Halfway through a Q4 planning meeting, someone always asks the question. How does our brand actually show up in AI search? There was no official answer before, so teams pieced together screenshots from third-party tools. Since September 16, there is an official report in Merchant Center, and Google calls it AI performance insights.

Google announced a batch of merchant tool updates at a retail event that day, and AI performance insights opened to eligible accounts in five countries. Ashish Gupta, Google's vice president and general manager of merchant shopping, said in a post that businesses need to learn how customers discover them through conversational experiences (the event and post via MediaPost; the availability news via Search Engine Roundtable).

For merchants, the verdict up front. The report is worth opening now, and the way to use it comes down to three moves. Keep a baseline, change one batch of feed data, and leave enough window for comparison. Read it as a sales report and the direction is wrong from the first line.

What the report gives you

The official help page is clear about positioning. It shows how your brand is discovered across Google's generative AI ecosystem, specifically for conversational queries with shopping intent on AI Mode and AI Overviews. You find it in Merchant Center under the Analytics tab, on the AI performance tab under Products.

Memorize the scope first. It is currently available to eligible accounts in Australia, Canada, India, New Zealand and the United States, and it covers English-language queries only. Other markets and languages are not visible yet, and that is not an account settings issue.

Four groups of metrics are worth knowing. Your own share of voice, based on your brand's or product's AI impression share for related queries. Competitors' average share. Frequency, which reflects how popular specific search types, intents or attributes have been. And products showing, the number of your products appearing for top queries.

The report also sorts conversational queries into three stages, discovery, evaluation and ready to buy, with share of voice shown for each stage. Below that sit opportunity scorecards for top terms, missing attributes and high-frequency search intents, each expandable.

The scope filters come with limits. The data covers organic AI traffic only, so paid ads are not included. Updates are daily with a few days of lag. Your comparison set is the competitor set Merchant Center defines for you, and it cannot be changed at present. And no single view covers all product categories, so record your filter settings when you export.

Two details are worth knowing early. When impressions are insufficient, share of voice shows as 0 or a dash; and when an account has no usable competitor data, share of voice shows 100%, which only means there is nothing to compare against. The report is also young, and Google says more features are coming, so keep your baseline and settings saved for future comparisons.

How to read it, starting with putting the share in its place

Share of voice measures visibility. It does not answer traffic or conversion questions, and it does not include revenue. The report itself suggests reading it alongside Search Console, your ads account and analytics, not instead of them.

Keep the comparison conditions tight. Same country, same category, same date window, same filters; if one is missing, before and after do not line up. Read the shopping stages separately too, because weak spots often show up in only one stage, for example a steady discovery share with a sharp drop at evaluation.

The opportunity scorecards are leads, not conclusions. The missing attributes and top terms they flag need to be checked against the product itself, and only accurate data should be submitted. And do not rush to judge from same-day readings; a lag of a few days is normal for this data pipeline.

Five feed moves to make first

First, pick products by the gap. High frequency, low share, commercially important. Start with combinations where all three hold, and do not roll changes across the whole store.

Second, map each gap to a concrete change. Missing structured details such as material, size or compatibility call for attribute completion. Accurate high-frequency terms absent from titles or descriptions should be added naturally. A repeated pre-purchase question becomes a question-and-answer entry. Muddled variants call for item group and variant option data.

Third, fill in conversational attributes where they help. The optional set includes question and answer, document link, related product, item group title, variant option and popularity rank. You can submit them through a supplemental data source, your primary data source or the Merchant API, and Google recommends starting small with a supplemental feed as a reversible path. Details already covered in the description or product detail do not need to be repeated. Adding these attributes will not impact the approval status of existing products, which the official docs state clearly.

Fourth, put the video attribute to work. The video link attribute accepts product videos, up to ten links, and the videos must follow Shopping policies. One detail worth noting, replacing the content behind the same link slows down recrawling, so use a new URL when the video itself changes.

Fifth, connect loyalty data only when it fits. Merchants with an existing loyalty program can connect it to surface member perks and pricing across Google's shopping surfaces. Building a token program just to make the report look better serves no one.

What about UCP integration for now

In the same batch of updates, Google simplified onboarding to the UCP (Universal Commerce Protocol) integration hub. Merchants who use the hub can enable cart transfer and enhanced checkout flow testing, with analytics coming soon after. UCP's core capabilities include cart transfer, native checkout on Google and identity linking, all aimed at bringing checkout into AI surfaces.

Watch the timing. The new capabilities are rolling out gradually in the United States, with Australia and Canada expected to follow next year. If your market is not in the first wave, there is nothing to do yet beyond staying informed.

The decision on moving early is not complicated. UCP and your ad campaigns run on separate tracks, so skipping it does not stop you from advertising. What decides the pace is whether your US checkout is a weak spot and whether your platform supports it. If you do move, treat it as an integration project and walk cart, variants, discounts, tax, shipping, cancellation and mobile flows through a sandbox first. Do not treat it as a toggle.

The timing before Q4

This week, confirm whether your account can see the report, and export the first baseline, recording the filters, definitions and date window together.

Next week, pick one category and make the first small batch of changes through a controlled path. Then give it one to two weeks for the reporting lag to settle, and recheck with the same settings, not from same-day readings.

Before the Q4 peak, finish the first test round and take the conclusions into holiday planning. Keep structures stable during peak and leave the second round for after the holidays. The lightest useful move right now is to save a baseline while the report is new and the definitions are still clean. Every later comparison starts from there.

What not to do

Do not read it as a traffic or conversion report. Share of voice is a visibility metric, sales are a transaction result, and mixing them produces wrong conclusions.

Do not blend paid and organic conclusions. The report covers organic AI visibility only, so take paid questions back to your ads account.

Do not change the whole catalog at once, and do not judge results the same day. A few days of lag is the rule, not a fault.

Do not copy suggested terms blindly, and do not stuff attributes for coverage. Accuracy is the premise of this dataset.

Keep the coverage boundary in mind too. Five countries and English queries are today's edge.