Google AI Mode: How to Use Merchant Center AI Insights

Check whether the report covers your store

On September 16, 2026, Google announced that Merchant Center AI performance insights were available to businesses in Australia, Canada, India, New Zealand, and the U.S. Google described visibility across AI Mode and AI Overviews.

The report currently covers English-language queries for eligible Merchant Center accounts. Those limits matter: don’t treat it as a worldwide picture of every shopping conversation.

Start by checking whether AI performance insights is available in your own Merchant Center account. Use it to guide an information review, then choose individual listings to inspect from your own catalog. Don’t build your workflow around an assumed itemized list of products appearing in AI answers.

If you don’t have access, you can still do the manual listing-to-page checks below. There’s no reason to postpone correcting a missing measurement or an inaccurate description while waiting for a report.

Read share of voice without mistaking it for sales

Treat the visibility percentage as context, not a sales result.

Google calculates share of voice using your AI impressions relative to impressions for your store and its Merchant Center-defined competitors. It isn’t a universal ranking against every retailer on Google.

The report also covers organic AI traffic, such as free listings, and excludes paid ads. So it isn’t an all-in view of your advertising and unpaid visibility either.

My recommendation: use the percentage as background for that particular comparison group. Don’t turn it into a revenue target or judge your marketing solely by whether it moves.

Keep your own traffic, order, and revenue reporting alongside this review. If visibility changes after an edit, investigate what happened next rather than assuming the edit produced more sales. A prettier percentage doesn’t pay for holiday inventory.

Build a manual checklist for missing product details

When Google previewed the report on May 27, 2026, it said product-attribute insights could help retailers identify products missing specifications shoppers seek. That’s the purpose I’d prioritize over score watching.

Choose a product from your catalog that you plan to promote, or one that prompts repeated customer questions. Compare its Merchant Center title, description, and attributes with its live product page. Your feed—the product information you send Google—and your website should both give a shopper accurate answers.

Picture a homewares store reviewing a storage basket. Suppose its listing describes the room it would look nice in but leaves out its material and dimensions. The useful correction isn’t more variations of “stylish storage.” It’s the verified information someone needs to decide whether the basket belongs on their shelf.

Use a short worksheet:

Treat these as inspection prompts, not instructions to add every possible attribute. Include only details that apply to the item and that you can verify.

Customer emails and support questions are useful places to start your own question list. Ask whoever answers them which missing detail they’re tired of explaining.

Fix listing and product-page mismatches first

Google’s guidance is straightforward: use the most current and complete product information you have. It also requires key landing-page details, including price and availability, to match submitted product data.

Before experimenting with extra AI-specific details, open a submitted listing and follow its link as a customer would. My first pass would cover:

• •Price: Is the live price the one submitted for this product?
• •Stock status: Does the page agree with the listing about availability?
• •Image: Does it represent the item being offered?
• •Selected variant: Are you inspecting the same size, color, or configuration?

Correct whichever side is wrong. Don’t change an accurate product page merely to make it agree with stale submitted data.

Then review any specification you plan to add. If you update a material or compatibility claim in Merchant Center, check that the linked page tells the same story. Don’t leave the customer to resolve the disagreement.

For search work on the product pages of an existing store, Quicklook’s website SEO service is the relevant option—not automatically a website rebuild.

Consider optional attributes after the core cleanup

Google says conversational product attributes are completely optional and complement primary Merchant Center product data. They aren’t a replacement for the ordinary catalog information you just checked.

Consider these additional details when a useful buying question remains unanswered after that cleanup. Don’t make filling optional fields the project before you’ve established what the shopper actually needs.

For each proposed addition, ask:

• •Does this answer a real question relevant to this product?
• •Can someone who knows the product verify it?
• •Does the answer apply to this exact item, rather than a similar model?

If the answer is uncertain, pause. Don’t turn “probably compatible” into a confident product claim because a field is available.

There is a useful real example, with a narrow meaning. Google reported that, during testing with lululemon, brand-submitted conversational attributes were incorporated into 50% of relevant AI Mode product recommendations.

That describes use of the submitted details in that test. It does not mean recommendations increased by 50%, and it isn’t a promised sales gain for your store. Take it as a reason to consider accurate, useful additions—not as a forecast to put in your holiday budget.

Keep a change log instead of rewriting the whole catalog

Make the output of your review a correction list someone can finish.

For each item, record the product, the suspected gap, the source used to verify the answer, and what changed. Note whether the correction affected submitted data, the product page, or both. Save the original wording so you can understand the edit later.

Separate confirmed errors from unanswered questions. An incorrect stock status belongs in the correction queue. An uncertain compatibility claim belongs with someone who can verify it—not with whoever can write the most confident sentence.

At the next catalog review, check that the corrections are still live. Look for another product-information gap worth checking. Don’t assume the next useful task is a broader rewrite.

Before the holiday rush, I’d prioritize products you intend to promote and can actually supply. Leave cosmetic copy changes behind factual errors and missing buying details. The aim is a manageable work queue, not a new standing meeting about a percentage.