Merchant Center AI Performance Insights: Share of Voice and Product Gaps

Google Merchant Center now has an AI performance report built around a question ecommerce teams have struggled to answer:

When shoppers use conversational Google experiences, which products and brands appear, at what stage of the shopping journey, and for which product needs?

The report is useful because it introduces first-party Google evidence for AI shopping visibility. It is also easy to overstate. A share-of-voice percentage is not universal market share. A popular product term is not the complete prompt a shopper typed. An AI impression is not a website visit, and none of those signals proves a purchase.

Short answer: Use Merchant Center AI performance insights as a provider-specific discovery instrument. Record availability, product category, country, time period, Organic AI scope, Google's fixed competitor set, and the exact metric definition. Review share of voice across discovery, evaluation, and purchase; connect popular terms and missing structured attributes to governed product-data changes; then validate the changed feed, landing page, and matched AI visibility Runs separately. Do not convert the report into a universal AI rank, traffic forecast, or revenue claim.

This guide explains the report as documented on August 6, 2026 and turns it into an auditable ecommerce AEO workflow.

What Google Currently Documents

Google's current Merchant Center AI performance insights help page describes a report focused on how a merchant's brand and products appear in AI Mode and AI Overviews.

The detailed page says the report is currently:

• A pilot for a limited number of Merchant Center accounts in the United States.
• Planned to expand to Australia, Canada, India, and New Zealand.
• Limited to conversational queries with shopping or brand intent.
• Filterable by product category, time period, country, and traffic scope.
• Limited to Organic AI traffic, such as free listings, rather than paid Ads traffic.

An earlier May 2026 rollout announcement described discovery across AI Mode, AI Overviews, and the Gemini app. The current detailed help page names AI Mode and AI Overviews when defining the report.

Treat that difference as a versioned reporting boundary. Record what the interface and current documentation say when you export the data. Do not silently expand a current total to every Google or Gemini surface because an earlier announcement used a broader example.

Report Unavailable Is Not Zero

Pilot access creates an important status distinction:

Preserve unavailable, insufficient, and measured_zero as separate values in any warehouse or spreadsheet. Converting all three to numeric zero destroys the meaning of the report.

The Merchant Center Metrics At A Glance

The report combines competitor-relative exposure with demand and product-data diagnostics.

Keep Google's native names in the raw export. Add internal labels in a separate field instead of overwriting the provider definition.

How Merchant Center AI Share Of Voice Works

Google defines the percentage around AI impressions for the target and the competitors available in Merchant Center.

For a simplified representation:

Merchant Center AI share of voice = target AI impressions ÷ (target AI impressions + available competitor AI impressions)

The formula needs four qualifications.

1. Google Controls The Competitor Set

The current help page says merchants cannot change the competitors used by the report. The cohort comes from the competitors available in Merchant Center.

That makes the metric different from a custom competitor benchmark in which the analyst freezes named brands and aliases. The provider-controlled cohort may not match the brands your merchandising, category, or executive teams consider strategically relevant.

Report it as:

Merchant Center AI share of voice for Google's available competitor set

Do not shorten it to:

Our AI market share

For a custom competitor cohort with explicit denominators, use the AI share-of-voice measurement guide and keep answer coverage, mention share, recommendation share, and citation share separate.

2. The Report Is Category-Scoped

Google says there is no single report covering every product category. A percentage for running shoes should not be combined with a percentage for hiking boots as if the populations were identical.

Export category as part of every row. If executives need one summary, show category-level results in a table rather than calculating an undocumented average.

3. Zero And One Hundred Need Diagnostic Checks

Google documents two display behaviors that can make a percentage look more decisive than it is:

• Insufficient impressions can display as 0 share of voice.
• Insufficient competitor data can display as 100 share of voice.

Add a status note beside both values. A bare 0% or 100% is not decision-ready evidence.

4. AI Impressions Are Not Answer Roles

The percentage does not tell you whether a product was:

• Named neutrally.
• Recommended.
• Compared favorably or unfavorably.
• Shown as a free listing.
• Used as a source for a product fact.
• Presented at a stage where the shopper was ready to buy.

That answer role requires separate evidence. A stable AEO Table Task can preserve exact buyer Questions, competitors, answer text, citations, market, and repeated Runs. Merchant Center supplies a Google shopping aggregate; it does not replace the answer record.

Use The Three Shopping-Journey Phases Deliberately

Google groups conversational shopping activity into three phases:

1. Discovery: the shopper is exploring possible product types or solutions.
2. Evaluation: the shopper is comparing options, features, or specifications.
3. Purchase: the shopper is closer to a transaction.

The phases are useful planning buckets, but they are not a complete behavioral funnel. Google notes that one complex query can map to more than one category.

Build a separate question set around each phase.

Do not use invented brand-leading prompts as the core baseline. A Question such as "Why is Acme the best shoe?" does not measure unprompted discovery. The AI search query set guide explains how to separate discovery, problem, comparison, proof, and branded Questions.

Freeze The Phase Mapping

If your team maps Questions differently from Merchant Center, retain both fields:

• google_journey_phase
• internal_intent_class

This prevents the internal taxonomy from being mistaken for a Google field and lets the team compare provider classifications without erasing them.

Turn Product Terms Into A Review Queue

Merchant Center describes frequently used AI shopping terms as wording around functional benefits or product features that shoppers prioritize. The report can show the term, its popularity, how many products match, and the corresponding share of voice.

A low-match, high-frequency term is a useful review candidate. It is not an instruction to insert the phrase everywhere.

Use this sequence:

1. Confirm that the term describes a real product property or supported use case.
2. Identify the relevant product IDs and variants.
3. Check the current feed value, landing-page copy, images, and applicable structured data.
4. Verify that the claim is accurate and supported.
5. Update only the fields and pages where the concept is genuinely relevant.
6. Validate feed acceptance and landing-page consistency.
7. Annotate the release and compare a later matched period.

For example, if "maximum cushioning" is frequent but only a few eligible shoes match, first determine whether the missing products actually meet the merchant's governed cushioning definition. Do not convert demand language into a product claim the item cannot support.

Google's product data optimization guidance recommends detailed, accurate, current product data and consistency between the feed and landing page. That is the operational standard to apply before chasing a new conversational term.

Treat Missing Attributes As Product-Data Work

The report's popular-attribute view focuses on structured properties such as size, color, or material. Google's detailed help page says the current report provides visibility for structured attributes.

Review the complete source chain:

source-of-truth catalog → Merchant Center data source → landing page → structured data → rendered offer

A feed-only edit can create disagreement with the site. A page-only edit can leave Merchant Center with stale or missing values.

Google's supported structured data mapping connects Merchant Center fields with schema.org properties and explains that structured data can help Google retrieve current product and offer details. It also notes that not every Merchant Center attribute has a schema.org equivalent.

Use a field-level audit:

Accuracy is more important than field completion. A complete but false attribute is a product-data defect, not an AEO improvement.

A Reproducible Merchant Center AI Workflow

Step 1: Record Access And Scope

Capture:

• Merchant Center account and authorized analyst.
• Report availability status.
• Export timestamp and timezone.
• Product category.
• Country.
• Date range.
• Traffic scope.
• Current documentation URL and review date.

If the tab is unavailable, stop the provider-report workflow and record that state. Do not fabricate a substitute total from ordinary Merchant Center or Search Console metrics.

Step 2: Save The Native Baseline

Export or transcribe the baseline before changing product data.

Preserve:

• Target share of voice.
• Competitor-average share.
• Journey-phase results.
• Query types.
• Frequent product terms.
• Popular structured attributes.
• Product counts and displayed popularity indicators.
• Any insufficient-data messages.

Screenshots help with interface context. A structured export is better for comparison. Use both when policy and access allow.

Step 3: Select One Governed Opportunity

Choose a term or attribute only when it is:

• Relevant to an important product category.
• Factually true for identifiable products.
• Connected to a meaningful shopping decision.
• Actionable in the catalog, feed, page, or offer data.
• Large enough to justify a controlled change.

Avoid editing dozens of unrelated fields in one release. A bundled rewrite makes later movement impossible to diagnose.

Step 4: Build The Product Evidence Map

For every affected product, map:

• Product ID and variant ID.
• Canonical landing URL.
• Current title and description.
• Relevant feed attributes.
• Current visible specifications.
• Current Product structured data.
• Price, availability, shipping, and return-policy sources.
• Owner and last verified time.

This map catches disagreement before the change reaches shoppers or Google.

Step 5: Define The Hypothesis Without A Guarantee

Use a falsifiable statement:

For products that genuinely have attribute X, completing X consistently across the catalog, feed, landing page, and structured data will remove a documented information gap. We will observe whether Merchant Center AI exposure and matched answer evidence change in a later comparable period.

Do not write:

Adding attribute X will increase AI share of voice by 20%.

The report does not provide a causal forecast.

Step 6: Make And Validate The Smallest Change

After publication:

• Confirm Merchant Center accepts the product data.
• Check affected product status and diagnostics.
• Render every changed landing page.
• Confirm the final canonical URL.
• Validate visible product facts.
• Validate Product and Offer structured data where used.
• Confirm price and availability agree across systems.
• Save the release timestamp and product IDs.

Step 7: Run Matched AI Visibility Checks

Use stable Questions tied to the product need, not the report's aggregated label alone.

Keep constant where practical:

• Question wording and intent class.
• Product category.
• Market and language.
• Answer-engine surface.
• Device or account conditions that affect the result.
• Competitor cohort.
• Attempt and retry policy.

Capture mention, recommendation, comparison, citation, accuracy, and handoff separately. The AEO content experiment guide explains how to use predeclared hypotheses, controls, pre/post Runs, and cautious interpretation.

Step 8: Compare Without Claiming Causality

Review three evidence lanes:

1. Merchant Center aggregate movement.
2. Prompt-level answer and citation movement.
3. Website visits, key events, and sales.

Movement in one lane can occur without movement in the others. Product demand, inventory, pricing, seasonality, competitor offers, model changes, and reporting coverage can all move during the same period.

Report the association and the alternative explanations. Repeat before treating a small change as durable.

Merchant Center, Search Console, GA4, And AEO Monitoring

These systems answer different questions.

Google's Search generative AI performance announcement lists impressions, pages, countries, devices, and dates for its dedicated Search Console views. That is a site-exposure report, not the same product and competitor instrument Merchant Center provides.

For cross-provider boundaries, use the Google-versus-Bing AI visibility report guide. For post-click measurement, use the GA4 AI referral tracking guide.

A CSV Schema For The Review

Keep provider data and analyst decisions in separate columns.

exported_at, report_version_note, account_id, report_available, availability_note, country, product_category, period_start, period_end, traffic_scope, google_journey_phase, query_type, product_term, attribute_name, target_ai_share, competitor_average_share, display_status, matching_product_count, popularity_label, product_ids, canonical_urls, internal_intent_class, source_of_truth_owner, change_id, release_at, validation_status, pre_task_id, post_task_id, analyst_note

Do not put raw customer data, private prompts, or unsupported inferred demographics into the export. Follow the account's data governance and access policy.

Common Mistakes

Do not call report unavailability zero visibility.

Do not call Google's available competitor cohort your complete market.

Do not average category percentages without a documented common denominator.

Do not treat query frequency as exact search volume or a verbatim prompt count.

Do not add popular terms to products that cannot substantiate the claim.

Do not update the feed while leaving the landing page or structured data inconsistent.

Do not combine organic AI exposure with paid Ads traffic.

Do not convert an AI impression into a session, conversion, or sale.

Do not attribute a later increase to one feed edit without matched evidence and repeated observation.

The Bottom Line

Merchant Center AI performance insights create a valuable first-party view of conversational shopping visibility. Their strongest use is not a headline score. It is the connection between a provider-defined competitor benchmark, shopping-journey phase, product language, structured attributes, and a governed product-data backlog.

Preserve the native scope. Diagnose 0% and 100% before celebrating or escalating. Make only accurate product-data changes. Validate the catalog, feed, page, and markup together. Then compare the provider report with stable answer Runs, GA4, and commerce outcomes without pretending those systems share one denominator.

Create an AEO Table account to monitor stable shopping Questions, compare brand and competitor answers, preserve cited evidence, and review whether product-data changes are associated with durable answer visibility.