AI & Automation

How Do You Measure Traffic From AI? Gemini Started Adding UTM Tags on Oct 1, and Only About 9% of App Clicks Keep a Referrer: How to Fill the Gap in GA4's AI Assistant Channel

2026.10.04 · 49 views
How Do You Measure Traffic From AI? Gemini Started Adding UTM Tags on Oct 1, and Only About 9% of App Clicks Keep a Referrer: How to Fill the Gap in GA4's AI Assistant Channel
“

On October 1, Google Gemini was spotted adding UTM parameters to outbound links, which Google's John Mueller confirmed seeing on Reddit. The next day, an apparent Google test of tracking parameters in AI Overviews turned out to be a Chrome extension rewriting URLs. We unpack the blind spots in GA4's AI Assistant channel, compare five tools, and show how small sites can measure AI traffic without buying SaaS. ”

Share:

Short answer: The most reliable way to measure AI traffic today combines GA4's AI Assistant channel, Gemini's new UTM parameters and server-side logging of AI crawlers. GA4 alone undercounts, because only about 9% of clicks from the Gemini mobile app carry a referrer; the rest mostly land in Direct.

On October 1, 2026, Search Engine Journal reported that a member of Reddit's SEO community noticed Google Gemini adding UTM tracking parameters to outbound links in its answers. Google Search Advocate John Mueller replied that he saw it too. Commenters in the same thread said only about 9% of traffic from the Gemini mobile app passes referrer information to websites. In other words, many sites that concluded "AI sends us no traffic" over the past year simply could not see it.

Why now? AI assistants have moved from chat tools to search entry points, and brands are asking what GEO investment actually buys. In May 2026 Google added an AI Assistant channel to GA4's default channel group, automatically labeling sessions from ChatGPT, Gemini, Claude and others with the medium "ai-assistant". But the channel depends on referrers sent by the browser; in-app browsers and copied links slip through. Our own September server logs showed ChatGPT-User fetching pages 501 times, 499 of them English pages and only 2 Chinese pages. Whether AI visited, and what it read, is invisible in GA4.

Meanwhile AI visibility tools are racing to fill the gap: Profound and Otterly.AI count brand mentions in AI answers, while Semrush and Ahrefs add AI visibility to existing suites. But October 2 brought a warning. Search Engine Roundtable briefly reported that Google was adding "st_source=ai_overview" to AI Overview links, before discoverer Tom Pool corrected it: a Chrome extension was rewriting URLs, presumably to sell data to brokers. Tracking parameters can be faked; measurement is expanding and getting messier.

For SMBs the impact is direct: if you are deciding whether to invest in GEO or English content, fix measurement first, or every decision rests on numbers missing half the picture. Below: how to read the parameters, which tools to consider and what to do for free.

Event details and numbers

Bottom line: Gemini's UTM tags give mobile-app traffic its first real chance at correct attribution, but Google has not documented the rules.

  • Oct 1: A Reddit user spotted UTM tags on Gemini outbound links; Mueller confirmed seeing them and asked, "Doesn't Gemini have a referrer too though?"
  • Key number: Commenters said the referrer mostly survives on desktop web, but only about 9% passes through from the Gemini mobile app.
  • Undocumented: Google has not said when parameters are added or whether values are fixed.
  • Oct 2: The "st_source=ai_overview" parameter on AI Overviews was not added by Google but by a Chrome extension; the reporter could not reproduce it in multiple browsers.
  • Existing rule: Per Google Search Central, clicks from AI Overviews and AI Mode are included in Search Console's Performance report under the "Web" search type and cannot be separated.

What three kinds of readers should do now

Bottom line: owners open the report, marketers build a custom channel, developers log on the server.

  • Brand owners: Open GA4 and record 90 days of AI Assistant sessions and conversions as a baseline.
  • Marketers / SEOs: Following the GA4 custom channel group guide, build an "All AI" channel matching medium ai-assistant plus sources containing gemini, chatgpt, perplexity, claude and copilot. Do not add utm_source=gemini to your own URLs; it mislabels other visitors.
  • Developers / agencies: Log AI crawler User-Agents (such as ChatGPT-User, PerplexityBot) server-side and review weekly which pages are fetched.

AI traffic measurement tools compared

Bottom line: free tools measure clicks, paid tools measure mentions, and they answer different questions.

ToolWhat it measuresCostBest for
GA4 AI Assistant channelAI clicks and conversions with a referrerFreeEvery site's first step
Google Search ConsoleTotal search clicks including AI OverviewsFreeOverall Google exposure; cannot isolate AI
Server logs / custom bot logWhich pages AI crawlers fetchedEngineering hoursTeams wanting to see what AI reads
Otterly.AI, ProfoundBrand mentions and citations in AI answersMonthly subscriptionCompetitive brands with marketing budget
Semrush, Ahrefs AI visibility featuresBrand and domain exposure in AI answersAdd-on to existing suitesTeams already using these SEO suites

What the hype won't tell you

Bottom line: more parameters do not mean more accurate numbers, and AI traffic should not be judged by clicks.

  • Contrarian view 1: tracking parameters create new problems. An extension could fake "st_source=ai_overview", and any URL parameter can be rewritten. Building a KPI on one parameter is risky.
  • Contrarian view 2: most AI value is not a click. Users see your brand in a conversation, remember it, then search your name or call you; reports show that as Direct or branded search. Judging GEO by AI-channel sessions always undercounts.

SMB alternatives without a SaaS subscription

Bottom line: one GA4 custom channel, one bot log table and a monthly prompt test cover about 80% of needs.

  • Step 1: Build an "All AI" custom channel in GA4 and set inquiry form submissions as a key event.
  • Step 2: Log AI crawler User-Agent, URL and language server-side for 90 days; in Laravel this is one middleware and one table.
  • Step 3: Each month, ask ChatGPT, Gemini and Perplexity the same 10 customer questions and record whether you are mentioned and which page is cited.
  • Step 4: Add "How did you hear about us?" to the inquiry form, with an AI assistant option.

FAQ

Can GA4 show traffic from ChatGPT?

Partly. Clicks with a referrer go to the AI Assistant channel; in-app opens and copied links mostly count as Direct.

Should I add utm_source=gemini to my URLs?

No. AI shows your original URL, so adding the parameter yourself mislabels visitors from other sources.

Can Search Console show AI Overview clicks?

Not separately. Google includes AI Overview and AI Mode clicks in the Web performance report.

Do SMBs need an AI visibility monitoring tool?

Most don't yet. Start with a GA4 custom channel, server bot logs and a monthly prompt test, then reassess after three months.

My take

My view: for SMBs, Gemini adding UTM tags does not mean "finally we can measure it"; it proves everyone underestimated AI traffic over the past year. Most people are waiting for Google to ship a complete AI traffic report before committing to GEO. I don't think that report is coming. Google already blends AI Overview clicks into regular search data in Search Console, because separating them does nothing for its ad business.

Second view: within 12 months, AI visibility monitoring will be absorbed into large SEO suites as an add-on, so SMBs need not rush to subscribe.

For ScriptWalker (a Taiwan-based Laravel/Flutter custom development studio), the takeaway: we already log AI crawlers on our own site. This "GA4 custom channel + bot log table + monthly prompt test" setup can be packaged as a one-off implementation of one to two working days, cheaper than any annual SaaS contract, and the client owns its data.

Sources

Share:
AI & Automation Back to Blog