Every e-commerce team has now been asked the same question by someone senior: do we show up when people ask an AI assistant what to buy? Until this year the honest answer was that nobody could tell you. Google has since built the measurement into Merchant Center, and paired it with six feed fields that change what the AI surfaces know about your catalogue.
Merchant Center's AI Performance Insights shows how your products appear across AI Mode in Search, AI Overviews and the Gemini app, using four sections: share of voice against similar merchants, shopping funnel performance, popular AI search terms, and missing product attributes. It launched as a limited US pilot around 13 July 2026 and is rolling out to the US, Canada, Australia, India and New Zealand. It is organic-only and reports no clicks metric.
What the report actually contains
AI Performance Insights was announced at Google Marketing Live 2026 and went live as a limited pilot for US accounts around 13–14 July 2026, with a rollout to the United States, Canada, Australia, India and New Zealand following. It reports on three surfaces: AI Mode in Search, AI Overviews, and the Gemini app. It sits inside Merchant Center rather than Search Console or Google Ads, which tells you what it is for — it is a catalogue-level view, not a page-level one.
Four sections do the work, and each one maps to a different action.
| Section | The question it answers | What to do with it |
|---|---|---|
| Share of voice | How often your brand appears in AI shopping responses relative to similar merchants | A relative benchmark, not a target. Track its direction over weeks; a single reading tells you almost nothing |
| Shopping funnel performance | Whether you appear at the discovery, evaluation or purchase stage | The most diagnostic section. Strong discovery and weak evaluation means your products are found but not comparable |
| AI search terms | Which conversational terms surface your products | Language input for product descriptions and attributes, not a keyword list to insert into titles |
| AI attributes | Which attributes are missing on products that appear in AI searches | The direct to-do list. Each missing attribute is a question an assistant could not answer about your product |
The funnel section is where most of the value sits, because it separates two failures that look identical in a share-of-voice number. Appearing in discovery queries ("good waterproof hiking boots") and disappearing by evaluation ("waterproof hiking boots under $200 with wide sizing") means your catalogue is visible but cannot be filtered. That is a data problem with a specific fix. Never appearing at all is a different problem with a different fix.
The limitation that should shape how you read it
In its pilot form the report is organic-only and carries no clicks metric. It tells you about appearance and share, not about traffic or revenue. This matters more than it sounds: you cannot tie a share-of-voice movement to a sales movement inside this tool, and anyone presenting it as an attribution report is over-reading it.
That constraint is easier to live with once you know which tool answers which question. This is the part most teams get wrong — they open one report expecting it to answer all four questions, then conclude that AI visibility is unmeasurable.
| Question | Tool that answers it | What it will not tell you |
|---|---|---|
| Do my products appear in AI shopping answers, and against whom? | Merchant Center AI Performance Insights | Clicks, revenue, or which specific query produced an appearance |
| Do my pages appear in AI Overviews and AI Mode? | Search Console generative AI report | Clicks, CTR or queries — see how to read that report |
| What do AI-referred visitors do once they arrive? | GA4, with the AI assistant channel separated | Anything about the answer that sent them, or about impressions you never converted into visits |
| What does an assistant actually say about my brand? | Manual prompting and third-party monitoring | Volume, share or any reliable sampling — treat it as qualitative evidence |
Used together these four cover the chain from appearance to revenue with one honest gap in the middle: nobody currently reports clicks from AI surfaces separately, in any of these tools. Plan on inference, not on precision, until that changes.
The fix side: six conversational attributes
Reporting a gap is only useful if there is a lever. The lever Google shipped alongside the report is a set of six optional conversational attributes in Merchant Center, launched at the same event and available in all countries. Three of them did not exist in the standard product feed at all: question_and_answer, related_product and document_link. The others specify product families, variant axes, and an internal popularity percentile.
| Attribute | What it supplies | When it earns its place |
|---|---|---|
| question_and_answer | Product-level Q&A pairs in your own words | Highest value first. Every pre-purchase question a customer emails you is an answer an assistant currently has to guess |
| related_product | Typed relationships between catalogue items | Accessories, refills, compatible parts, and anything where "what else do I need" is part of the decision |
| document_link | Crawlable PDFs: manuals, spec sheets, sizing and care guides | Technical and considered purchases where the detail lives in a document, not the description |
| Product family title | The name of the family a variant belongs to | Catalogues where one product exists as dozens of near-identical listings |
| Variant axes | Which dimensions actually vary: size, colour, capacity | Anywhere assistants need to distinguish variants rather than treat them as competitors |
| Internal popularity percentile | Your own ranking of what sells | Broad catalogues where the assistant otherwise has no basis to pick a representative product |
Two mechanics matter. Google recommends supplying these through a supplemental feed joined on id rather than rewriting your primary feed, which keeps the change additive and reversible. And these attributes do not affect Shopping approval — they are not a compliance risk, so the usual argument for deferring feed work does not apply.
Which products to fix first
Nobody completes 40,000 SKUs of attributes. The prioritisation that works is a three-filter pass, in this order.
- Revenue and margin percentile. Start with the products that carry the business: the top decile by contribution margin, not by unit volume. Attribute work is a fixed cost per SKU and the return scales with the value of the product.
- Products already appearing in AI searches with missing attributes. The AI attributes section names these directly. They are proven to surface, so completing them raises quality of appearance rather than gambling on new appearance.
- The answerable-question test. For each candidate SKU, ask whether a customer's pre-purchase question has a factual answer you can state in one sentence. "Does this fit a 2019 model?" qualifies. "Is this good?" does not. Products that fail this test gain little from Q&A attributes and should get relationship and document data instead.
Sequence the fields as Google's own guidance implies: product Q&A first, then catalogue relationships, then linked documents. The first is the highest-information addition per unit of effort, because it encodes knowledge that currently only exists in your customer service inbox.
Pull the last 300 pre-sale customer service enquiries and sort them by frequency. The top 20 questions, answered factually against the SKUs they concern, will do more for how an assistant describes your products than another round of description rewriting. It is also the one input a competitor cannot copy from your product page.
A 30-day sequence
- Days 1–3. Confirm whether AI Performance Insights is available in your Merchant Center account and market. Screenshot or export the current share of voice and funnel readings; this is your only chance to record a pre-change baseline.
- Days 4–7. Export the AI attributes gap list. Join it to your revenue data and cut it to the top decile by contribution margin.
- Days 8–14. Build the supplemental feed. Start with Q&A pairs drawn from real customer enquiries, then variant axes and family titles for any catalogue with heavy variant duplication.
- Days 15–21. Add related products and document links. Check that every linked PDF is crawlable and not behind a login or a JavaScript viewer, which is the most common reason this field does nothing.
- Days 22–30. Re-read the funnel section. You are looking for movement from discovery-only appearance toward evaluation, which is the change these attributes are designed to produce. Share of voice will move more slowly and more noisily.
Product page work still matters underneath all of this — a feed cannot describe a product that the page itself never explains. Our guide to Shopify product page SEO covers the page-side half, and the feed should agree with it rather than substitute for it.
Three things not to do
- Do not paste AI search terms into product titles. The terms report describes how people phrase things conversationally. Keyword-stuffing a title with conversational phrasing degrades the shopping listing you already have, and title quality is a Shopping ranking input in its own right.
- Do not treat share of voice as a KPI with a target. It is a relative measure against a competitor set you do not control, on surfaces that are changing weekly. Use its direction over a quarter, not its level in a month.
- Do not fabricate attributes to fill gaps. An assistant confidently repeating a wrong specification is a returns problem and a trust problem. Blank is better than wrong, and this is one place where the incentive to complete a field can quietly cost you money.
Frequently asked questions
What is AI Performance Insights in Google Merchant Center?
It is a Merchant Center report showing how your products perform across Google's AI shopping surfaces: AI Mode in Search, AI Overviews and the Gemini app. It covers share of voice against similar merchants, shopping funnel performance across discovery, evaluation and purchase, popular AI search terms, and missing product attributes on items that appear in AI searches.
Who can access the report and in which countries?
It was announced at Google Marketing Live 2026 and went live as a limited pilot for US accounts around 13 to 14 July 2026, with a rollout to the United States, Canada, Australia, India and New Zealand following. Availability is account by account, so check your own Merchant Center rather than assuming market-wide access.
Does the report show clicks or revenue from AI surfaces?
No. In its pilot form it is organic-only and carries no clicks metric. It measures appearance and share, not traffic or revenue, so it cannot be used as an attribution report. For page-level AI impressions use Search Console's generative AI report, and for post-click behaviour use GA4.
What are conversational attributes and do I need them?
They are six optional Merchant Center fields launched at Google Marketing Live 2026 and available in all countries: product question and answer pairs, related products, document links, a product family title, variant axes and an internal popularity percentile. They supply AI surfaces with detail your standard feed does not carry. They are optional and do not affect Shopping approval.
How do I add conversational attributes to my feed?
Google recommends a supplemental feed joined to your primary feed on the product id, rather than rewriting the primary feed. This keeps the change additive and reversible. They can also be supplied through the primary feed or the Merchant API if that suits your setup better.
Which products should I complete attributes for first?
The top decile by contribution margin, filtered to products the AI attributes section already flags as appearing in AI searches with gaps. Within those, prioritise products where customers ask factual pre-purchase questions you can answer in one sentence, since product Q&A carries the most information per unit of effort.
The takeaway
The useful shift this year is not that AI shopping surfaces exist; it is that you can finally see whether you are in them and act on a named list of gaps. Read the funnel section rather than the share-of-voice headline, accept that no tool currently reports AI clicks, and spend the effort where it compounds: real answers to real pre-purchase questions, attached to the products that actually carry your margin. That work improves your product pages, your Shopping listings and your AI visibility at the same time, which is the only kind of AI optimisation worth doing.
Sources & further reading
- Google Merchant Center Help, “Insights for AI-powered shopping experiences”
- Search Engine Land, “Google launches AI Performance Insights and Conversational Attributes in Merchant Center”
- Google Merchant Center Help, “How to use conversational attributes”
- Productsup, “Google introduces six conversational attributes in Merchant Center”