How to track AI referral traffic properly in GA4
Jump to a section 8 sections
AI tools are starting to show up in ecommerce analytics, but for most sites, the traffic is still small enough to be misleading if you look at it in isolation.
TL;DR. Since May 2026, GA4 puts traffic from the main AI assistants into its own AI Assistant channel, so look there first. A custom channel group is still worth ten minutes if you want to compare against older months or catch assistants Google's list misses. For most shops it's under 1% of sessions, so treat it as a diagnosis, not something to optimise for yet.
Does GA4 track AI traffic on its own now?
Mostly, yes. On 13 May 2026 Google added an AI Assistant channel to GA4's default channel group. Open Reports > Acquisition > Traffic acquisition and it sits alongside Organic Search, Email and the rest.
There are two gaps. Google doesn't publish the full list of assistants it recognises: its channel definitions name ChatGPT, Gemini, DeepSeek, Copilot and Grok, and the release note adds Claude, but neither mentions Perplexity. And if you want to compare against the months before May, a custom channel group is the safer route, because Google applies custom groups to your past data as well.
How to set up your own AI channel in GA4
- Go to Admin > Data display > Channel groups
- Copy your default channel group (three little dots > Copy to create new)
- Add a new channel called something like AI
- Set the channel condition to Source matches regex
- Paste the regex below, and save
- Click Reorder, drag your AI channel above Referral (and above Google's AI Assistant channel, if your copy has one), then Apply
^.*ai|.*\.openai.*|.*chatgpt.*|.*gpt.*|.*copilot.*|.*edgeservices.*|.*gemini.*|.*perplexity.*|.*claude.*|.*grok.*|.*deepseek.*|.*you\.com.*|.*poe\.com.*|.*writesonic.*|.*nimble.*|.*outrider.*|.*astastic.*|.*copy\.ai.*|.*bnngpt.*$
Step 6 is the one people miss. GA4 puts each visit into the first channel it matches, working down the list, so an AI channel sitting below Referral never sees most of its traffic.
The regex is built on Google's own example, with Grok, DeepSeek, Claude, You.com and Poe added, and Bard and Neeva taken out because neither exists any more. New assistants keep turning up, so check it against your Session source report every few months.
Why it’s worth grouping AI traffic at all
Right now, AI-driven sessions are small for most ecommerce sites, often well under 1% of total traffic.
Because the volume is low, ungrouped data is easy to dismiss or misread. One spike looks meaningful, or the whole thing gets ignored.
Grouping AI traffic gives you a clean baseline.
Instead of asking “is this big or small”, you can ask better questions:
-
Does this traffic behave differently?
- Does it convert at a higher or lower rate?
-
Does it assist conversions even when it’s not the last click?
-
Is its share stable, creeping up, or volatile?
Without a baseline, every future conversation about AI becomes speculative. With one, you can tell whether something has genuinely changed.
What the data usually shows
For most ecommerce sites today:
-
AI traffic volume is negligible compared to organic, email, paid, and direct
-
Conversion rate is often higher than average
-
Assisted journeys are more common than last-click wins
This doesn’t make AI a better channel. It tells us something about intent.
By the time someone clicks through from an AI answer, a lot of filtering has already happened. Questions have been narrowed, comparisons made, and uncertainty reduced.
You only see this pattern clearly when the data is grouped properly.
How this helps inform decisions
The value of this channel isn’t optimisation, it’s diagnosis.
Pages that perform well for AI traffic usually share the same traits:
-
Clear structure
-
Explicit answers to common questions
-
Obvious delivery, returns, and reassurance information
Those are the same traits that help human visitors decide.
Tracking AI traffic properly helps you spot where clarity already exists, and where it doesn’t, without chasing a new channel or rebuilding your site for a trend.
When you should care about this
If your fundamentals are shaky, weak product pages, unclear propositions, poor conversion, this should not be a priority.
This data becomes useful once your core channels are stable and your pages already do a decent job of helping people decide.
At that point, AI traffic becomes an early signal, not a distraction.
What not to do
This is not a reason to rebuild for AI, chase GEO tactics, or deprioritise organic search.
Track AI traffic to stay informed, not to stay busy.
Want more thinking like this?
This post is adapted from Ecommerce, Prioritised, a short weekly newsletter focused on helping ecommerce teams make clearer decisions about growth.
It’s written for Shopify businesses that are growing, but don’t want more tools, tactics, or noise, just better judgment.
Get the next one in your inbox
Ecommerce Prioritised is a free weekly read on growing an ecommerce store: what to prioritise, what to ignore, and the changes worth acting on.