How to build one that doesn’t

I recently dug through a Google Analytics 4 account, broke down traffic by source and medium, and saw something that made me stop scrolling. The same source, chatgpt.comwas sitting on three different channels at once. Not three different sources. A source, scattered across three channels, in the same report. If you have GA4’s new AI Assistant channel, there’s a good chance the same thing is happening to your data right now. And that means the AI ​​traffic number you’re reporting is almost certainly fake.

AI traffic still only represents a small share for most sites, but it converts well above its weight: Clickstream data from Similarweb ChatGPT referrals convert at around 7%, ahead of organic search and not far behind paid search. Such a small high-intention channel deserves to be measured correctly rather than observed. Let me show you why it fragments and how to fix it.

So what has Google actually changed?

On May 13, 2026, Google added a native AI assistant channel to the GA4 default channel group. The idea is simple. When GA4 spots a referrer that it recognizes as an AI assistant, it marks the session with support ai-assistantdrops it into the AI ​​Assistant channel and marks the campaign as (ai-assistant). No configuration, no regex, nothing to build. It was deployed gradually and reached most establishments at the beginning of June 2026.

If you’ve spent the last year putting together custom regexes just to see your AI’s traffic, this is a real win. Before that, these visits were Sponsored or Direct when the referent was missing.

But here’s the catch, and it’s hidden in the word “recognizes.” The list of platforms recognized by Google continues to evolve. At launch, it named ChatGPT, Gemini and Claude. In June, the live documentation listed a different set (ChatGPT, Gemini, DeepSeek, Copilot and Grok), with Claude having been quietly abandoned. Perplexity, one of the most intentional sources of AI, is still not on the list and continues to land in Referral. So don’t hardcode a list of platforms into a customer report. Check Google Channel Definitions the day you post, as they change.

1 source, 3 channels: the problem hiding in your reports

Let’s go back to this screenshot. The reason a single source is split across three channels is that GA4 decides the channel using the source. And medium together, not source alone. Add Session source/medium as dimension, and you can look chatgpt.com divide into three:

• chatgpt.com / ai-assistant lands in AI Assistant. This is the recognized and tagged GA4 slice.
• chatgpt.com / referral lands in Reference. These are sessions that GA4 didn’t tag, plus anything that happened before the channel was activated for your property (remember, the rollout was delayed until June).
• chatgpt.com / (not set) lands in Unassignedthe channel almost no one ever opens. Google’s own rule is brutal here: when the source/medium appears as (undefined), there is no chain rule to catch itso it falls into Unassigned.

Why would a source you clearly recognize appear without support? In my experience, this is usually the ChatGPT app and its built-in browser. As MarTech has documentedlinks opened in these embedded browsers tend to drop the referrer, so GA4 clings to the source but loses support.

Your real ChatGPT number is therefore not in the same place. It’s spread across three channels, and one of them is a bucket you probably never check. Read the AI ​​Assistant channel alone and you’ll undervalue it every time.

Why Obvious Fixes Don’t Work

There are three knee-jerk solutions here, and I wouldn’t rely on any of them:

• Just read the AI ​​Assistant channel. It misses reference and unassigned slices of the same traffic and completely ignores Perplexity.
• Compare this month to last. The native channel only counts from the deployment date and is activated at different times for different properties. So any range going back to spring 2026 compares marked traffic to unmarked traffic. This is not a trend; it’s an artifact.
• Check your ranking tracking. Entirely different question. Rankings tell you about position, not whether an assistant actually sent someone to your site.

The fix: 1 custom channel that matches the source

Here’s the move that actually solves the problem. Create a custom channel group and match the source while completely ignoring the media. The second you do that, the ai-assistant, referral And (not set) versions of chatgpt.com reduce to a single line.

You get two more bonuses. A custom channel group applies its rules retroactively across your entire date range, so it saves all those old ChatGPT sessions stuck in sponsorship. And because you write the rule, you can include the platforms that Google leaves out, including Perplexity.

Here’s how I configured it:

1. Go to Admin > Data View > Channel Groups and create a new group.

2. Add a channel and call it AI.

3. Set the condition to Source matches regex, with a pattern covering AI domains.

4. Drag the AI ​​channel above Referral and Organic, so that it claims these sessions first.

5. Save, then apply the group as the primary dimension in any acquisition report.

The regular expression is where you can quietly sabotage yourself, so construct it carefully. Keep entries tied to recognizable domains or service-specific host tokens. Never throw away a bare chip like gpt on its own, because it will match any source containing those three letters and result in false positives, which is exactly the sort of thing that helps separate your data. Here’s a boundary-aware template covering the main AI sources you want to track:

.*(^|(/.:@?&=))(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 this. gemini.google.com is a specific host, so it won’t grab all your Google organic items, which is exactly why you should never add a simple google to the list. And consider the motive perishable. I would review it quarterly, as platforms come and go and their domains may change.

The Problem Even a Perfect Channel Can’t Fix

This is the part that most guides skip over, and it’s the part that keeps you honest. A custom channel group fixes the way traffic is classifiednot if it happens collection in the first place. A source rule can only intercept sessions that presented themselves with a source that it can actually read.

The biggest blind spot is AI traffic without any referrer, which lands in Direct. Most of this comes from AI mobile apps and in-app browsers, which don’t feed anything to GA4, so there’s no source to match on.

On top of that, AI Previews and Google AI Mode are counted as organic search (google / organic), and Google deliberately excludes them from the AI ​​Assistant channel. For many sites, this is the largest AI surface area there is, and is invisible as “AI” in your reports. Don’t try to drag it into your AI channel either, or you’ll accidentally gobble up regular Google organics.

So be specific about what you claim. With this custom channel in place, you have a comprehensive and consistent number of AI sources that you can actually identify. What you don’t have is a measure of AI’s total influence. Whether an assistant recommended you to someone who never clicked is a completely different problem, and no amount of channel grouping is going to solve it.

What I would do this week

Consider the AI ​​Assistant native channel as a starting point, not a finished product. SO:

• Add a session source/media to a traffic acquisition report and find your AI domains split between AI Assistant, Referral, and Unassigned. Honestly, seeing the split for yourself is half the lesson.
• Create the source-based AI custom channel group to gather the fragments and recover the history left by the native channel.
• Write down a monthly baseline. With a moving list of platforms and a big dark slice you can’t see, the monthly trend will tell you much more than any headline number.

Google finally putting AI traffic on the dashboard is a real step forward. Remember that the default view gives you the easy slice. The real work now is knowing exactly which slices he discreetly leaves out.

More resources:

---

Featured image: Roman Samborskyi/Shutterstock