I was digging into a Google Analytics 4 account recently, breaking down traffic by source and medium, and I saw something that made me stop scrolling. Same source, chatgpt.comit was sitting on three different channels at the same time. They are not three different sources. One source, spread over three channels, in the same report. If you have a new GA4 AI assistant channel, chances are the same thing is happening to your data right now. And it means that the AI traffic number you’re reporting is probably wrong.
AI traffic is still a small slice of most sites, but it’s punching above its weight: Similarweb’s clickstream data has ChatGPT conversions at around 7%, ahead of organic search and far behind paid. A channel with a small high target should be measured properly rather than eyeballing it. Let me show you why it breaks, and how to fix it.
So What Has Google Really Changed?
On May 13, 2026, Google added a native AI Assistant channel to GA4’s Default Channel Group. The idea is simple. When GA4 recognizes the referrer as an AI assistant, it marks the session as medium ai-assistantdownload it to the AI Assistant channel, and stamp the campaign with (ai-assistant). No setup, no regex, nothing to do. It started slowly and reached many places in early June 2026.
If you’ve spent the last year cobbling together custom regex just to see your AI traffic, that’s a real win. Prior to this, those visits were always Referral, or Direct when the referrer was not available.
But here’s the catch, and it’s hidden in the word “you see.” The list of platforms that Google recognizes goes on and on. When it was launched, it named ChatGPT, Gemini, and Claude. In June, the live documents listed a different set (ChatGPT, Gemini, DeepSeek, Copilot, and Grok), Claude quietly stepped down. Confusion, one of the most highly targeted AI sources, is still unlisted and keeps coming up on Referral. So don’t hard code the field list into a client report. Check the Google channel descriptions on the day you publish, because they change.
1 Source, 3 Channels: Problem Hiding in Your Report
Go back to that screenshot. The reason that one source separates the three channels is that the GA4 determines the channel using the source again medium together, not the source alone. Add Session source/central like size, and you can watch chatgpt.com split three times:
chatgpt.com / ai-assistantstays in the middle AI assistant. This is a GA4 piece spotted and tagged.chatgpt.com / referralstays in the middle Transfer. These are the times that GA4 can tag, as well as anything that came before the channel opened for your area (remember, the release was pushed to June).chatgpt.com / (not set)stays in the middle Not providedchannel almost no one ever opens. Google’s own rule is not clear here: if the source/medium appears as (not set), there is no channel rule to catch it, so it falls to Unassigned.
Why did a well-known source appear without a producer? As far as I know the most common is the ChatGPT app and its in-app browser. As MarTech has written, links opened within those embedded browsers tend to strip the transmitter, so GA4 clings to the source but loses the medium.
So your real ChatGPT number is not in one place. It’s spread across three channels, and one of them is a bucket you probably never check. Learn the AI assistant channel yourself, and you’ll count less every time.
Why Obvious Fixes Don’t Work
There are three natural solutions here, and I won’t depend on any of them:
- Just read the AI assistant channel. It misses the Referral and Unshared pieces of the same traffic, and ignores Perplexity entirely.
- Compare this month with last. The native channel only counts forward from the release date, and airs at different times on various features. So any range going back to spring 2026 compares tagged traffic versus untagged traffic. That is not a habit; an artifact.
- Check your follow rate. It’s a completely different question. The rankings tell you about the position, not whether the contributor actually sent someone to your site.
Fix: 1 Custom Channel Same as Source
Here’s a move that really solves it. Create a custom channel group and match the source while completely ignoring the medium. The second time you do that, the ai-assistant, referral again (not set) versions of chatgpt.com die in one line.
You get two bonuses at the top. A custom channel group applies its rules repeatedly throughout your date range, so it rescues all those old ChatGPT sessions stuck in Referral. And because you write the rule, you can include fields that Google excludes, including Perplexity.
Here’s how I set it up:
1. Go to Manager > Data display > Channel groups and create a new group.
2. Add a channel and call it AI.
3. Set the condition to Source matches a regex, with a pattern that includes AI domains.
4. Drag the AI channel above Referral and Organic, so that it searches for those times first.
5. Save, and use the group as your dimension in any discovery report.
Regex is where you can hurt yourself in peace, so build carefully. Keep entries associated with virtual domains or service-specific host tokens. Never throw a blank token gpt itself, because it will match any source that might contain those three characters and drag out false positives, which is exactly the kind of thing that makes your data fragmented. Here’s a border-conscious pattern covering major AI sources you might want to follow:
.*(^|[/.:@?&=])(chatgpt.com|chat-gpt.org|openai.com|perplexity|gemini.google.com|copilot.microsoft.com|edgepilot|edgeservices|claude.ai|deepseek.com|grok.com|you.com|nimble.ai|iask.ai|aitastic.app|bnngpt.com|writesonic.com|copy.ai)([/.:@?=]|$).*Two quick notes on that. gemini.google.com it’s a specific host, so it won’t cover all of your Google stuff, so never add anything empty. google in the list. And treat the pattern as something perishable. I would update it quarterly, because platforms come and go and their domains may change.
A Bit Even a Perfect Channel Can’t Fix
This is the part that many directors skip, and it’s the part that keeps you honest. A custom channel group adjusts traffic patterns is dividednot even if it finds collected in the first place. A source rule can only capture instances that appear from a source that you can actually read.
The biggest blind spot is AI traffic with no referrer at all, which is always Direct. Most of that comes from mobile AI apps and in-app browsers, which don’t pass anything to GA4 to read, so there’s no source to match.
In addition, Google’s AI Overview and AI mode are calculated as Organic Search (google / organic), and Google intentionally keeps them out of the AI Assistant channel. On most sites, that’s one big area of AI there is, and it doesn’t show up as “AI” in your reports. Don’t try to drag it into your AI channel either, or you’ll accidentally swallow Google’s regular organic.
So, be very specific about what you want. With this custom channel in place, you have a complete and consistent number of AI sources to target. What you don’t have is a measure of the full impact of AI. That the assistant recommended you to someone who has never clicked is a completely different problem, and no channel team will solve it.
What I will do this week
Treat the native AI Assistant channel as a starting point, not a finished product. Then:
- Add a source/intermediate to the Traffic discovery report and you’ll find your AI domains spread across the AI Assistant, Referrals, and Unassigned. Honestly, experiencing the breakup yourself is part of the lesson.
- Build a source-based AI custom channel group to pull the pieces together and restore the history that the native channel left behind.
- Write down the monthly basis. With a moving platform list and a big black piece you won’t see, the month-to-month trend will tell you more than any single article.
Google finally putting AI traffic on the dashboard is a real step forward. Just remember that the default view gives you a simple slice. The real work now is to know exactly which pieces are left out silently.
Additional resources:
Featured image: Roman Samborskii/Shutterstock
