Are your products showing up in AI shopping? How to read Merchant Center's report
Merchant Center's AI performance report shows how often your products appear in Google's AI shopping experiences, such as AI Mode and AI Overviews, compared with a set of competitors. It covers organic appearances only, so it tells you nothing about your Shopping ads. Before acting on it, check what each number counts. A 100% share of voice, for example, is what Google shows when your account has no defined competitors, which means there was nothing to compare against.
Read carefully, the report is a good way to decide which product facts to check next. This guide covers where to find it, what each metric means and how to turn one finding into one verified feed change.
Where to find the report
In Merchant Center, open Analytics, then Products, then the AI performance tab. According to Google's documentation, it currently covers English-language queries for Merchant Center accounts in Australia, Canada, India, New Zealand and the United States. Historical data updates daily with a lag of a few days.
- Merchant Center
- Analytics
- Products
- AI performance
- Country
- Time period
- One product category at a time
You filter by country, time period and one product category at a time. Google says there's no report covering every category, so a catalog-wide picture means working through categories one by one. Write the country, category and dates beside any screenshot you share, because two people looking at different categories can reach opposite conclusions from charts that look alike.
If the tab isn't there, check eligibility before reading anything into its absence. An account outside the supported markets has no report, and that tells you nothing about what shoppers see.
What each number means
| Metric | What Google says it measures | Watch for |
|---|---|---|
| Your share of voice | Your AI impressions divided by the total impressions for your defined competitors and you, for related queries. | The three special readings in the next section. It isn't shown on the popular attributes scorecard. |
| Competitors' average share | The share of voice captured by your defined Merchant Center competitor set. | You can't choose the competitors. When the set changes, your share and theirs can move together. |
| Frequency | How popular a search type, term, intent or attribute has been. | This describes shopper demand. Your own performance shows up in the other columns. |
| Products showing | How many of your products appear for top terms, popular attributes and search intents. | Google counts cumulative appearances, so one product can be counted several times. Zero means none of your products showed. |
| Shopping stages | Discovery (exploring options), evaluation (comparing options or checking specifications) and ready to buy (close to a transaction). | Each stage groups the questions shoppers were asking at that point in the journey. |
Products showing is the number most likely to be misquoted in a meeting. Because it's cumulative, a reading of 40 doesn't mean 40 different products appeared. Use it as a trend within the same filters, and don't add it up across categories.
Check for zero, a dash and 100% first
Google documents three share-of-voice readings that cause most of the confusion:
- 100%Your account doesn't have defined competitors. Report it as "no competitor comparison available."
- 0You don't have enough impressions to calculate a share. That's worth investigating as low visibility, though it isn't a measured zero.
- -A dash: There's no impression data at all for the selection. Check the filters and eligibility before concluding anything about shoppers.
Misreading these costs real time. A team that takes 100% as dominance stops working on a category that may need help, and a team that takes a dash as invisibility can spend a sprint rewriting healthy product pages to fix a gap in the report.
Use the trends to choose what to check
The report lists top terms, popular attributes and top search intents, each with a frequency and a products-showing count. Together they tell you what shoppers in the category are asking about. What your products actually are still has to come from your own specifications, so the useful move is to take one popular attribute or intent and check whether your product data answers it truthfully.
| Report signal | Check first | Change if a gap is confirmed |
|---|---|---|
| A popular attribute, such as a material or finish | The supplier specification for each variant | Add the verified value to the matching feed attribute and the product page. |
| An intent about a use or setting | Product documentation, care and installation instructions | Explain the use plainly in the description, including its limits. |
| A size or compatibility question | Dimensions and compatibility data | State dimensions and compatibility explicitly in the feed and on the page. |
| A share of voice that looks too good | Whether competitors are defined, and the filters used | Qualify the finding before deciding the category needs no work. |
A popular attribute is a prompt to check your data. If your product doesn't have the feature, leave it out, however often shoppers ask for it.
A worked example: one material, one verified change
A hypothetical furniture merchant filters the report to outdoor seating in the United States and sees frame material among the popular attributes. The example uses no real impression counts and predicts no lift.
The team pulls one dining chair. Its supplier specification says the frame is powder-coated aluminum. The submitted feed calls it a "metal chair," and the product page never names the material. That's a real gap, and it can be verified.
The change is small: add the material to the feed's material attribute, name it in the description and on the product page so all three agree, and leave the seat fabric, finish and care details as they are. "Rustproof," a weight rating and "all-weather" each need their own evidence before they go anywhere near the feed, because filling gaps from general category knowledge is how accurate-looking feeds end up wrong.
Before the edit goes live, the team records the product ID, the field, the old and new values, the report filters and the dates it will compare. Our guide to conversational attributes in Merchant Center covers the implementation side.
What the report can and can't prove
After the change, answer three questions separately:
- Did the corrected data reach Merchant Center, and does the product page still match it?
- Did the relevant reading move in a comparable period, with the same country, category and competitor situation?
- What happened to visits and confirmed purchases, measured in your analytics and your store?
The AI performance report can help with the second question only. It holds no visit or purchase data, so it can't show that a feed change earned sales, and a share-of-voice rise during a month when the competitor set changed may have nothing to do with your edit. If the comparison stays muddy, report what changed and what you observed, and leave the causal claim open.
Two related reads: our comparison of popular products and popularity rank covers Google's demand signals, and our guide to Merchant Center brand profiles covers brand-level controls.
Where to start this week
Pick one category that matters commercially and read its share of voice with the 100%, 0 and dash cases in mind. Then take its top popular attribute and check ten products against their specifications, which is usually enough to show whether the feed has a pattern of missing facts. If it does and you want help fixing it across the catalog, talk with us about Merchant Center enrichment.
