Google Merchant Center AI Performance Insights: AI Share of Voice and the 6 Conversational Attributes

Until now, the only way to answer "how often does our product get mentioned in AI search?" was to estimate it with third-party tools. On September 16, 2026, Google announced that AI performance insights in Merchant Center is generally available to businesses in Australia, Canada, India, New Zealand, and the United States^[2]. Google now reports brand share of voice inside AI answers directly to sellers. The same announcement pushed merchants to submit conversational attributes and shipped changes to the UCP integration hub^[4].

This page closes the loop on four things: what the metrics actually count, which filters and edge cases will wreck your reading of them, how to get the six conversational attributes into your feed, and what a Korean account outside the eligible markets can use in the meantime.

30-second definitions

• AI performance insights is a Merchant Center report that compares your brand's and products' share of voice against competitors, scoped to shopping-intent conversational queries in Google AI Mode and AI Overviews^[1].
• Your share of voice is your AI impressions on relevant queries divided by the sum of your impressions and the impressions of competitors defined in Merchant Center^[1].
• Conversational attributes are six optional feed attributes that help AI systems understand product nuance. They supplement the existing product data specification^[3].
• UCP (Universal Commerce Protocol) is the commerce integration spec that connects a merchant's product catalog, checkout, and payments to Google Search, AI Mode, Gemini, and other surfaces^[2].
• Business Agent lets a brand answer a shopper's multi-part questions directly inside an ad or a search result. The YouTube ads beta opened in the United States^[2].

How the report is built

Metric and filter definitions

The report runs on four metrics and four scope filters. Reading the metric names alone will mislead you almost every time, so start with the denominator^[1].

Source: Google Merchant Center Help documentation on AI performance insights (2026)^[1].

Four edge cases are worth knowing before you open the report. Share of voice shows 0 when impressions are insufficient and a dash when there is no impression data at all. Accounts with no competitors defined in Merchant Center see 100%, which means the comparison set is empty rather than that you own the category. The competitor set itself is not user-editable^[1].

Timeline, from pilot to general availability

In July 2026 a pilot was spotted on a subset of US Merchant Center accounts, alongside the Help documentation, with expansion to Australia, Canada, India, and New Zealand already signalled^[6]. In early September 2026, AI Search intent, AI Search terms, and AI attributes sections appeared in the report^[5], and Google made general availability official on September 16^[2]. MediaPost reported that the announcement came at the Rethink ROI Summit^[7].

Two months from pilot to general availability is fast. The report surface keeps moving, which means an internal reporting template pinned to dashboard screenshots will break quickly. Anchor on the metric definitions instead.

The six conversational attributes, and how to submit them

Conversational attributes belong to the Merchant Center product data specification, and all six are optional. You can add them to your primary product data, though a separate supplemental data source is the recommended route, and the Merchant API accepts them as well. Adding them does not affect the approval status of products already in your account^[3].

Source: Google Merchant Center Help documentation on conversational attributes (2026). Example values are taken from that documentation^[3].

The only public number behind the effect is a Lululemon test: Google says brand-submitted conversational attributes were reflected in 50% of AI Mode's related product recommendations^[2]. In the same announcement, Google added that merchants who adopted core feed best practices saw conversions rise 5% on average the following month^[2]. Both figures come from Google, so treat them as claims to re-validate against your own data.

What changed in the UCP integration hub

The UCP integration hub in Merchant Center picked up new capabilities. Google says UCP already supports direct checkout for hundreds of thousands of brands and retailers across its surfaces, and merchants who use UCP checkout can now switch on cart transfer to their own site and an improved checkout flow test through the hub. Analytics is listed as coming later. The new features roll out gradually in the United States, with Australia and Canada flagged for early next year^[2]. Search Engine Roundtable lists three integration items in the hub: merchant cart transfer, native checkout on Google, and identity linking^[4].

Tapestry, owner of Coach and Kate Spade, was named as an example of a merchant that connected UCP checkout and began selling through Search, AI Mode, and the Gemini app^[2]. MediaPost read the update as a response to shrinking consumer decision timelines and cited NetElixir's forecast that US holiday ecommerce revenue will grow 9% year over year^[7].

Measurement surfaces compared, and what Korean accounts use

AI performance insights is limited to English-language queries in accounts based in the five eligible countries^[1]. Korean brands either cannot open the report at all or, if they can, look at data that says nothing about domestic demand. Separating the measurement surfaces makes it obvious where the gap is.

Among global tools, Profound^[8] and Peec AI^[9] fall into that category. In Korea, Next-T^[10], LeadGenLab^[11], Ascent AI^[12], and BOIDA^[13] state that they handle Korean-language queries. BOIDA is a brand operated by Designovel, and its product is BVI (Brand Visibility Index)^[13][14]. Every profile and function description here follows what each company states on its own site and is not third-party verified. Pricing follows each vendor's published rates and shifts over time, and BOIDA is quoted on inquiry.

The short version: a platform report tells you your relative share inside Google's own surfaces, while third-party measurement covers the rest of the engine set and Korean-language queries. The two divide the work rather than substitute for each other. If share-of-voice measurement is new to you, read How to measure AI search share of voice and Google Search Console generative AI performance reports first.

Execution sequence

1. Confirm you can open the report. Sign in to Merchant Center, go to Analytics, then Products, then the AI performance tab at the top. No tab means you should check the eligible country list and the eligibility requirements^[1].
2. Check your competitor set. A 100% share of voice means no competitor data exists. Since you cannot change the set, drop share of voice as a metric in that case and work from products showing and frequency instead^[1].
3. Find the gaps by stage. Across discovery, evaluation, and ready to buy, prioritize search types where frequency runs high but your share of voice runs low. That is the judgment rule the Help documentation gives^[1].
4. Fill in popular attributes. The popular attributes scorecard surfaces items missing from your product data, things like size, color, and material. Work down from the top. Share of voice does not appear on this scorecard, so treat it purely as a list of missing attributes^[1].
5. Put top search terms into titles and descriptions. Write the scorecard's top search terms into product titles and descriptions in natural language. Google's recommended order puts top search terms and missing attributes first^[1].
6. Submit conversational attributes. Create a new supplemental data source and start with question_and_answer, variant_option, and item_group_title. Existing approval status is unaffected, so you can roll out incrementally^[3].
7. Connect video links and loyalty data. Google lists two best practices here: use the video link attribute so shoppable ad formats can be built automatically, and connect loyalty data to surface member benefits and pricing^[4].
8. Review the UCP hub options. Merchants already on UCP checkout should decide whether to enable cart transfer and the checkout flow test. Outside the United States, check the rollout schedule^[2].
9. Cover the Korean-language gap. If you are not in an eligible country, track Korean-query exposure separately with Search Console and third-party AI visibility measurement. The feed work overlaps with Agentic commerce GEO strategy and the ChatGPT shopping feed setup guide, so bundle it into one pass.

In practice, the value of this report sits in the scorecards, not the share numbers. Missing attributes and top search terms convert straight into feed edit tickets. The numbers are for the report deck; the list of things to fix is what the team actually uses.

Wrap-up

AI performance insights is Google officially reporting brand exposure inside its own AI surfaces back to sellers. The scope, though, is English queries in five countries, the data is organic AI traffic only, and the competitor set is fixed^[1]. The accurate way to use this report is therefore not as a measurement tool for total AI visibility but as a work order telling you where to start fixing your feed inside Google's surfaces. The six conversational attributes and the UCP hub options are the implementation side of that work order^[3][2]. For a Korean brand, the job right now is covering the Korean-query territory this report does not yet reach with separate measurement.