Track AI Traffic in GA4: What October’s Outage Exposed

Your ChatGPT traffic did not fall in October

The second week of October looked alarming in most analytics accounts. ChatGPT-attributed sessions dropped hard across the web between October 8 and October 15. But almost none of that drop was real. The visits kept arriving, and they simply stopped carrying the label that identified them.

Here is the mechanism. GPT-6 rolled out on October 7 and 8. After that, ChatGPT began opening links in a new tab with a browser setting that withholds the referrer, and it also left off the utm_source tag. So analytics had nothing left to match on. Those sessions fell into direct traffic instead.

What three independent measurements found

They landed in roughly the same place:

• Graphite: ChatGPT-attributed GA4 sessions fell between 22% and 57% on sites with meaningful AI Assistant traffic.
• LLM Pulse: ChatGPT sessions ran about 46% lower across October 9 to 11 than the previous fortnight.
• Adobe: the drop in ChatGPT visits was almost exactly matched by a rise in direct visits.

That last pairing is the tell. Traffic does not move from one channel to another because shoppers changed their minds. It moves because the label changed. Meanwhile the share of ChatGPT answers carrying at least one tagged source link fell from about 98% on October 8 to 66% by October 14.

The tag returned on October 15, and tagged clicks were back to roughly 95% of the prior week within hours. One caveat belongs here, though. The confirmation that this was a bug is secondhand, reported by Adobe’s Cedric Huesler. OpenAI has not addressed it publicly, and its release notes for early October do not mention it.

Why one missing tag could do that much damage

Because there is no backup signal. Your AI traffic number exists almost entirely because ChatGPT volunteers utm_source=chatgpt.com on the links it puts in an answer. That is a courtesy, not a standard, and the vendor can withdraw it without telling anyone.

The referrer would normally be the fallback. But the referrer is still missing on most ChatGPT clicks even now that the tag is back. So the October fix restored the label without restoring the safety net underneath it.

There is a second, older gap worth knowing about. The ChatGPT mobile apps frequently pass no normal web referrer at all. Those sessions have always landed in direct traffic and blended into everything else unattributed. In other words, your AI number was an undercount before October and it remains one now.

How to track AI traffic in GA4

Good news first: there is nothing to configure. Google added a native AI Assistant channel to GA4’s default channel group on 13 May 2026, and it applies automatically.

Google’s own definition is narrow and worth reading literally:

A session lands there when the medium exactly matches ai-assistant. It also lands there when the referrer matches Google’s list of AI assistants, in which case the medium is set to ai-assistant and the campaign to (ai-assistant). To see it, open Reports, then Acquisition, then Traffic acquisition, and group by Default channel group.

The blind spot nobody mentions

The same help page is blunt about what the channel leaves out. It “excludes Google’s AI Overviews and AI Mode.” That traffic stays inside Organic Search, mixed in with ordinary blue-link clicks.

So the AI Assistant channel counts assistants that send you a click. It does not count the AI surface most of your shoppers actually see, because Google does not separate it. If you are judging your AI visibility by that one channel, you are reading a partial number and treating it as the whole.

The same check inside your Shopify admin

You do not need GA4 for a quick read. Shopify’s own reports will do it:

1. Open Analytics, then Reports.
2. Open Sessions by referrer, one of Shopify’s four acquisition reports.
3. Group by Referring platform, the platform a visitor arrived through.
4. Check Referring channel and Referring medium as well, because an assistant can land in either depending on how its link was tagged.
5. Fall back to Referrer path, which shows the raw session data, when the grouped views hide what you need.

Why this matters for your store: these are the same visits GA4 is counting, arriving through a different door. So when your two tools disagree, you have found a tagging problem rather than a traffic problem.

Read it with the same caution, though. That report works because of the UTM tag, not because of a clean browser referrer. It went quiet in October for exactly the same reason GA4 did.

Which guide do you want?

This post is about counting AI traffic. Two neighbours cover the other halves of the question, so go sideways if that is what you came for:

• Whether the traffic is worth pursuing at all, with the conversion data: AI Search ROI: Is AI Traffic Worth Chasing Yet?
• How to earn the citation in the first place: How to Get Cited by ChatGPT

What the number is worth once you have it

Small, growing fast, and better quality than its size suggests. Shopify’s own Q1 2026 commerce data puts AI chatbot referral sessions at more than 8x year-over-year growth on Shopify storefronts. Those visitors convert at roughly 50% higher rates than organic search on product detail pages, with average order values about 14% higher.

Treat those figures carefully, because Shopify discloses no sample size or methodology beyond “Q1 2026 commerce data”. They are first-party numbers from a party with an interest in the answer.

One more trap, and it catches people constantly. Last-click referral figures and whole-journey clickstream figures measure different things, so they disagree without either being wrong. We pulled that apart in detail in the AI search ROI post, and it is worth reading before you quote any AI conversion number at your team.

Three things to do this week

Practical, and none of it takes long:

1. Annotate October 8 and October 15 in your reporting. Any year-on-year comparison that spans that week is distorted, and in six months nobody will remember why.
2. Check direct traffic alongside AI traffic, not on its own. When one falls and the other rises by a similar amount, suspect labelling before you suspect shoppers.
3. Stop treating a single AI number as truth. Compare GA4’s AI Assistant channel against Shopify’s referrer reports, because when they disagree the gap is the story.

The deeper lesson from October is not about one bug. It is that AI attribution rests on voluntary tags from vendors who owe you nothing. So build the habit of cross-checking now, while the stakes are still small. For the search side of the same job, Google Search Console remains the one source nobody can silently switch off.