Google Gemini’s UTM-tagged links change how referral visits appear

Google Gemini’s UTM-tagged links change how referral visits appear

A referral visit used to arrive in GA4 with a small identity crisis. During 2025, Google began appending UTM parameters to some Gemini outbound links, a change covered in Search Engine Journal’s announcement. That gives store teams a clearer record of a Gemini click, though the business value of that click still needs to be earned, even with that clearer record.

A tagged Gemini session can pass campaign source and medium into GA4 instead of showing up as an unhelpfully bare referral URL. Google’s manual campaign dimensions documentation explains how UTM parameters carry campaign details into reporting. The labeling gets better. The interpretation is still your job.

AI-referral identification should extend beyond Gemini. Review recognizable traffic from ChatGPT, Perplexity, Microsoft Copilot, and other AI platforms alongside tagged Gemini sessions, then group each platform consistently by source, medium, referrer, and landing page. Some platforms provide campaign details, while others may appear as referral, direct, or unassigned traffic.

UTM tags identify a click, not the value created by that click. Someone may reach a waterproof trail running shoe guide through a Gemini link and become a clearly recognizable assisted-search visitor, but that session needs to open a relevant item or begin a cart before it belongs in a revenue conversation.

Preserve Gemini campaign values. Review them as their own acquisition segment. Channel labels are useful evidence, especially when they were previously buried inside broader referral traffic, but your assisted-discovery conversion definition determines whether that evidence should guide ecommerce investment decisions, including content updates.

What GA4 AI search attribution has to prove

An assisted-discovery conversion is an action showing that an informational visit moved a shopper closer to buying, even when the order comes through another channel later. It matters. Gemini’s tagged links make that path easier to isolate in GA4. They don’t prove every guide reader influenced a sale.

Lean teams need two conversion tiers:

1. A qualified product-path action happens when someone reads a trail running shoe fit guide and opens the waterproof model they were considering.
2. A high-intent action occurs when that shopper adds the shoe to cart, even if they return after checking the size chart.

An AI referral earns credit when it creates a measurable route toward purchase. Credit must be earned. Set the first tier as a secondary key event, then treat add-to-cart activity as stronger evidence. This keeps a quick article exit from receiving the same credit as a session that reaches a purchasable SKU.

A purchase-only model misses much of the research behind a $160 waterproof trail runner. Shoppers compare fit with outsole grip, then weigh weather protection before settling on a size, while the final order may arrive through email or a direct return visit and branded search can also contribute.

In our audits, teams often give every AI-referred article visit equal weight, only to discover that many of those sessions never reach a catalog item. The useful signal sits in the route from helpful guidance into a relevant model. That’s worth counting.

Keep informational entry pages separate from revenue pages

Before reviewing Gemini referral performance, create two page groups: discovery pages that answer a shopper’s question, and purchase pages that help them choose or buy. Keep them distinct. That distinction sets a fair expectation for each visit.

A trail running shoe outsole-grip article can link directly to a mud-focused waterproof footwear collection. That collection visit is its expected action. A size chart has a different assignment. Measure size selection and cart activity instead.

Entry-page intent determines the conversion you should expect. When you apply one success event to every educational URL, you hide the routes that deserve more editorial attention and make weak internal links look like weak referral traffic. That’s an expensive misunderstanding.

Build a simple GA4 exploration around the landing page first. Then add next-page path, so you can see whether the shopper enters the relevant catalog area, and include key-event completion. When those campaign parameters are present, limit the audience to Gemini-tagged sessions.

Most stores find that their strongest educational content needs one clearer route into the catalog before referral data becomes useful. A grip article ending with generic footer navigation tells you very little. Not much at all. A visible link to mud-ready waterproof options gives the visitor direction, while also giving GA4 something meaningful to measure.

Annotate Gemini referral changes before comparing performance

Campaign-tagged Gemini links can create a newly visible referral segment in GA4. Treat its first appearance as a measurement change. A larger labeled segment may reflect cleaner classification while shopper demand remains flat.

Use this setup process:

1. Add an annotation on the exact date tagged sessions first appear.
2. Compare post-click behavior against earlier direct traffic and generic referrals over matching windows.
3. Review landing-page engagement, product-detail progression, and the entry page’s defined key event.
4. Validate the same grouping rules across Gemini, ChatGPT, Perplexity, and Microsoft Copilot before comparing platform performance.

For example, a trail shoe brand may see newly labeled Gemini sessions land on its waterproof running shoe comparison guide after those visits previously appeared as direct traffic. The useful question is whether these recognized sessions behave differently after arrival. Do they continue into a waterproof collection or view a size-specific SKU?

A reporting change. It can look like a demand increase. Keep that sentence beside the annotation so nobody turns a tracking shift into a triumphant content-performance slide.

Across browsers and apps, referrer handling remains incomplete. The W3C Referrer Policy specification explains how websites can limit the referrer details sent with outgoing requests, leaving analytics platforms to work with partial signals. Cross-device journeys, consent choices, and later channel touches also limit platform-level attribution.

In our experience, equal date ranges and matching landing pages make these comparisons far more credible, and a short annotation log prevents teams from treating a measurement update as proof that their content suddenly started pulling harder.

Build an assisted-discovery conversion model for ecommerce content

The reporting shift matters because, before attribution means much, ecommerce content needs a clear commercial destination that lets you define an assisted-discovery conversion as the next buying-intent action fitting the question a shopper arrived to solve. That’s the goal.

Give each priority educational entry point a short conversion contract. Keep it focused. Write down the visitor’s question, where they should go next, and one tracked action that signals purchase progress, whether that action is selecting a variant or opening an item detail view.

Every discovery page needs a defined route toward merchandise. A trail shoe sizing guide should link directly to waterproof models available in half sizes when that’s the shopper’s practical need, rather than leaving them to search for those options elsewhere. Sending them back into a broad catalog makes the guide less useful and the data less revealing.

Start small. Begin with the small group of informational URLs that bring in the most visits, because a dozen well-defined routes are manageable and can be reviewed with confidence. Hundreds of loosely tracked articles, by contrast, create reports that nobody trusts, including the person who built them.

Inspect internal links near the answer. Before the visitor leaves, check whether product context appears, then make sure a fit guide leads into the relevant product family with availability and size options visible at the handoff. Make the handoff visible.

Flag pages that attract attention without moving shoppers toward merchandise. Give them stronger product context, place a relevant collection link beside the answer, add a model selector, or change the page’s role in the catalog. This review often exposes content that earns visits. No job after the first click.

Frequently asked questions

Sources

• Search Engine Journal’s announcement

Written by Richard Newton, Co-founder & CMO, Sprite AI.