Merchant Center AI Performance Insights in 2026: A Practical Share-of-Voice Audit

Retailers can now see part of a shopping journey that was previously difficult to measure: whether their products appear when people ask conversational shopping questions in Google AI Mode and AI Overviews. Merchant Center AI performance insights turns that visibility into a report with share of voice, shopping-stage, term, intent, attribute, and product-appearance signals. The useful question is not simply whether the number is high. It is which product-data gap is limiting discovery and whether a controlled correction improves qualified visibility without damaging catalogue accuracy.

Google's current AI performance insights documentation says the report covers organic AI traffic, not paid ads. It currently applies to English-language queries for eligible Merchant Center accounts in Australia, Canada, India, New Zealand, and the United States. A retailer operating from Armenia should therefore confirm the account country, target market, language, and report availability before promising an AI-visibility dashboard.

Define what the report measures before setting a target

The report focuses on conversational queries with shopping intent in AI Mode and AI Overviews. It does not measure every mention of a brand across the open web, every Gemini interaction, paid Shopping traffic, or completed revenue. Its share-of-voice percentage compares your AI impressions with the impressions of the competitor set defined by Merchant Center for related queries.

This makes the report a discovery diagnostic, not a complete attribution system. Use it to identify where product data may be weak, then connect the resulting work to Merchant Center diagnostics, site analytics, checkout data, and business outcomes. A visibility lift without accurate price, availability, margin, and conversion evidence is not yet a commercial win.

Confirm eligibility and build a clean baseline

Google made AI performance insights generally available in five markets in September 2026, according to its agentic commerce update for retailers. Availability can still differ by account, query language, category, and data sufficiency. Open Merchant Center, go to Analytics, select Products, and look for the AI performance tab. Record the account, country, product category, time range, and export date for every review.

Do not combine unlike categories into one conclusion. The report requires a single product-category view, while demand, competitor sets, attributes, and shopping paths can differ sharply between shoes, electronics, cosmetics, and furniture. Build a category baseline and preserve it before changing titles, descriptions, or attributes.

Read the core metrics together

Signal	What it can tell you	What it cannot prove
Your share of voice	Your relative AI impressions within Google's defined comparison set	Revenue, profitability, or visibility across every AI platform
Competitors' average share	Whether your visibility is above or below the current comparison baseline	Why a competitor appears or whether the set reflects your strategic competitors
Frequency	Which terms, intents, attributes, or search types carry more demand	Exact query volume or guaranteed conversion value
Products showing	How often products appeared cumulatively during the selected period	A count of unique products or completed purchases

A 100% share can be misleading when Merchant Center lacks sufficient competitor data. A zero can mean insufficient impressions, while a dash can mean no impression data. Competitor and brand averages may also rise together when the competitor set changes. Document these conditions rather than turning every movement into a performance story.

Diagnose the shopping stage, not only the average

Merchant Center groups conversational activity into discovery, evaluation, and ready-to-buy stages. A brand can be visible during broad exploration but disappear when shoppers compare specifications, look for reviews, or approach a transaction. The corrective action should match the stage.

Discovery

High-frequency feature or category questions with low share of voice can indicate that product titles and descriptions do not express how the product is used or what problem it solves. Improve useful, factual language without turning the feed into a keyword list.

Evaluation

Weak visibility for comparisons or specifications often points to missing material, size, compatibility, technical, image, or variant data. Make the structured attributes, feed, landing page, and visible product details agree.

Ready to buy

Late-stage gaps can come from incomplete price, availability, shipping, return, seller, or landing-page information. Before changing copy, run the parity checks described in Boost iT's product structured data audit guide.

Turn terms, intents, and attributes into an evidence queue

The report surfaces top terms, popular attributes, and top search intents. Treat them as three different inputs. A term describes language shoppers use. An intent explains the job behind the query. An attribute is structured product information that may be missing. The strongest opportunity is a high-frequency pattern with weak share of voice, relevant products, and a verifiable data gap.

Create a backlog with category, stage, signal, frequency, current share, affected products, missing source field, proposed change, owner, release date, and expected business behavior. Reject suggestions that are irrelevant, promotional, unsupported, or inconsistent with the physical product. Google's product data specification requires accurate titles, descriptions, identifiers, variants, images, links, and commercial data; an AI insight does not suspend those requirements.

Protect one product source of truth

Do not edit the feed, structured data, landing page, and internal search independently. Decide which system owns each field and how it reaches every channel. If a material is added because the report shows demand for it, that material must be true, visible to the shopper, associated with the correct variant, and maintained when the catalogue changes.

For complex catalogues, Boost iT's web-development service can help connect catalogue fields, server-rendered product pages, structured data, feeds, analytics, and release checks. The goal is a reliable data pipeline, not a separate layer of AI-oriented copy that drifts away from inventory and checkout.

Use conversational attributes selectively

Google's conversational attributes guidance describes optional question-and-answer, document-link, related-product, item-group-title, and variant-option inputs. They can add product nuance for conversational surfaces, but they complement the core product specification rather than replacing it.

Start only where the source is authoritative: a verified compatibility table, care guide, ingredient sheet, technical manual, or genuine product FAQ. Do not manufacture answers because a query appears popular. For regulated, safety-sensitive, or high-return categories, require product-owner and legal review before publishing new claims.

Run a controlled product-data experiment

Select one eligible category with enough data and a clear high-frequency, low-share opportunity.
Choose a limited product group and a comparable holdout group where practical.
Capture titles, descriptions, attributes, landing pages, share of voice, products showing, diagnostics, clicks, conversions, returns, and margin before release.
Correct only verified gaps and keep feed, structured data, visible page, and checkout information consistent.
Allow for the report's daily update and several-day lag; do not judge the change immediately.
Compare visibility and downstream quality over a stable period, accounting for promotions, stock, price, and seasonality.
Expand only when accuracy remains intact and the business signal is directionally useful.

Do not make dozens of simultaneous catalogue changes and then credit one dashboard movement to the project. Preserve release notes so the team can explain what changed and roll back unsupported edits.

Build a measurement ladder beyond share of voice

Use four levels. First, data quality: missing attributes, disapprovals, page-feed mismatches, and stale inventory. Second, AI discovery: share of voice, frequency, stage, terms, intents, attributes, and products showing. Third, qualified behavior: landing-page engagement, product views, add-to-cart, assisted conversions, and new-customer quality where measurable. Fourth, economics: completed orders, margin, cancellations, returns, and support cost.

The report is organic-only, so do not merge it with paid Shopping results without clear labels. Keep country, language, category, date range, and competitor-set caveats visible in every review. This prevents a useful diagnostic from becoming a false attribution claim.

Run a focused 30-day AI visibility audit

Days 1-7: access and baseline

Confirm account eligibility, English-query scope, countries, categories, data lag, competitor limitations, and current product diagnostics. Export a baseline and map product-data owners.

Days 8-14: opportunity mapping

Prioritize high-frequency, low-share terms, intents, attributes, and stages. Inspect the affected products and verify each proposed field against a trusted catalogue source.

Days 15-21: controlled release

Correct a limited product set, maintain landing-page and feed parity, document the release, and check for disapprovals, broken variants, price conflicts, or indexing regressions.

Days 22-30: evaluate and operationalize

Review visibility after the reporting lag, compare downstream quality, keep only accurate improvements, and assign a monthly category review with owners and acceptance criteria.

Conclusion: use AI visibility as a product-data feedback loop

Merchant Center AI performance insights gives retailers a practical view of conversational shopping discovery, but its value comes from disciplined interpretation. Confirm availability, segment by category and stage, diagnose verifiable data gaps, test controlled corrections, and connect visibility to customer and commercial outcomes. Do not confuse an organic AI impression share with paid-media performance or revenue.

For a bilingual audit of Merchant Center AI visibility, product-data quality, landing-page parity, measurement, and a controlled optimization backlog, request an e-commerce SEO audit from Boost iT. The output should be a prioritized implementation plan, not a promise of guaranteed placement in AI results.