webtracking.org
New feature GA4 published 2026-08-16

GA4 now buckets AI-assistant traffic natively — check your custom channel groups

Impact Moderate — worth knowing, act opportunistically
What changedGA4 now measures AI-assistant referrals natively: recognized assistant traffic gets a new 'ai-assistant' medium value, an 'AI Assistant' default channel, and '(ai-assistant)' campaign labels (rolled out May 13, 2026). Historical data is not reclassified, and custom channel groups do not update themselves.
Who's affectedAnyone reporting on acquisition in GA4 — AI referrals stop polluting Referral/Direct from mid-May onward, which bends every trend line. Especially relevant if you built a custom 'AI' channel (including with our generator): it can now disagree with the native one.
ActionAnnotate May 13, 2026 in your reporting. If you run a custom channel group with an AI channel, add a medium = ai-assistant condition to it (or retire it deliberately) so custom and default groupings stop telling different stories.

What changed

Until May, a click out of ChatGPT or Perplexity landed in GA4 as generic Referral — or, via in-app browsers that strip the referrer, as Direct. As of May 13, 2026, GA4 classifies traffic from assistants it recognizes with three coordinated labels: a new ai-assistant medium, a dedicated “AI Assistant” channel in the default channel group, and (ai-assistant) campaign naming.

This is processing-time classification, Google-side. You configure nothing — and you control nothing about it.

What it does not do

  • No backfill. Sessions before May 13 keep their old classification. Every AI-adjacent trend has a step in it at that date; annotate it or someone will read the jump as growth.
  • No published registry. Google has not documented which assistants qualify or how the list evolves. You cannot audit what’s in the bucket — which is the structural difference from a rules-based custom channel, where the source list is explicit and versioned.
  • No dark-traffic recovery. Assistant visits that arrive with no referrer at all — webviews, copied links, “the assistant summarized you and the user typed the URL” — still land in Direct. Native classification narrows the gap; it does not close it. First-touch capture (the dataLayer snippet our generator emits) remains the only way to keep those.

The conflict to fix

If you built a custom channel group with an AI channel keyed on source (the pre-May approach, including ours), you now have two definitions in play: your source-regex channel in custom groups, and Google’s medium-based channel in the default group. They will disagree — Google’s registry and yours won’t match exactly — and a report built on the default group will contradict one built on your custom group.

The fix is one condition: add medium exactly matches ai-assistant (OR’d with your source rules) to your custom AI channel. You keep the auditability and the assistants Google may not recognize; you inherit the native classification where it fires. Our generator now amounts to that hybrid: explicit sources + the native medium.

What to do

  1. Annotate 2026-05-13 in GA4 and any BI layer that reads its exports.
  2. Reconcile custom channel groups with the condition above — or deliberately retire the custom channel and note the definition change.
  3. Keep measuring the dark share. Compare “AI Assistant” channel volume against branded-search and direct lift; the visible channel is the floor, not the total — method in the AI traffic guide.
  4. Adobe users: nothing changed for you — the processing-rule approach is still the only route.

Sources

Every claim in this brief traces to a source above or to our own published crawl data. Corrections: contact — applied and dated.

The next brief, in your inbox

One email a week: what changed across the tracking stack, with the same what/who/action structure.

Weekly. Free. One-click unsubscribe. Opens Substack in a new tab — you won't lose your place here.