The first question every operator asks about AI search is how much traffic it is sending. GA4 now looks like it answers that, because there is a channel called AI Assistant sitting in the default channel group. It answers a narrower question than it appears to.
What the channel definition actually says
Google's documentation defines AI Assistant as the channel by which users arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok. Then it adds one sentence that changes how the whole report should be read: it excludes Google's AI Overviews and AI Mode.
Those are in Organic Search. The same documentation defines Organic Search as arrivals via non-ad links in organic search results, including Google's AI Overviews and AI Mode.
That split is defensible. A click from an AI Overview is a click from a Google search results page, and treating it as organic keeps a long time series intact. But it means the number labelled AI Assistant is not your AI traffic. It is your non-Google assistant traffic, and the Google half is invisible inside a much larger organic figure.
The three ways this gets misread
First, someone reports that AI traffic is small. It usually is small in that channel, because the largest AI surface most businesses are exposed to is AI Overviews, and those clicks are not in it.
Second, someone reports that organic is holding up fine. It may be holding up because AI Overview clicks are propping up the same bucket that ordinary blue-link clicks are falling out of. Two opposite movements inside one number look like stability.
Third, someone builds a dashboard on the named list and assumes it is complete. Google's wording is sources like, followed by five names. Perplexity is not among them. If Perplexity matters in your category, find out where its sessions are actually landing in your property rather than assuming the channel caught them.
The traffic you cannot see at all
Channel assignment depends on a referrer arriving with the session. A meaningful share of assistant traffic does not carry one, because of how links are opened from desktop apps, mobile apps and copied URLs. Those sessions land in Direct.
We are not going to put a percentage on that, because the published figures for it are vendor estimates with methods that are not comparable to each other. What can be said is directional and checkable in your own data: if Direct sessions have grown without a corresponding brand-search or campaign explanation, some of that is very likely assistant traffic.
How to set this up so it tells you something
Keep the default channel group for continuity, and build a custom channel group beside it that reflects the question you actually have. If you want to know how much of your acquisition originates in an AI answer, you need Organic Search split by whether Search Console reports generative AI impressions for the landing page, plus the assistant channel, and you have to accept that the join is approximate.
Add a referrer-based segment for the assistants that matter in your category rather than relying on the named list.
Then judge all of it on outcomes, not sessions. Assistant traffic tends to be low in volume and late in the decision, which means a session comparison flatters ordinary organic and a qualified-lead comparison often does the opposite. The bucket that produces five leads from two hundred sessions is the one worth understanding.
What this means for an operator
Stop reporting a single AI traffic number. Report the assistant channel and the Google generative surface separately, judge both on qualified leads, and check whether Direct has been quietly absorbing the difference.