Google gave you a channel. It did not give you a measurement.
On May 13, 2026, Google Analytics added a native AI Assistant channel to the GA4 Default Channel Group, with broad availability across properties reached around June 7, 2026 (Google Analytics Help, What's new; Digital Applied). Marketing directors are already pulling this line into board decks as the answer to "how much AI traffic are we getting." It is not that answer. It is a referrer allowlist wearing the costume of a metric, and treating it as settled data is the riskiest move available right now.
The incumbent way
Most teams will open GA4, see a new AI Assistant row in the Default Channel Group report, and read it as the AI traffic number. Google's own description encourages exactly that reading: the channel exists so marketers can "identify how users are discovering your site through chatbots like ChatGPT, Gemini, and Claude" (Google Analytics Help). The mechanic behind that sentence is narrow. A session gets tagged medium ai-assistant and rolled into the AI Assistant channel only when its referrer matches a specific list Google maintains (Google Analytics Help, Default channel group). Default channel groups cannot be edited (Google Analytics Help), so whatever Google puts on that list is the report, full stop, until you build something else.
That is the incumbent way: take the new row at face value, drop it into the quarterly deck next to Organic and Paid, and move on.
The cost
The cost shows up as soon as you ask where the rest of your AI traffic went, because it did not disappear. It scattered.
Perplexity is not on Google's recognized list for the native channel and its visits still land in Referral (Digital Applied). Google's own custom channel documentation includes Perplexity in its example regex for a self built AI channel (Google Analytics Help, Custom channel groups). Meanwhile Google's own AI surfaces, AI Overviews and AI Mode, are filed under Organic Search by design, not under AI Assistant ("Organic Search is the channel by which users arrive at your site/app via non-ad links in organic-search results, including Google's AI Overviews and AI Mode") (Google Analytics Help). And any AI session that arrives without a referrer GA4 recognizes cannot match the rule at all, so it falls into Direct or Unassigned instead of AI Assistant (Madx Digital).
One buyer action, a person who found you through an AI assistant, now has four possible homes in your reporting: AI Assistant, Referral, Organic Search, or Direct. A single line item cannot be the total when the same behavior is scattered across the report by design.
The timing compounds the damage. The channel is forward only. GA4 does not retroactively reclassify historical data, so any period over period comparison that straddles the May 13 launch, or the gradual rollout that continued toward June 7, is comparing a metric to its own absence, not measuring growth (Digital Applied). A line that jumps from zero to something in a mid period chart looks like a spike. It is a measurement event, not a traffic event.
What the evidence supports
The qualitative case for undercounting is direct and well supported. Madx Digital states plainly that the channel "only captures AI visits that arrive with a recognised referrer, so it undercounts AI's real influence" (Madx Digital). Credible practitioner sources agree on the direction of the problem and disagree sharply on its size. No defensible universal percentage exists across the sources reviewed here, and printing one would manufacture false precision. The honest summary stays qualitative: treat the channel number as a floor, not a ceiling (Digital Applied).
There is also a genuine open question the industry has not settled, and pretending otherwise would be dishonest. Google's own documentation for custom and primary channel groups states that a new grouping applies going forward: "Data will populate using the new criteria from that point forward" (Google Analytics Help, Custom channel groups). At least one independent practitioner source claims the opposite for a channel group built directly in the reporting UI, stating it "applies to historical data, so you see past AI traffic reclassified straight away" (Madx Digital). These two claims conflict. Nobody outside Google has published a controlled test that resolves it. The correct move is not to pick a side. The correct move is to validate retroactive behavior inside your own property before you rely on it for anything.
The measurement playbook
Build a custom channel group in Admin, under Data display, Channel groups. Add an AI channel using a matching regex that includes the sources Google's native list omits, most notably Perplexity, and reorder it above Referrals so it captures those sessions before the Referral rule does (Google Analytics Help, Custom channel groups).
Keep the native AI Assistant channel active as a controlled subset rather than replacing it. Cross check what you see through three separate views: source and medium, the channel group itself, and landing pages, then add a comparison period such as the trailing 28 days against the prior 28 to catch drift (Madx Digital). Annotate the launch date in your reporting so nobody on the team mistakes the new line for a performance spike (Digital Applied). Document your dark traffic assumption in writing, including which sources informed it and why, so the number carries its caveat wherever it travels. Test whether your custom group reclassifies history in your own account rather than assuming either Google's forward only language or a third party's retroactive claim applies to your setup.
Treat the early numbers as directional. They are a useful signal, not a complete measurement model (Madx Digital).
The binary
That is not an attribution report. That is an unvalidated allowlist with a chart attached.
You have two paths. Report the native AI Assistant number as-is and let a board decision ride on a figure that omits at least one major AI source, scattered across three other channels, and distorted by its own rollout date. Or rebuild the definition first, validate it against your own property, label its floor, and only then let it move a dollar.
The verdict
The teams that win the AI traffic conversation in 2026 will not be the ones who found the number in GA4. They will be the ones who rebuilt it, labeled its floor, and only then let it move a dollar.
Budget decisions built on a platform's default label instead of a validated data layer are the same last click mythology attribution has always produced, just with a new channel name attached. Reconciling AI traffic against one clean, validated data layer, instead of whichever report looks most official this quarter, is what Magnet's Attribution and Measurement work is built to do.
Sources
- Google Analytics Help, What's new in Google Analytics
- Google Analytics Help, Default channel group definitions
- Google Analytics Help, Custom channel groups
- Digital Applied, GA4 AI Assistant channel playbook
- Madx Digital, GA4 launches AI Assistant channel
