On October 1, 2026, Search Engine Journal reported something small that's going to cause a lot of GA4 dashboards to shift without anyone changing a setting: Gemini has started appending UTM parameters to links in its answers. Not announced. Not in a release note anywhere Google publishes. A Reddit user spotted it first, Google's John Mueller showed up in the thread and confirmed he saw it too, and that's the entire paper trail right now (Search Engine Journal, Oct 1, 2026).

Here's the honest state of it. Google hasn't documented which UTM keys Gemini adds, what values go in them, or under what conditions the tagging fires. Mueller's own comment in the thread, as quoted by SEJ, was "Nice. I see it too. Doesn't Gemini have a referrer too though? I thought you could already track them." That's a Google employee speaking informally on Reddit, not an engineering spec. Anyone telling you the exact utmsource and utmmedium strings Gemini uses, with total confidence, is guessing or hasn't checked their own GA4 data recently. So that's step one of this whole piece: go look at your own Session source / medium report, filter for anything with "gemini" in it, and see what's actually showing up. That's more reliable right now than any blog post, including this one.

What isn't guesswork is the infrastructure this lands on top of. GA4 built a home for AI referral traffic back in May 2026, and understanding that first makes the Gemini news easier to place.

The AI Assistant channel already exists, and it predates this

In May 2026, Google Analytics shipped a new default channel group called AI Assistant. It wasn't a UTM feature, it was a referrer-recognition feature: GA4 started checking incoming referrer domains against a list of known AI assistants, and when it found a match, it classified that session's medium as ai-assistant and its campaign as (ai-assistant), then rolled it up into the AI Assistant channel group in your acquisition reports (Google Analytics Help: Default channel group). Google's own release language names ChatGPT, Gemini, and Claude, and Search Engine Journal's coverage of the launch cites earlier guidance covering five platforms total: "ChatGPT, Gemini, Microsoft Copilot, Claude, and Perplexity," while noting plainly that "Google hasn't published the full list of recognized AI assistant referrers" (Search Engine Journal, May 14, 2026). The same coverage makes a point worth remembering before you build the custom channel group later in this piece: "GA4 property owners no longer need to build custom channel groups with regex patterns to separate AI assistant visits from referrals," at least for the platforms already on Google's recognized list.

This matters for how you read the Gemini UTM news. Before any UTM showed up, Gemini referral sessions were already landing in a named channel, because GA4 was reading the referrer header, not a campaign tag. A UTM parameter doesn't create that classification. It can change it, though, because GA4's acquisition model generally gives campaign parameters priority over referrer data when it's deciding session source and medium. If Gemini starts sending utm_source and utm_medium values that don't say "ai-assistant," your reports could start splitting what used to be one clean bucket into two: an older referrer-based ai-assistant row, and a newer, UTM-tagged row with different wording. Google hasn't published whether that's actually how it's behaving, so this is a prediction worth testing against your own account, not a settled fact.

There's a second piece of GA4's AI model worth knowing before you build anything: Google's AI Overviews and AI Mode are not broken out as their own channel at all. Google's documentation is explicit that "Organic Search is the channel by which users arrive at your site/app via non-ad links in organic-search results, including Google's AI Overviews and AI Mode" (same default channel group reference above). A click from an AI Overview card lands in Organic Search next to every ordinary blue-link visit, indistinguishable in the acquisition report. More on why that's a real limitation, not a footnote, later in this piece.

How to actually build a report for this in GA4

You have two paths, depending on how much detail you need.

The fast path, for most people most months: open Reports → Acquisition → Traffic acquisition, and look at the Session default channel group table. AI Assistant is already there as its own row if you've had any qualifying traffic. Click into it, add Session source / medium as a secondary dimension, and you'll see the actual source strings GA4 is recognizing, which is also your fastest way to spot whether Gemini's new UTM values are showing up as something other than the referrer-based entries you're used to.

The detailed path, for when you need to slice by landing page, by exact source string, or build something you can export and hand to someone else, is an Exploration.

  1. Go to Explore → Blank.
  2. Add dimensions: Session source / medium, Session campaign, Session default channel group, and Landing page + query string.
  3. Add metrics: Sessions, Engaged sessions, and whichever key event or conversion you report on (purchases, leads, signups).
  4. Drag Session default channel group into Rows, then add Session source / medium as a nested row so you can see the specific sources sitting inside the AI Assistant bucket.
  5. Add a filter on Session source / medium containing any of: gemini, chatgpt, perplexity, copilot, claude. This catches referral-based sessions even if the channel group logic hasn't caught up to a new naming pattern yet, which is exactly the gap a brand-new, undocumented UTM rollout could open up.
  6. Save the exploration and check it monthly for the first quarter after any AI platform ships a tagging change. Patterns that look stable in month one can shift in month two without any warning, because none of this is governed by a changelog you can subscribe to.

That filter in step 5 is the part worth keeping even after today's Gemini news stops being news. New AI products will keep showing up, and GA4's default channel rules will always be a step behind whatever's shipping that week.

Comparing it to how ChatGPT already handles this

ChatGPT has had UTM tagging in its outbound links for longer than Gemini, and it's narrower than you'd assume. Per the same Search Engine Journal reporting on the Gemini news, ChatGPT "adds UTM parameters to outgoing links but only to links that are in the context of answers that are grounded from web sources, not from information that comes from training data." In other words, if ChatGPT answers purely from what it learned during training, with no live retrieval behind the answer, there's no outbound link to tag in the first place. The tagging only shows up on answers where the model went and fetched something from the web to ground the response, which in practice is a meaningful share of commercial and current-events queries but not all of them.

The practical result in GA4 for most sites: look for chatgpt.com appearing as a session source. Several practitioners tracking this, including a recent Practical Ecommerce walkthrough (Ann Smarty, Oct 11 2026), recommend going to Traffic acquisition, filtering the Session source / medium dropdown by typing "gpt," and reviewing what comes back. That same piece flags the thing that actually loses you data here: "Google Analytics incorrectly classifies a percentage of ChatGPT traffic as either Direct or Unassigned because ChatGPT often adds a 'noreferrer' attribute to external links" (Practical Ecommerce). A rel="noreferrer" link strips the referrer header entirely before the browser ever sends it, so GA4 has nothing to classify the session against unless a UTM parameter happens to survive on the URL itself. When both the referrer and any campaign tag are missing, the session falls into Direct or Unassigned, and no exploration you build will recover it after the fact. This is worth sitting with for a second: your "Direct" channel in GA4 right now almost certainly has AI chatbot traffic sitting inside it, mislabeled, with no way to retroactively sort it back out.

Perplexity, Claude, and Copilot: referrer-based, not UTM-based (for now)

As of this writing, there's no public reporting that Perplexity, Claude, or Microsoft Copilot are appending UTM parameters to outbound links the way ChatGPT does and Gemini apparently now does. That means their referral traffic depends entirely on the referrer header making it through intact, which GA4's AI Assistant channel rule reads automatically when the referring domain matches its recognized list.

Practically, that means two things for you. First, you don't need to build anything special to catch these three; if the referrer survives, they already land in AI Assistant, same as Gemini did before this week. Second, they're exposed to exactly the same failure mode ChatGPT has: any in-app browser, any link-stripping proxy, any noreferrer attribute on the publisher's end of a shared link, and that session silently becomes Direct or Unassigned instead. If you want to sanity-check whether you're under-counting any of these three specifically, cross-reference your server logs or a CDN's bot/referrer log against what GA4 shows for the same week. A gap between the two is your stripped-referrer tax, and for some sites it's larger than the visible AI Assistant number itself.

Where this breaks down regardless of what you build

Three limits apply no matter how carefully you set up the exploration above, and none of them are fixable from the GA4 side.

App traffic. A link opened inside the ChatGPT, Gemini, or Perplexity mobile app, inside an in-app browser rather than the system browser, frequently sends no referrer at all, or sends an app-internal identifier GA4 doesn't recognize. That traffic lands as Direct. There's no setting to turn this off, because it's a platform-level decision about what the app's WebView passes along.

Stripped referrers generally. Beyond apps, any rel="noreferrer" link, any HTTPS-to-HTTP downgrade (referrers don't cross that boundary by default), and some corporate and VPN proxies, all erase the referrer before your analytics ever see it. UTM parameters survive this because they live in the URL itself, not in a header, which is exactly why a platform choosing to add UTMs, like Gemini apparently just did, is a meaningfully more durable signal than relying on referrer alone.

AI Overviews and AI Mode clicks count as Google Organic Search, full stop. This is the one people get wrong most often. If someone clicks through from an AI Overview box on a Google results page, or from an AI Mode conversation, GA4 files that session under Organic Search, identical to a click on a standard blue link (Google Analytics Help). You cannot separate "came from an AI Overview" from "came from a regular ten blue links result" inside GA4's acquisition reports. If you need that split, the only path is Google Search Console's Search Appearance filters, which can show impressions and clicks tied to AI Overview appearances specifically, reported separately from GA4 entirely. Our piece on tracking AI Overview mentions covers that side of the measurement problem in more detail.

A custom channel group, if you want AI traffic broken out everywhere you look

Google's position, per the reporting above, is that you shouldn't need a custom regex group anymore for the platforms already on its recognized list. That's fine as far as it goes, but it's also a fixed definition you can't edit, covering a list Google hasn't published in full, and it won't automatically absorb a new UTM pattern the moment a vendor starts using one, which is exactly what may be happening with Gemini right now. A custom channel group gives you a rule you control instead of one you're trusting to keep up. In Admin → Data display → Channel groups → Create new channel group, add a channel, name it something like "AI Answer Engines," and set a condition using a regex match against Session source:

chatgpt\.com|gemini\.google\.com|perplexity\.ai|copilot\.microsoft\.com|claude\.ai

Pair that with an OR condition on Session default channel group exactly matches AI Assistant, so you catch both the sources GA4 already recognizes by referrer and anything slipping through on a UTM string that happens to use a different source value. Put this custom group next to Default channel group in your regular Traffic acquisition report as a comparison, and you've got a single view that survives the next undocumented tagging change instead of needing a rebuild every time one ships.

FAQ

Does Gemini's UTM change mean I'll see more AI traffic in GA4?

Not necessarily more traffic, more correctly labeled traffic, assuming the UTM values line up with what GA4's channel rules expect. If they don't line up, you could actually see some Gemini sessions split off into an "Unassigned" or differently-named bucket until Google documents the behavior or GA4's default rules catch up.

Is utm_source=chatgpt.com confirmed by OpenAI?

No. It's a pattern widely observed and reported by SEO practitioners checking their own Session source / medium reports, not something OpenAI has published documentation confirming. Treat it the same way as the current Gemini situation: real, observable in your own data, officially undocumented.

Why does my GA4 show so little AI referral traffic compared to what I'd expect?

Check Direct and Unassigned before assuming the traffic isn't happening. Stripped referrers, in-app browsers, and noreferrer links all route AI-sourced sessions into those two buckets instead of AI Assistant, often invisibly.

Can I track AI Overview clicks separately from regular organic search in GA4?

No, not inside GA4. Google counts both under Organic Search. Google Search Console's Search Appearance filtering is the only place to see AI Overview-specific performance data.

Should I set up a custom channel group now or wait for Google to document the Gemini change?

Set it up now. Waiting on documentation that may never arrive at the level of detail you want means losing however many weeks or months pass in between with no visibility at all. A regex-based custom channel costs you ten minutes and gets updated the moment you decide it needs to be.

If you're responsible for a GA4 property that gets any meaningful AI referral traffic, the one-hour version of this whole piece is: open Traffic acquisition today, check what's sitting in Direct that has a landing page matching a high-intent, commercial page you'd expect an AI assistant to cite, and compare that against your AI Assistant channel numbers for the same period. The gap between those two tells you more about your real AI-referral volume than anything a vendor will document this quarter. For the broader measurement picture beyond referral traffic specifically, our guide to AI visibility KPIs covers the metrics worth tracking once mentions start happening whether or not anyone clicks through at all, and how to get your brand mentioned in Gemini covers the earlier half of this problem, getting cited in the first place. If this is the first time you've gone looking at AI referral traffic specifically, our guide to getting cited by ChatGPT and Google AI Overviews and this week's AI search news roundup are good next stops.