GA4 quietly shipped a new default channel in May 2026 called AI Assistant. Traffic from ChatGPT, Gemini and Microsoft Copilot now gets its own row in the traffic acquisition reports instead of drowning in Referral or Direct. Within a week of the rollout people started sending me the news with some version of "so this is handled now, right?"
It's a real improvement and I'm glad it exists. It also covers only part of the problem, and because the new row looks official and complete, it invites you to stop asking questions right at the point where the interesting ones start. This article goes through what the native channel actually catches and the gaps it leaves open. Then what I build on top for clients who make decisions based on this data.
The mechanism is simple. When a session arrives with a referrer that matches Google's internal list of AI platforms, GA4 assigns it to the AI Assistant channel at processing time. ChatGPT, Gemini and Copilot are on that list. Before May those sessions showed up as Referral when the referrer survived the click, or Direct when it didn't.
Note what's happening there. It's a labeling rule on top of the same referrer data GA4 was already collecting, and there's no new signal involved. The sessions that were invisible before the channel launched are exactly as invisible now.
The channel started classifying traffic on the day it reached your property. Everything before that date keeps its original channel, because GA4 doesn't reprocess history when default channel definitions change. So if you want to answer a question like "how much has our ChatGPT traffic grown this year," the native channel can't help you. The first half of the year is still sitting in Referral and Direct with no label on it.
For a brand trying to decide whether AI search deserves budget, the trend line is the whole point. A channel that starts at zero on rollout day can't show you a trend.
As I write this, Perplexity isn't on Google's list. Neither is Claude, and neither is Meta AI. Perplexity is the one that hurts, because for ecommerce stores it tends to carry more purchase intent than Gemini traffic does. All of those sessions keep landing in Referral or Direct, so if you're reading the AI Assistant row as "our AI traffic," you're undercounting by whatever those engines send you, and you have no way to know the size of the miss from inside the report.
Google will presumably extend the list over time, but I wouldn't build reporting on presumably.
Clicks from Google's AI Overviews carry a google.com referrer, so GA4 files them under Organic Search along with every classic blue-link click. The AI Assistant channel doesn't claim them, and Google gives you no parameter to split them out. Nobody outside Google can separate an AI Overview click from a regular organic click with certainty, and I'd be suspicious of any tool claiming otherwise. What you can watch is directional evidence, like shifts in branded query behavior, but that slice of AI influence stays estimated no matter what you buy.
GA4 lets you build custom channel groups, and they behave differently from the default one in the way that matters most here. A custom channel group applies retroactively to all the data your property has already collected. Define a channel with conditions covering the full engine list, including Perplexity, Claude and Meta AI, plus the UTM patterns some assistants append to their links, and you can open February's report and see AI traffic labeled correctly, months before the native channel existed.
I'm deliberately not pasting matching conditions here. The engine landscape shifts monthly, hostnames get added and renamed, and a stale condition list quietly leaks traffic back into Referral without telling anyone. The concept takes a paragraph to explain. Keeping the definitions current is the actual work, and that maintenance is part of what clients pay me for.
Everything above classifies sessions that arrive with a referrer. A large share of AI assistant clicks don't. Vendor measurements put the no-referrer share anywhere from roughly a third to two thirds depending on the assistant and the platform, with in-app clicks on mobile among the worst. Those sessions land in Direct with nothing for any channel group condition to match on. I went deep on that failure mode in Your ChatGPT Traffic Is Hiding in Direct, worth reading if your Direct channel has been growing without explanation.
For clients where AI traffic matters commercially, I add a server-side layer through server-side GTM. It picks up signals a browser-only setup drops and keeps the AI source in durable first-party data, so attribution survives past the point where a normal browser cookie dies. The same source then feeds GA4 and Meta's Conversions API with proper deduplication against the browser events, which means your Meta campaigns learn from these conversions too. In practice this recovers roughly half of the AI clicks that are currently invisible. I won't promise a fixed percentage, because it depends on your traffic mix, and anyone quoting an exact number before seeing your data is guessing.
There are limits I put in writing for every client. Safari's tracking prevention still caps cookie lifetime in several common setups regardless of how the cookie is set, and long persistence takes real same-origin infrastructure on your own domain. For EU visitors everything runs behind Consent Mode v2, since a cookie set by a server is still a cookie under ePrivacy. And you can't inject tracking parameters into a URL an AI model wrote on its own, so referrer and first-party stitching stays probabilistic at the edges. The setup beats the default by a wide margin, but it has edges, and I'd distrust anyone selling it without mentioning them.
Some buyers see your brand in an AI answer, never click, and come back days later through branded search. No analytics setup attributes that per person, mine included. I measure it in aggregate with a post-purchase survey and branded search lift, and those numbers are labeled modeled in my reports.
If you run an online store and Direct has been climbing while nothing in your marketing explains it, some of that is probably AI assistant traffic filed wrong.
I'm Rytis Balys, a conversion tracking specialist working with GA4, GTM, server-side tagging and Meta CAPI. If you want to know what your AI traffic is actually worth, write to me and I'll look at your setup.
Partly. Since May 2026 GA4 files referrer-carrying clicks from ChatGPT, Gemini and Copilot under an AI Assistant channel. Clicks without a referrer still land in Direct, and data from before the rollout stays unlabeled.
Perplexity isn't on Google's platform list yet, so its sessions stay in Referral or Direct. A custom channel group with your own conditions is the current fix, and it also covers Claude and Meta AI.
The native channel can't, it only classifies traffic from rollout onward. A custom channel group applies retroactively to everything your property has already collected, so it can label months of past AI traffic.
No. They carry a google.com referrer, so GA4 keeps them inside Organic Search, and Google provides no way to separate them from classic organic clicks.
The native channel is a start, not the answer. If you want to know what your AI traffic actually looks like across the engines GA4 skips, send your numbers below and I will tell you what your property is most likely hiding, no charge. If the native channel already covers your traffic mix, I will say so.
or email work@rytisbalys.com · every rated job 5.0 · verify on Upwork