AI Performance Insights Merchant Center: AI Mode Visibility
A chunk of the buying journey has moved somewhere you have no search terms report and no click data. Someone asks AI Mode for “lightweight road running shoes with a wide toe box under $150”, gets a product shortlist back, and either you are in it or you are not. Until recently there was no way to find out which — until AI Performance Insights arrived in Merchant Center.
Google has started closing that gap with a Merchant Center report called AI performance insights. It shows how your brand and products surface in AI Mode, AI Overviews and the Gemini app, and introduces a share-of-voice metric — the conversational equivalent of auction insights. It is currently a pilot limited to a small number of US accounts, with stated expansion to Australia, Canada, India and New Zealand.
Here is what it measures, what it structurally cannot tell you, and — the useful part — what to do with your feed right now while you wait for access. Preparation takes weeks; access can arrive any day.
What AI Performance Insights in Merchant Center is
AI performance insights is a reporting surface in Merchant Center, not an ad product. It describes organic visibility on AI surfaces: how your products show up in Google’s conversational answers. Paid placement is explicitly out of scope.
Location: Merchant Center → Analytics → Products → the “AI performance” tab at the top of the page. No tab means your account is not in the pilot, and there is no documented way to request access.
Google’s help text frames the purpose as helping merchants understand “how your brands show up in conversational results” and “view trends to help optimize your product data”. The second half matters more than the first: this is designed as a pointer to gaps in your feed, not as a trophy cabinet of metrics.
The four metrics
| Metric | Definition | How to read it |
|---|---|---|
| Share of voice | Your AI impressions divided by total impressions across you and your competitors for related queries | Relative. A drop can mean you slipped or that a strong competitor entered |
| Competitors’ average share | The share held by the competitor set configured in your Merchant Center | The benchmark is only as good as that list. Wrong list, meaningless comparison |
| Query frequency | How popular particular query types, product terms, journey phases and attributes are | A demand signal — the most useful input for assortment decisions |
| Query type | How the question is phrased: browsing a category, researching specs, looking for reviews | Independent of funnel stage. Tells you the language people use |
All four are impression-based. There are no clicks, no CTR, no conversions and no revenue — more on that below.
The journey breakdown
A genuinely useful cut: three stages, each with sub-categories.
- Discovery — category, features, style. The shopper does not yet know what they want.
- Evaluation — specs, reviews, comparisons. Choosing between options.
- Purchase — ready to buy, pricing, local availability.
Share of voice usually differs sharply by stage, and that difference is diagnostic. Strong at purchase and weak at discovery means something specific: you get found by people who already decided, but you are absent while they are deciding. That tells you to fix descriptions and attributes, not price.
Attribute completeness
The most actionable block is the attribute view: which product characteristics (colour, style, material) come up most often in conversational queries, plus an attribute completeness score showing which of your products are missing them. It is effectively a generated task list — “shoppers keep asking about material, and 40% of your SKUs have it blank.”
The report does not tell you how to rank in AI answers. It tells you which attributes people ask about and which of your listings lack them. That is the difference between chasing “SEO for AI” and doing ordinary data quality work.
What the report does not contain
This matters more than the metric list, because it determines which decisions you must not base on it:
- Clicks. No click metric appears in any description of the report. Impressions are not tied to site visits.
- Conversions or revenue. The documentation does not connect AI impressions to sales and describes no attribution model.
- Keywords. You get query types and product terms, not a list of actual phrasings.
- Paid traffic. Data is “strictly limited to Organic AI traffic (for example, free listings).” Paid results on AI surfaces are excluded.
- Non-shopping queries. Only conversational queries that are “mainly of shopping intent and brand related” are counted. Everything else is excluded.
- An all-category rollup. The filter works one product category at a time, and Google states there is no report covering all categories. Account-level totals have to be assembled by hand.
Which leads to the single most important caveat: AI Mode share of voice is not a KPI you should bonus a team against. It is a leading indicator of data quality, not an outcome metric. Share of voice rising while revenue stays flat is an entirely normal state of affairs, not a win to report.
Availability
As of the most recent public reporting: a pilot with a limited number of Merchant Center accounts in the US. Google’s text mentions expansion to Australia, Canada, India and New Zealand “in the coming months”, with no date attached. The report first appeared in Merchant Center help documentation in late May 2026; live screenshots from advertisers surfaced in mid-July.
No brand-verification or minimum-traffic requirement appears in any public description, and there is no documented way to apply. The tab simply appears for selected accounts.
For merchants outside those markets the practical conclusion is simple: you do not have this report and may not for some time. That is a reason to prepare, not a reason to ignore it.
Five things to do before you have access
Everything that drives visibility on AI surfaces happens in the feed, and all of it can be done today. The report only reveals the result.
1. Fill the attributes people actually ask about
Conversational queries are built differently from keyword searches. Nobody types “Nike running shoes size 10” into a conversation — they describe a problem: “lightweight road shoes with a wide forefoot”. Answers to that are assembled from attributes, not titles. The minimum set worth driving to 100% coverage: material, colour, size system, age group, gender, compatibility, intended use. Do it at scale with feed rules and supplemental feeds rather than rebuilding your CMS export.
2. Adopt conversational attributes
A dedicated set of fields Google built specifically to match products to conversational queries. These are the input and AI performance insights is the output: with the attributes empty, the report will show you nothing. Full breakdown in our article on conversational attributes in Merchant Center.
3. Configure your competitor set
The “competitors’ average share” benchmark is computed from the competitor list you configured in Merchant Center. If it still holds whatever retailers someone added during an audit two years ago, your benchmark is noise from day one. It is a five-minute fix that also improves your price competitiveness and best sellers reporting.
4. Take an AI traffic baseline in Search Console
When the report arrives you will need something to compare it against. Search Console records organic impressions and clicks, including from AI surfaces — capture a product-page snapshot now so that in three months you have a before. On reconciling organic with paid, see our guide to the paid and organic report.
5. Rewrite titles into descriptive language
A title like “SKU 4471-B Black” takes no part in conversational matching at all. The structure “brand + product type + key attribute + size/volume” works in Shopping and on AI surfaces alike. Structure guidance is in our piece on product titles in Merchant Center.
How conversational queries differ from search queries
Feed work only makes sense once you understand what actually changes in the phrasing. Three structural differences:
Length and structure. A search query is two to four words, often brand plus model. A conversational one is a sentence full of conditions: “I need a laptop for video editing that runs six hours on battery and weighs under 1.5kg.” That is three filters at once, and only a product with populated technical fields can answer it.
Constraints instead of keywords. People describe boundaries rather than products: budget, delivery date, compatibility, size limits. Half of those constraints live in operational feed fields — price, availability, shipping settings — not in the description. An empty or stale shipping field removes a product from “I need it by Friday” answers even when the specs are a perfect match.
Comparison happens inside the answer. In classic results the user does the comparing. In conversational results the system does, and for that it needs comparable fields across products. If one SKU lists material as “100% cotton”, another as “cotton”, and a third leaves it blank, you lose the comparison on data consistency rather than on product quality.
Which yields a dull but decisive rule: consistency of attribute values matters more than the number of attributes. Ten fields populated to one standard outperform thirty populated ad hoc.
A one-hour feed audit for AI surfaces
A sequence you can run today for measurable improvement, with or without report access:
- 15 minutes — feed diagnostics. Open Merchant Center diagnostics and write down the warnings, not just the errors. Warnings are where unpopulated optional attributes hide, and those are precisely what conversational matching uses.
- 10 minutes — coverage of key fields. Calculate the share of SKUs with material, color, size, gtin, age_group and gender populated. Anything under 90% goes in the backlog.
- 10 minutes — value consistency. Export unique values per attribute. If the colour field holds 400 variants instead of 20, your problem is normalisation, not completeness.
- 10 minutes — titles. Check the first 70 characters of your top 50 revenue products: do they contain product type and a key attribute, or do they open with a SKU code?
- 5 minutes — competitor list. Tidy it in Merchant Center settings.
- 10 minutes — baseline. Capture Search Console data for product pages: impressions, clicks, average position, 28-day window. Save the file with a date on it.
The underrated step is the third. Normalising values usually delivers more than adding new fields, and it can be done with feed rules without involving developers.
Building a proxy metric without report access
You cannot compute true share of voice without the report, but you can get close. Three sources that together give a workable picture:
- Search Console. Track impressions for product pages separately from sitewide traffic. A sharp impression change at a stable average position is circumstantial evidence of movement on AI surfaces.
- A manual query panel. Build a list of 20–30 typical conversational phrasings for your category, run them monthly, and record whether your products appear. Crude and laborious, but directional — and unlike the report it gives you the actual phrasings.
- Merchant Center cross-platform reporting. It shows how products behave across Google surfaces and helps separate a general decline from a localised one. Covered in our piece on Merchant Center cross-platform reporting.
The crucial part is documenting method and dates. When the official report arrives, a homemade baseline lets you interpret what came before instead of starting observation from zero.
Why this matters for paid campaigns
The report is organic, but the work it highlights feeds straight into paid performance. The reason is technical: modern formats increasingly assemble ad copy from feed attributes. AI Max already generates Shopping ad copy from fields such as material and fit. A richer feed means better raw material for whatever the platform writes on your behalf.
That produces a useful prioritisation argument: investment in feed completeness pays twice — in organic AI visibility and in the quality of automatically generated paid ads. It is a rare case of one task covering two channels. For the paid side of the same surface, see our article on advertising in Google AI Mode.
How not to fool yourself with share of voice
- It is a share, not a volume. Yours can rise simply because a competitor trimmed their catalogue. Always read it alongside query frequency — a rising share in falling demand is not an achievement.
- It is impressions, not money. With no clicks or revenue in the report, any link to sales stays a hypothesis. Test it against your own organic traffic and revenue in the same categories.
- It is one category at a time. There is no rollup, so “store-level share of voice” is a number you constructed. Document your method or next quarter’s comparison will be meaningless.
The fair analogy is auction insights in Google Ads: useful for understanding the competitive field, useless as a target. How to read that class of metric is covered in our piece on impression share and auction insights.
The takeaway
AI performance insights is the first official Google tool that measures product visibility in conversational results. It is raw: a US pilot, impressions only, one category at a time, no clicks. But the direction matters more than the current feature set — Google is acknowledging that AI surfaces are a distinct channel and giving merchants a metric for it.
If you do not have access yet, none of the preparation requires it. Complete attributes, conversational fields, descriptive titles, a current competitor list and a captured baseline add up to a few weeks of work that pays off in organic visibility and in paid campaigns at the same time. When the tab appears, you want it showing results rather than an empty report.
If you want help wiring the feed to your campaigns, PPC Rebels works through these setups directly — and launching Shopping campaigns without review interruptions is considerably easier on a stable base, which is what Google Ads agency accounts provide from day one.
FAQ
What is AI performance insights?
A Merchant Center reporting section showing how a merchant’s brand and products appear in AI Mode, AI Overviews and the Gemini app. It reports organic visibility and is not an ad format.
Where do I find it?
Merchant Center → Analytics → Products → the “AI performance” tab at the top of the page. The tab is visible only to accounts included in the pilot.
Which countries is it available in?
As of the latest public reporting, a pilot with a limited number of accounts in the US. Google has mentioned expansion to Australia, Canada, India and New Zealand “in the coming months” without naming dates.
How do I apply for the pilot?
There is no documented way to request access. The tab appears automatically for selected accounts, so check Analytics → Products periodically.
How is share of voice calculated?
Your AI impressions divided by total impressions across you and your competitors for related queries. The exact formula and whether it is computed at brand or product level are not fully disclosed in public documentation.
Where does the competitor list come from?
From the competitor settings already configured in Merchant Center. The “competitors’ average share” benchmark depends entirely on that list, so clean it up before access arrives.
Does the report include clicks or conversions?
No. Every described metric is impression-based; there is no mention of clicks, CTR, conversions or revenue, and no attribution model is described.
Is paid traffic included?
No. Data is limited to organic AI traffic, such as free listings. Paid results on AI surfaces do not appear here.
Which queries count?
Only conversational queries that are mainly shopping-intent or brand related. Per Google’s wording, anything else is not factored into these insights.
Can I see all categories at once?
No. The filter works one category at a time and Google states no all-category report exists. Account-level views must be assembled manually.
What is the attribute completeness score?
An indicator showing which of your products lack the attributes shoppers most often mention in conversational queries. In practice it is a ready-made backlog for feed work.
Should AI Mode share of voice be a target metric?
No. It is a leading indicator of data quality rather than a performance measure: it is disconnected from clicks and revenue, expressed in relative terms, and dependent on a competitor list you chose yourself. Sales and profitability remain the targets.
