You open your order list in the morning. There is a new purchase, payment is in, and the item needs shipping. You look for the customer’s journey in your site analytics, but there is no visit. If checkout happened on another platform, both reports can be telling the truth.
On 4 November, Google announced that it was rolling out direct checkout in Google Search AI Mode and the Gemini app for eligible Canadian retailers. Checkout uses the Universal Commerce Protocol (UCP), and the merchant remains the seller. The rollout starts with specific partners; the announcement does not say it is already available to every store or in Latvia.
Where did they find the product, and where did they pay?
Those are two different questions. A shopper can discover a product in an AI conversation and pay on your website. They can also finish checkout on the platform itself. In the second case, the order can exist in your system without a corresponding website session. Record the source only when the integration actually provides it. If it is unknown, label it as unknown.
Example: 12 orders, 10 website purchases
Imagine the store system records 12 paid orders in one day. Ten completed on the website and two on an external platform. In this example, site analytics correctly shows ten website purchases, while the order report shows all twelve. To reconcile them, match records by order ID and check where checkout happened. Track refunds separately so that the revenue comparison accounts for them.
The same order may also appear in the platform’s report. When reconciling sources, check that you have not counted the same ID twice. And keep purchases completed off-site out of your website conversion rate if the denominator still counts only website sessions.
Before launch, test one order from start to finish
Choose a product with variants and a discount. Check that the variant, availability, final price and shipping charges agree across the catalogue, the offer and the order. Then check that the order ID reaches your reporting and that your team can find, cancel or refund the order. If you cannot tell where checkout happened, document that limitation before comparing channels.
The best time to sort out reporting is while you still have just one test order. Later, the difference between ten and twelve may look like an analytics bug when it actually reflects two different checkout paths.