Merchant Center AI insights: turn visibility into a product-data test

Google announced broader availability for Merchant Center's AI performance insights on September 16, 2026. The announcement names Australia, Canada, India, New Zealand and the United States. Türkiye is absent from that list. For agencies managing international retail accounts, the immediate task is to identify which clients can use the report and decide what business decision it should inform. Google's announcement is the release record.

Our recommendation is to use the report to select a small product-data improvement, then evaluate that work against a defined starting point. Keep the sales forecast separate. A team can make useful progress by correcting one poorly described product family without promising a measurable revenue gain from a newly available report.

Agree what the number means before presenting it

The report documentation limits coverage to English queries and organic AI results. Share of voice depends on the competitor set. A zero can mean insufficient impressions, while 100% can reflect missing competitors. Data arrives with a delay of a few days. These qualifications belong next to the chart.

For a client review, write a short description above the screenshot: the account, country, product category, reporting period and comparison being made. Add the date you retrieved it. Agree on the same scope for the next review. If a different colleague cannot reproduce the view, postpone the performance claim and fix the report definition first.

Suppose a hypothetical footwear retailer sees a rising visibility percentage after improving its product descriptions. Check the comparison conditions before crediting the copy. Note any price promotion, stock change, assortment change or altered competitor context. Present the movement as an observation. Reserve a causal conclusion for a test designed to support one.

Keep an incomplete period out of the headline comparison. Choose a review date that gives the data time to settle, and label anything provisional. Do not replace missing observations with zeros in the spreadsheet. Put a short explanation in the cell so the person preparing next month's report knows why the comparison was withheld.

Pick a product question you can actually answer

Start with one commercially relevant category and a small set of products the merchandising team understands well. Ask the team to identify information a buyer needs before choosing: dimensions, compatibility, materials, care requirements or what comes in the box. Turn one of those gaps into the first editing assignment. Use evidence from the product owner, rather than asking a writing tool to guess.

For example, consider an invented retailer selling storage containers. A useful assignment would be to establish which sizes fit a particular shelf depth and whether lids are included. Have someone check the specifications and the actual product page. Avoid broad claims such as “fits every kitchen” unless the business can substantiate them. The practical deliverable is a correct answer attached to the correct product.

Google describes conversational attributes as optional additions to existing product data. It recommends a supplemental data source and says information already supplied in descriptions, highlights or product details need not be duplicated. Plan the edit around a real omission. Adding the same sentence to several fields gives the reviewer more work without answering another buyer question.

Keep a compact change log with product IDs, the original information, the approved revision and the person responsible. Treat that log as the handover between merchandising and marketing. For a catalogue managed by an external supplier, request a sample export before the full update so both sides can inspect the same records.

Proposed workflow for a limited product-data pilot. The diagram describes a process, not a promised performance result.

Make the test small enough to interpret

Choose the product group before looking for a favourable result. Write down why it was selected and what you plan to change. Where practical, keep a comparable group unchanged for reference. Describe that comparison honestly: a reference group selected by the team does not, by itself, establish causality. If the budget decision needs stronger evidence, involve the analyst before launch.

Give the pilot two separate acceptance criteria. First, the revised information must be accurate and appear in the intended product records. Second, the next reporting review must use the same scope as the starting point. Neither criterion requires visibility to increase. A cleanly executed test with no clear movement still tells you whether the proposed work deserves more time.

Set a spending limit for the work itself. Include merchandising review, agency editing and developer time if an integration change is needed. Agree what evidence would justify expanding to the next category. Avoid a catalogue-wide rewrite whose success is defined only after the hours have been billed. A short pilot brief should name the owner, the products, the revision, the review date and the next decision.

Use the separate Competitors report for its documented product-category comparisons. Do not paste its metrics beside AI performance numbers under one unlabeled “visibility” heading. In the client deck, give each measure its own definition and purpose. Keep paid-media budget recommendations tied to the evidence used for those recommendations.

Use the preparation work in Türkiye without inventing access

For a Türkiye-only account, record that the new report's announced availability does not cover the market. Proceed with a product-information review if it has a clear commercial purpose. Do not change account details merely to chase an interface. For an exporter, ask the account owner to verify access and eligibility for the actual target market before allocating reporting work.

The question_and_answer specification permits that optional attribute in all countries. This is a separate scope from report availability. Use the distinction in the project brief: supplying useful product information and receiving this particular AI report are separate tasks. A supported attribute does not establish access to every reporting feature.

Check the basics alongside any new field. Google's free-listing guidance says eligibility does not guarantee that products will appear. Its requirements include product information and shipping arrangements. Assign obvious data errors to the relevant owner before spending the editing budget on speculative changes. Keep the original export so the team can inspect or reverse its own edits.

At the review, show the approved changes, the observation window and the result, including an inconclusive result. If the pilot exposes a source-data problem, make correcting that problem the next task. If it produces a useful and repeatable editing process, extend it carefully. The aim is to make the next decision easier to defend.

Use AI visibility to choose a product-data test, then judge the test on its own evidence.

What to do this week

Confirm report access for each relevant account and record the target market, language and category.
Save one reproducible starting view, with the reporting period, retrieval date and metric definition.
Select one product family and have its owner verify a specific missing piece of information.
Approve a limited update with a change log, time budget and clear responsibility for checking the feed.
Book the review, define the comparison in advance and agree what would justify further work.

Sources

For help connecting this work to an SEO and GEO programme, start with the distinction between visibility and visits in our earlier article on Search Console's AI report.