Google has started sending notifications to Merchant Center users whose e-commerce platforms support Universal Commerce Protocol, informing them that their products are now eligible for native checkout. The key detail buried in that notification: it is opt-out, not opt-in. If you have not actively turned it off, your products may already be purchasable directly inside AI Mode and Gemini.
That is a significant shift. Not because native checkout is new as a concept, but because it is now appearing inside AI search surfaces - the very surfaces that brands are trying to measure, attribute, and build visibility programmes around. The mechanics of AI search measurement were already challenging. This makes them harder.
What Native Checkout in AI Mode Actually Means
When a user finds your product through an AI Overview, AI Mode, or a Gemini recommendation and completes a purchase inside that surface, the transaction happens without a session on your website. No landing page load. No GA4 session. No server-side event fired from your domain. The order exists in your fulfilment system, but the journey that produced it is invisible to your standard analytics stack.
This is not a tracking gap in the usual sense - a pixel that failed to fire, or a cookie that was blocked. It is a structural absence. The traffic never came to your site. If your attribution model depends on sessions, page events, or on-site conversion tags, it will record nothing. The sale will appear to have no source.
For brands running AI search visibility programmes, this is the core problem. You may be earning genuine recommendations and citations in Gemini. Those citations may be driving real commercial outcomes. But if those outcomes complete inside the AI surface, your CRM will show revenue without a channel, and your AI search programme will show no measurable return.
The Attribution Problem Is Not Theoretical
AI search attribution is already difficult. GA4 underreports direct AI referrals. Many AI surfaces do not pass referrer strings consistently. Traffic from Perplexity, ChatGPT, and AI Overviews frequently lands as direct or unattributed. Practitioners running AEO programmes have been building workarounds - UTM-enriched links in cited content, server-side tracking, CRM source tagging, branded search lift analysis - precisely because the default setup does not capture AI-influenced journeys.
Native checkout in AI Mode removes the click entirely. There is no URL to attach a UTM to. There is no session to tag. The order data that arrives in your system will carry whatever identifiers Google passes through its Universal Commerce Protocol integration - and whether that includes source attribution meaningful enough to use in CRM reporting depends entirely on how the integration is built and what your platform exposes.
The immediate practical question is: what order-level data does your platform receive when a sale completes via UCP? If your platform supports UCP, check the order source field in your back-end right now. If it is blank, or tagged in a way that does not distinguish AI Mode from other Google surfaces, you have a measurement gap that will compound as AI search volumes grow.
Opt-Out Has Its Own Costs
The obvious response is to opt out - and for many brands, that may well be the right call. Keeping the customer journey on your own site preserves session data, enables on-site personalisation, protects your first-party data collection, and keeps the conversion event within your tracking infrastructure. For any brand serious about measuring AI search revenue, site-side journeys are significantly easier to attribute.
But opting out is not cost-free. Native checkout reduces friction. A user who can buy inside Gemini without navigating to a new page converts at a different rate than one who must click through, wait for a page to load, find the product again, and re-enter payment details. If competitors remain opted in, they may take transactions from users who prefer the lower-friction path.
The decision should not be made on convenience or default inertia. It should be made deliberately, with a clear view of what you can and cannot measure in each scenario, and what that means for how you report AI search programme performance internally.
What to Do Before You Decide
First, check whether you have received the notification. If your platform supports UCP, the auto-enrolment may already be active. Log into Merchant Center and review your checkout settings. If native checkout is enabled and you did not consciously turn it on, that is the starting point.
Second, audit what data your platform receives with UCP orders. Talk to your platform provider or development team. Understand what fields are populated on orders that originate from Google's native checkout flow, and whether those fields allow you to distinguish AI Mode or Gemini as a source. If the answer is no, factor that into the decision.
Third, if you do remain opted in, build a compensating measurement approach. That means looking for order volume patterns that correlate with AI visibility changes, monitoring branded search lift in GSC alongside Merchant Center performance data, and flagging UCP orders as a distinct segment in your CRM even if source attribution is incomplete. Imperfect data is still useful data if you label it honestly.
The Broader Signal for AI Search Programmes
What this development illustrates is that AI search is no longer just an awareness and consideration channel. It is completing transactions. That has been the direction of travel with agentic commerce and AI shopping features for some time, but native checkout inside Merchant Center - auto-enabled, rolled out quietly via email notification - is a concrete step in that direction rather than a hypothetical one.
For brands that have been treating AI search visibility as a top-of-funnel concern and deferring measurement, this is a prompt to move faster on attribution infrastructure. If transactions are completing inside AI surfaces and you are not capturing them, the gap between what your programme actually delivers and what you can demonstrate it delivers will widen.
The measurement problem in AI search has always been that the surfaces sit outside the infrastructure brands built for web analytics. Native checkout makes that problem more acute, not less. Getting ahead of it - understanding your data, making the opt-in or opt-out decision deliberately, and building compensating measurement where needed - is the practical work that needs doing now.