What GA4's AI Assistant Channel Sees, and What It Still Misses

Yes, GA4 now labels some AI traffic. Since 13 May 2026 its default channel group has an AI Assistant channel, filled when a visit's referrer matches Google's list of assistants. It still misses visits with no referrer, ChatGPT links tagged with a source but no medium, assistants not on the list, Google's own AI Overviews, and every AI crawler.

This post was first written before that channel existed, and its title said GA4 could not see your AI traffic at all. That is no longer true, so here is the updated picture: what the new channel counts, where AI-related visits still land somewhere else, and which gaps no GA4 setting can close.

Yes. Google's release notes for 13 May 2026 announce "a new AI Assistant channel in your Default Channel Group reports" for "chatbots like ChatGPT, Gemini, and Claude". Three traffic-source values change when the referrer matches a recognised assistant:

• Medium becomes ai-assistant.
• Channel group becomes AI Assistant.
• Campaign becomes (ai-assistant).

The channel definition page, checked on 30 September 2026, describes the channel as traffic "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok" and says it "excludes Google's AI Overviews and AI Mode". Those two count under Organic Search instead. Claude appears in the release note but not in that list, and Google does not publish the full set of recognised hosts: the source-category spreadsheet linked from the page has no AI assistant category.

This is a real improvement. Before May, an AI referral with an intact referrer could land in Referral, Organic Search or Unassigned depending on the host. Now the recognised ones collect in one place, and you can compare that row with your other channels without building anything.

The channel is a referrer rule. It answers "did the browser say this visit came from a known assistant?", and every gap below follows from that one design choice.

If a visit carries neither a referrer nor campaign tags, GA4 files it under Direct. That is correct behaviour: GA4 reports what the browser sent. But links opened inside an assistant's own app often reach you with no referrer, and the reasons are mostly in browser and app design, not in anything GA4 controls. No setting recovers a header that was never sent.

Some ChatGPT links arrive tagged utm_source=chatgpt.com (the field guide to assistant referrals covers when). Search Engine Journal reported in July 2026 that when a link carries that source and no utm_medium, the session shows as chatgpt.com / (not set) and lands in Unassigned, "the channel almost nobody ever opens". So the same assistant can be split across AI Assistant, Unassigned and Direct depending on how the click travelled.

Recognition depends on Google's list, and the list is not published. The same SEJ test found Perplexity "still isn't on the list and keeps landing in Referral". Treat the AI Assistant row as the assistants Google recognises today, not as all AI referrals.

AI Overviews and AI Mode clicks count as Organic Search, so the AI Assistant channel will never show them. For those surfaces, Google now offers a separate tool: the Generative AI performance report in Search Console, rolled out to all sites on 31 August 2026, shows impressions from AI Overviews and AI Mode by page.

GA4 cannot see crawlers for two independent reasons. A crawler fetches the HTML and leaves without running your scripts: Vercel and MERJ found that "none of the major AI crawlers currently render JavaScript", including GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot and PerplexityBot. And even if one did, GA4 excludes known bot traffic automatically, and Google's help page says "you cannot disable known bot traffic exclusion". Crawls can only be counted on your server or edge, which is a separate measurement problem with its own directory of AI crawlers.

This is the mechanic that turns a gap into an error. Google's attribution documentation states: "All attribution models exclude direct visits from receiving attribution credit, unless the path to key event consists entirely of direct visits."

For a visit filed under AI Assistant, that is fine: it has a channel and can earn credit. The problem is the AI visit that arrived with no referrer. It is Direct, so it earns nothing, and the credit goes to whatever earlier channel touched the buyer.

The result is not a missing number. It is a wrong one in a specific direction: organic search, email and paid campaigns are over-credited by the amount that referrer-less AI visits are under-credited. The new channel does not change this, because it only helps visits that already carried a referrer.

Two limits matter when discovery and purchase are weeks apart.

Session timeout. A session ends after 30 minutes of inactivity by default. Someone who reads your pricing page, closes the laptop and returns that evening starts a new session, and only the first one carries the source.

Data retention. Standard GA4 properties keep event data for 2 or 14 months. The setting only affects explorations and funnel reports; standard aggregated reports are unaffected. Raising it to 14 months also applies to data you still hold, but anything already deleted under the shorter setting is gone. Check which one your property uses before you run any long-window analysis.

All of these are worth doing, and none closes every gap above.

1. Read the AI Assistant row, and label it honestly. It is the recognised, referrer-carrying subset. Report it as that, not as total AI traffic.
2. Build a custom channel group that matches on source. Match assistant hosts such as chatgpt.com, perplexity.ai, claude.ai and gemini.google.com in the source, ignoring medium, and place the channel above Referral. Google applies the first channel a visit matches, and custom channel groups apply to reports retroactively, so this rescues the Unassigned and Referral slices of past data too. Standard properties can have two custom groups.
3. Set retention to 14 months. One setting, and the cost of leaving it on 2 months is paid later and cannot be undone.
4. Segment Direct by landing page. Direct visits that land on deep pages such as pricing or comparison pages are unlikely to be typed URLs. That segment is where referrer-less AI visits tend to collect, and its trend is the best proxy GA4 offers.
5. Use Search Console for Google's AI answers. The Generative AI performance report covers AI Overviews and AI Mode, which the AI Assistant channel excludes.

GA4 is a good tool for the job it was built for, and the new channel is an honest step. Four things are outside what any setting can change:

• Crawlers. AI crawlers read your pages long before anyone clicks a link in an answer, and GA4 has no record of them.
• Confidence and method. A GA4 channel is a label. There is no field saying which signal produced it or how sure the rule is, so a sudden jump cannot be traced to real growth or a rule change.
• Revenue from payments. GA4 reports revenue from the purchase and refund events you send it. Joining a visit to what a payment processor actually charged, and netting out refunds, is work you do outside GA4.
• Signals that were never sent. No tool can read a referrer that the browser did not transmit.

Traceten is built around those gaps. It names the assistant behind each visit from the referrer, UTM tags and user agent, and stores a confidence score and the detection method with every classification. It records AI crawlers with server or edge middleware and checks them against published IP ranges and reverse DNS where the crawler's vendor supports them, and it joins visits to revenue from Stripe, Shopify, Lemon Squeezy, Polar and Paddle. Most teams keep GA4 for marketing and Google Ads and run both; the GA4 comparison sets out which tool answers which question.