For most of 2026, measuring how much traffic AI assistants actually send you has been guesswork. Referrers pass through inconsistently — a large share of Gemini clicks arrive with no referrer at all and get filed as anonymous 'direct' visits — so the real number has been hidden inside a bucket you can't attribute. Google's Gemini quietly changing that is the news: it has begun appending UTM parameters to the links it surfaces, a tracking tag that rides in the URL and survives where a referrer doesn't. It's undocumented, Google hasn't published the format, and a Reddit user found it before anyone at Google confirmed it — so it's an early signal, not a finished feature. This guide explains what the tag is and isn't, why referrer data misses so much AI traffic in the first place, how Gemini already lands in GA4's native AI Assistants channel, the crucial difference between Gemini-app traffic and AI Overviews (which are not the same thing and shouldn't be counted together), the one tagging mistake that quietly corrupts your data, and — once you can finally see the number — what the measurement actually tells you to do about it.
Measuring how much traffic AI assistants actually send you has been a guessing game all year. The honest answer for most sites has been "more than analytics shows, but we can't prove how much" — because the visits arrive without a reliable fingerprint and get swept into the "direct" bucket alongside bookmarks and typed-in URLs. Google's Gemini quietly changing that is the development worth acting on: it has started adding UTM parameters to the links it surfaces, which is the first attribution signal from Gemini that actually survives the trip from the assistant to your analytics.
Two things to hold in mind before you get excited. First, it is undocumented — a Reddit user spotted it in late September 2026 and Google's John Mueller confirmed he sees it too, but Google has not announced the format or the rules, so it can change without notice. Second, a tracking tag measures traffic; it does not create it. The number it reveals is diagnostic, and the useful question is what you do once you can finally see it. This guide covers both halves: how to read Gemini traffic accurately in 2026, and what the reading actually tells you to do.
To understand why a UTM tag matters, you have to understand why the thing it replaces fails. Web analytics has traditionally attributed a visit using the referrer — the header a browser passes that says "this person came from over there." It works well for an ordinary link click from one web page to another. It works badly for AI assistants.
The problem is that referrers from assistants pass through inconsistently, and from Gemini specifically a large share arrive stripped. Clicks that originate in a mobile app often carry no referrer at all, and invocations through Android's assistant layer tend to drop it entirely. When the referrer is gone, your analytics has nothing to attribute the visit to, so it files it as "direct" — the same bucket as someone who typed your URL by hand. The result is a systematic undercount: real, qualified visits that a person reached by following a Gemini citation, invisible as such, hiding inside a number that looks like brand loyalty. This is the exact gap the UTM tag closes, because a UTM parameter is part of the URL itself, not a header the browser may or may not forward.
A UTM parameter is a short string appended to a link — the "?utm_source=…" style tag marketers have used for two decades — that analytics tools read to attribute a visit to a campaign or source. What Gemini has begun doing is adding such a parameter to the links it surfaces to websites, identifying Gemini as the source. Because it rides inside the URL, it reaches your analytics even when the referrer has been stripped, which is precisely the case referrer-based detection was failing on.
Now the caveats, because they matter for how much you trust the number. Google has not documented this behavior: no announcement, no published parameter format, no stated rule for when the tag is applied. The reasonable assumption is that it fires when Gemini cites or links a source in an answer, but that is inference from observed behavior, not a contract. So the right posture is to treat a UTM-tagged Gemini visit as a confirmed Gemini referral, and the total of them as a floor rather than a complete count — some Gemini traffic may still arrive untagged, and Google could change the format or stop tagging tomorrow. It is a genuine improvement in visibility and an unstable one. Build dashboards that can tolerate it shifting.
Gemini is not first here, which is useful context for calibrating expectations. ChatGPT already appends a documented tag — utm_source=chatgpt.com — to links that are grounded in live web sources rather than drawn from its training data, so anyone tracking AI traffic has had a clean ChatGPT signal for a while. Perplexity and Claude, by contrast, have not consistently tagged their outbound links, leaving referrer-based detection as the only (unreliable) option for those two.
So Gemini adopting UTM tags narrows a measurement gap rather than inventing a new category. The practical upshot for your reporting is that you can now compare tagged traffic from at least two major assistants — ChatGPT and Gemini — on a consistent basis, instead of trusting ChatGPT's clean numbers and squinting at everyone else's. If you maintain a cross-assistant view of where your audience actually comes from, this makes two of the columns trustworthy. For the strategic picture of where visibility effort pays off across assistants, we cover that in AI visibility across the major AI assistants.
Here is the part people over-engineer: in Google Analytics 4 you mostly do not need to build anything. GA4 ships a native AI Assistants channel that recognizes a list of AI assistants — Gemini explicitly among them — by their referrer, commonly gemini.google.com, and automatically files those sessions with the medium set to ai-assistant and the campaign to (ai-assistant). No custom channel group, no regex, no manual utm mapping required for the referrer-detected portion.
What the new UTM tags add is coverage for the sessions the referrer method was missing — the stripped-referrer mobile and Android clicks described earlier. The genuinely useful exercise, and the one worth doing this month, is a comparison: look at your Gemini sessions as detected by referrer in the AI Assistants channel, then look at the sessions carrying Gemini's UTM tag, and see how far apart they are. The gap is an estimate of how badly the referrer-only method was undercounting Gemini before the tags existed. That single comparison turns "we think Gemini sends some traffic" into a defensible range.
This is the measurement error most likely to make your AI-traffic reporting wrong, and it has nothing to do with UTM tags. Gemini the assistant and AI Overviews — the AI-generated summaries that appear at the top of Google Search results — are different surfaces that behave differently in analytics, and counting them together produces a number that means nothing.
Gemini lives at gemini.google.com, and a click from it is a true referral filed under the AI Assistants channel. AI Overviews live inside the Search results page, generate no distinct referrer, and count as organic search — Google explicitly excludes them from the AI Assistants channel. So a visit that came because your page was cited in an AI Overview shows up in your organic numbers, and its impression data lives in Search Console, not in any assistant channel. A Gemini UTM tag tells you about assistant traffic; it tells you nothing about AI Overview visibility. Keep the two questions — "how much does the Gemini assistant refer" and "how am I doing in AI Overviews" — in separate reports, because they are answered by separate systems.
When site owners learn that Gemini tags its links, a tempting-but-wrong instinct follows: "I'll add utm_source=gemini to my URLs so it's tracked properly." Do not do this. It is the fastest way to ruin the exact number you are trying to measure.
The reason is that Gemini links to your real, canonical URL — the same URL people reach from organic search, from your email newsletter, from a social post, from a bookmark. If you bake utm_source=gemini into that canonical URL, every one of those visitors gets stamped as Gemini traffic, not just the ones who actually came from Gemini. You would inflate Gemini to absurd levels and strip attribution from every other channel in the process. UTM tagging is something the referring source does to a link it controls; it is not something the destination applies to itself. Let Gemini add its tag when it surfaces your link, and leave your canonical URLs clean. Your job is to read the tag in analytics, never to reproduce it.
Attribution is diagnostic. Once the Gemini tag confirms what most sites suspected — that AI assistants are now a real and growing referral source, not a rounding error — the number stops being the interesting part. The interesting part is the lever it points at, and that lever is not an analytics setting. It is whether your brand is the one Gemini chooses to cite in the first place, and whether it describes you the same way every time.
That is a production-and-consistency problem, and it is where Kompozy earns its place in this workflow — not as a tracker, but as the engine that makes the thing the tracker measures. Assistants ground their answers in brands that publish clear, current, on-brand content across the surfaces they read, and that keep their claims, name, and positioning identical everywhere so the model isn't choosing between three contradictory versions of you. Doing that consistently, at the cadence assistants reward, is precisely what stalls most teams. Kompozy is a full AI content generation and multi-platform publishing engine: it turns one source into finished blog articles, newsletters, platform-native text, persona and avatar video, and brand-exact graphics, all governed by a single Persona Brief so the brand reads the same in every output, then schedules and publishes them across the eight social platforms plus blog and email through autopilot.
The discipline behind that — keeping ChatGPT, Gemini, and Perplexity describing you one consistent way — is worth its own read in AI search brand consistency, and the question of why assistants name a competitor instead of you is covered in why AI recommends your competitor. The through-line is simple: the UTM tag lets you prove AI referral is real and size it; a consistent, high-volume content engine is how you grow it. Measurement without production is just a better-instrumented plateau.
Gemini quietly adding UTM tags is a real upgrade to AI-traffic measurement and an unfinished one. Use it: let GA4's native AI Assistants channel catch what it can by referrer, let the UTM tag catch the stripped-referrer clicks it was missing, and compare the two to size the undercount you've been living with. Keep Gemini-assistant traffic strictly separate from AI Overviews, which are organic and live in Search Console. Never tag your own canonical URLs. And treat the whole exercise as diagnosis, not destination — the number tells you AI referral is worth winning, and winning it is a content and consistency job, not an analytics one.
They are campaign tracking parameters that Google's Gemini has quietly begun appending to the links it surfaces to websites. The tag identifies Gemini as the source in the URL itself, so analytics can attribute the visit to Gemini rather than filing it under 'direct' traffic. Google has not documented the exact format or the conditions that trigger it — a Reddit user found the behavior and John Mueller acknowledged seeing it — so treat it as an early, unofficial signal rather than a settled feature you can build permanent reporting on.
You largely don't have to configure it. GA4 has a native AI Assistants channel that recognizes Gemini by its referrer (commonly gemini.google.com) and files those sessions automatically with the medium set to ai-assistant. The new UTM tags supplement that for the cases where the referrer is stripped — mobile apps and Android assistant invocations in particular. Compare your UTM-tagged Gemini sessions against the referrer-detected ones to see how much the referrer-only method was undercounting.
Because referrers pass through inconsistently. A large share of Gemini clicks, especially from mobile apps and Android's assistant, arrive with the referrer header stripped entirely, so analytics has nothing to attribute them to and files them as anonymous 'direct' visits. That is the core problem UTM tags solve: a UTM parameter rides inside the URL and survives the trip even when the referrer doesn't, so the visit can still be traced back to Gemini.
No, and conflating them is a common mistake. Gemini is the standalone assistant at gemini.google.com; its clicks are real referrals filed under GA4's AI Assistants channel. AI Overviews appear inside Google Search results, produce no distinct referrer, and count as organic search — Google explicitly excludes them from the AI Assistants channel. So a Gemini UTM tag tells you about assistant traffic, not about AI Overview impressions, which live in Search Console as part of your organic picture.
No. Gemini links to your real canonical URL, so manually tagging that URL with utm_source=gemini would stamp the tag onto every visitor — organic, email, social — not just people arriving from Gemini. You would mislabel all of them as Gemini traffic and destroy the accuracy you were trying to gain. Let Gemini apply its own tag; your job is to read it, not reproduce it.
Gemini has quietly started appending UTM parameters to the links it surfaces, giving analytics a reliable way to spot Gemini referrals even when the referrer is stripped — which happens often on mobile and Android. It's undocumented, so treat tagged visits as a floor, not a full count. In GA4, Gemini already lands in the native AI Assistants channel by referrer; the UTM tag supplements it. Don't tag your own URLs with utm_source=gemini — you'll mislabel organic traffic.
Get started → · ← All guides · Compare Kompozy vs other tools