolated percentage.

Conclusion

GA4 made AI traffic more measurable in 2026, not completely measurable. Without requiring a custom setup, the new “AI Assistant” channel shows sessions with recognized referrers from platforms such as ChatGPT, Gemini, and Copilot, as well as traffic tagged manually with ai-assistant. It is a practical basis for analyzing landing pages, engagement, and key events.

Looking at that one row alone will still undercount traffic from Google AI Overviews, no-click mentions, and visits without a clear referrer. Check the channel now, validate your key events, and build a 90-day baseline combining traffic, business value, and repeatable visibility tests.

This article was drafted and translated with AI assistance.