Gemini UTM Tags: How a Solo SaaS Should Recount AI Referrals

Gemini started appending UTM tags to some outgoing links, and Google’s John Mueller confirmed on Reddit that he could see them. Search Engine Journal reported the change on 1 October 2026. An SEO Pulse note on 3 October added that Google still has not documented when the tags appear. For a one-person SaaS, that is the useful part. A slice of AI-referred sessions that used to fall into Direct can now be counted, if you read the parameter instead of waiting for a named channel in the default report.

This is not a new acquisition channel. It is a measurement correction. The visits were already happening. The ledger was wrong.

Why the old Gemini number was mostly a desktop story

Chat products often pass a referrer on desktop web and drop it inside an in-app browser. A community reply on the same Reddit thread said Gemini’s referrer mostly survives on desktop web, with reported pass-through from the mobile app around 9 percent, and that Android assistant invocations strip it entirely. Treat that 9 percent as a practitioner observation, not a Google figure.

When the referrer dies, GA4 files the session as Direct. Direct is the leftover bucket: no referrer, no campaign parameter, no click identifier. A founder who checks Organic Search on Monday never sees those sessions. A founder who checks Direct sees a blob that also contains typed URLs, bookmarks, email clients that strip referrers, and dark social. Gemini was mixed into that blob.

UTM tags survive webviews more often than the Referer header. They also land in raw server logs. That is why the change matters more than a small badge in the Gemini interface.

ChatGPT already tags some outgoing links, and only on answers grounded in web sources, not on answers drawn only from model memory. Google has not said whether Gemini follows the same rule. Until it does, assume the tag is partial. A missing tag does not prove Gemini never cited you. A present tag does prove that particular click came through a tagged Gemini link.

What to count, and what not to invent

Do not hard-code a parameter name you have not seen in your own logs. Google has not published the schema. Open a link from a Gemini answer on your own site, copy the query string, and store the keys you actually received. Common campaign keys are utm_source, utm_medium, and utm_campaign, but the live string is the source of truth.

Keep three counts separate for the next four weeks:

1. Sessions with a Gemini UTM, regardless of referrer.
2. Sessions whose referrer host is a Gemini property and that have no UTM.
3. Sessions that are still Direct, which you should not reassign.

The first count is the new floor. The second is the desktop-web remainder. Adding them is a better Gemini total than either number alone. Do not subtract them from Direct and call the remainder true direct. You still cannot see untagged in-app clicks.

A worked example makes the bookkeeping concrete. Suppose last week a docs site recorded 1,400 sessions. Default reports showed 620 organic, 410 direct, 180 referral, and 190 other. After you parse query strings you find 46 sessions with a Gemini UTM. Of those 46, 31 were already in Referral because the desktop referrer survived, and 15 were in Direct. You also find 22 referrer-only Gemini sessions with no UTM. The corrected Gemini total is 68, not 46 and not 22. Direct falls by 15, to 395, and you label the 15 as reclassified, not as new demand. Organic is unchanged. If you told yourself Gemini was about 30 visits, you were counting the referrer path only and missing the in-app slice.

That correction changes a staffing decision. A 30-visit channel does not earn a page rewrite. A 68-visit channel that converts to trial at half the rate of organic might still earn a single docs page, if the landing URLs cluster. It does not earn a new content program.

A one-person measurement pass

You can do this in an afternoon. No new analytics vendor is required.

1. Capture the raw query string. In GA4, keep the full page location on page_view. If you are on a lightweight stack, log request URL, referrer, and status for HTML responses only. Strip emails and tokens before storage. Keep 28 days.

2. Build a Gemini flag, not a channel rewrite. A session is gemini_utm if any observed key matches the string you copied from a live Gemini link. A session is gemini_referrer if the referrer host matches the Gemini host you actually see, and gemini_utm is false. Leave the default reports intact so you can compare them with the flag.

3. List landing URLs for gemini_utm sessions. Sort by sessions, then by sign-up or trial events. For a small SaaS the list is often short: a pricing page, one comparison page, and two docs articles. That list is the work queue. A citation that never receives a click is not on it.

4. Infer the job from the landing URL. Gemini does not hand you the user prompt. If three tagged visits land on a page titled “export invoices to CSV,” the cited job is export, not your homepage slogan. Pair that with queries you already rank for in Search Console, without mixing the two systems into one number.

5. Compare behavior only after a minimum sample. Wait until a landing URL has at least 30 tagged sessions before you rank it against organic. Under that, a single enterprise trial will distort the rate. Compare engaged sessions, trial start, and early churn if you have it.

6. Write one page for the cluster, not a sitewide refresh. If tagged visits concentrate on a missing answer, add that answer on the existing URL. Keep the title stable. Add a short worked example and a limit section that says what the product does not do.

7. Recheck after 14 days. Parameter names can change while the behavior is undocumented. If the key you stored stops arriving, do not delete the old flag. Add the new key and keep both. A broken flag looks like a traffic drop and is actually a parser gap.

Search Console does not count a Gemini click as a Search click, and a UTM on a Gemini link does not create a Search Console impression. A rise in tagged Gemini sessions with flat Search Console clicks means the assistant surface moved, not that rankings moved. Do not append your own campaign tags onto URLs you hope Gemini will cite. Keep the canonical clean and leave internal links parameter-free.

A small decision rule for the next month

Use the corrected count to pick one of three actions.

If tagged plus referrer Gemini sessions are under 2 percent of engaged sessions, record the flag and stop. The measurement is worth having. A content sprint is not.

If they are above 2 percent and cluster on one or two URLs, revise those URLs. Add the example a buyer would need after an assistant summary: price boundary, limit, setup step, and who it is not for.

If they are above 2 percent and scatter across the homepage, the assistant is citing the brand and not a job. Homepage citations rarely convert. Publish one specific page the assistant can point at, then watch whether the landing URL shifts over the following two weeks.

Ignore vanity totals. A week with 80 Gemini visits and one trial is a weaker signal than a week with 25 Gemini visits and four trials on the same docs URL.

FAQ

Does a Gemini UTM mean Google Search sent the visit?
No. It means the click came from a tagged Gemini link. Search Console clicks stay in their own report. Do not add the two numbers together.

Should I create a GA4 channel called Gemini on day one?
Not yet. Keep a separate flag until you have seen the live parameter for at least two weeks. Default channel groups hide parser mistakes. A flag does not.

What if I never see a UTM on my own site?
You may have no tagged clicks yet, or the tag may be limited to some answer types. Check server logs as well as GA4. Absence is not evidence that Gemini never mentions you.

Will the tag hurt rankings?
Unlikely if the canonical stays clean and you do not generate internal links with the parameter. Do not let the tag become a second indexable URL.

How is this different from pasting Search Console into a chatbot?
Search Console describes queries and pages on Google Search. The UTM describes a click from Gemini. One says what people searched. The other says which assistant click reached a URL.

After this week’s tagging change, an indie SaaS can stop treating a chunk of Gemini clicks as unexplained Direct, name the pages that actually receive them, and spend the next revision on those pages only.