Back to “Technical setup”
Technical setup4 min read

Configuring GA4: the reports that actually matter

Isolating referral traffic from ChatGPT, Perplexity or Gemini in GA4 with a custom channel, and tracking how it evolves over time.

To track traffic from generative AI in GA4, isolate the referring domains (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com...) in the acquisition report, ideally through a custom channel, then follow its trend over time rather than looking at a single snapshot.

Isolating referral traffic from generative AI

By default, GA4 classifies incoming traffic using predefined channel rules (Organic Search, Direct, Referral, Paid Search...). Traffic from a click on a link cited by ChatGPT or Perplexity usually lands in the "Referral" channel, lumped together with any other site linking to you — which makes it invisible unless you dig further.

To isolate it, open the Acquisition > Traffic acquisition report, then add a secondary dimension for "Session source". Filter on the relevant domains: chatgpt.com, chat.openai.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai. This list evolves with usage and should be reviewed regularly.

Creating a custom channel or a dedicated segment

Manually filtering every time you check gets tedious fast. Two more durable options:

  • A custom channel group (Admin > Data channel groups), where you create a rule like "if session source contains chatgpt.com OR perplexity.ai OR gemini.google.com OR copilot.microsoft.com OR claude.ai, classify as 'Generative AI' channel." This channel then shows up like any other in standard reports, allowing ongoing tracking without redoing the filter every time.
  • A custom exploration (Explore menu), letting you build a table crossing session source, number of sessions, conversion rate, and landing pages, with export available for external tracking.

The custom channel is the more durable solution over time: once created, it applies retroactively to some extent and stays available across all standard reports without repeated manual work.

Reports worth checking beyond acquisition

Once the channel is isolated, three angles add more value than a simple session count:

  • Landing pages: which pages on the site actually receive this traffic? This reveals, indirectly, which content is being picked up by AI.
  • Engagement rate and session duration: does AI-driven traffic behave differently from classic organic traffic (shorter sessions, higher bounce rate, or on the contrary stronger engagement)? There's no universal rule here — it depends heavily on industry and content type, and is worth observing in your own data rather than assuming.
  • Conversions by channel: if a conversion event is set up, compare its rate on the generative AI channel against other channels, to assess the real value of this traffic rather than just its volume.

Tracking the trend over time

A one-off reading of AI traffic has little value on its own — the trend is what matters. Set up simple monthly tracking (number of sessions on the generative AI channel, compared to the previous month and to the previous year if volume allows). A gradual rise in this channel, even from a small base, is a signal that your content is starting to get picked up by answer engines — a signal that complements AI visibility scores measured elsewhere, but this time measured on real traffic rather than prompt tests.

Setting up an automatic alert

Rather than relying on remembering to check manually, GA4 lets you create custom insights (Admin > Custom insights) that trigger an email notification when a defined condition is met — for example, a change in session count on the "Generative AI" channel beyond a certain threshold over a week. This kind of alert removes the need to remember to check the report every month, and lets you react faster if an unusual spike or drop shows up, particularly after a robots.txt change or a content overhaul.

Limits worth knowing

This tracking has blind spots worth keeping in mind:

  • Some traffic from AI answers isn't necessarily attributed correctly: links shared from a mobile app or copy-pasted into a new tab can land as "Direct" traffic, invisible in the dedicated channel.
  • Consent banners (see the lesson on installing GA4) can reduce collected data volume if some visitors decline tracking.
  • Referring domain naming changes over time: a new AI engine or a domain change on the provider's side can make traffic disappear from your custom channel without warning, which is why the channel rule is worth revisiting regularly.

With this tracking in place, you have a direct measure of AI traffic to weigh against the technical fixes already made through Google Search Console and your site speed checks. It's also a solid foundation before tackling the next module's core question: why AI sometimes ignores your site.