How to Measure AI Referral Traffic in GA4
By default, GA4 scatters AI-assistant traffic across Referral and Direct, mixed in with everything else. Most teams have no idea how much they are getting.
Here is how to isolate it — and, just as importantly, what this measurement cannot tell you.
TL;DR
- AI assistants appear as referrals from a handful of identifiable hostnames.
- Build a GA4 exploration filtered to those hostnames to get a working view today.
- Create a custom channel group so AI traffic is separated permanently.
- AI referral sessions are typically small in volume but high in intent.
- The measurement has a hard ceiling: it counts only the people who clicked. The people who read the answer and never clicked — usually the majority — are invisible to analytics entirely.
Step 1 — Find the traffic you already have
In GA4: Reports → Acquisition → Traffic acquisition, then switch the dimension to Session source / medium.
Look for hostnames belonging to the assistants. The common ones include:
chatgpt.com/chat.openai.comperplexity.aigemini.google.comclaude.aicopilot.microsoft.com
Search the source field for each. Most teams find a non-zero number they had never noticed.
Two caveats. Some assistants strip or omit the referrer, so a share of this traffic lands in Direct and is unrecoverable. And in-app browsers sometimes report inconsistently. Treat whatever you measure as a floor, not a total.
Step 2 — Build an exploration
For a working view now:
- Explore → Blank.
- Dimension:
Session source. Metrics:Sessions,Engaged sessions,Average engagement time,Conversions. - Add a filter:
Session sourcematches regexchatgpt|openai|perplexity|gemini\.google|claude\.ai|copilot\.microsoft - Add
Landing pageas a second dimension.
That last dimension is the valuable one. It tells you which pages assistants are actually sending people to — which is a direct, observed signal of which of your content earns citations.
Step 3 — Make it permanent with a channel group
Explorations are ad hoc. To track this properly:
Admin → Data display → Channel groups → Create new channel group.
Add a channel named AI Assistants, placed above Referral and Organic in the ordering, with the condition: Source matches regex, using the pattern above.
Ordering matters — channel groups evaluate top down, so if Referral comes first it will swallow the traffic before your rule sees it.
Note that custom channel groups apply going forward and to a limited window of historical data. Set this up now rather than later.
Step 4 — Watch the right metrics
Volume is the least interesting number here. Compare quality instead:
| Metric | What to look for |
|---|---|
| Engagement rate | Frequently higher than organic search |
| Avg. engagement time | Often longer — these visitors arrived deliberately |
| Conversion rate | Commonly stronger, because intent is pre-qualified |
| Pages per session | Varies; a single deep read is a normal pattern |
The recurring finding across teams that measure this: AI referral volume is small and its per-session value is high. Judging the channel on sessions alone badly undersells it.
What this measurement cannot tell you
This is the part that matters most, and it is where analytics-driven teams go wrong.
You are only counting clicks. Every person who asked an assistant about your category, read an answer that named you, formed an impression, and never clicked, is invisible. In most categories, that group is far larger than the group that clicks.
You cannot see the answers you lost. GA4 shows traffic that arrived. It cannot show the questions where a competitor was recommended instead of you — which is the more important number.
Attribution is lossy. Stripped referrers push an unknown share into Direct.
So AI referral traffic in GA4 is a genuine, useful signal — and a severe undercount of AI's actual influence on your funnel. It answers "how many people clicked through from an assistant?" It does not answer "how visible are we inside AI answers?" Those require different instrumentation: measuring the answers themselves.
Use GA4 for the click-through story. Do not mistake it for the whole picture.
Frequently asked questions
Why is my AI traffic so low? Partly genuine, partly measurement loss from stripped referrers, and partly because most AI interactions end without a click by design.
Should I build a separate conversion goal for it? Not initially. Get the channel group in place and observe for a quarter first.
Does this work in other analytics tools? Yes — the same hostname list works in Plausible, Fathom, Matomo, and Adobe. The mechanics differ; the principle does not.
How do I measure visibility rather than clicks? By measuring the answers directly — running category questions across assistants and recording whether you are named. That is a different exercise from web analytics.
Published by the SIQA Editorial Team.