Merchant Center AI shopping: the feed hygiene checklist
Merchant Center AI shopping works when the product data Google holds for your store is complete and matches what a shopper sees on the page right now. Gemini and Google's other shopping surfaces draw on that data, so a feed full of warnings costs you before anyone reads a word of your site. Most of the fixes are dull, and most of them fit in a week.
Why the feed matters more as shoppers use assistants
Locus reported in Q2 2026 that 45% of US online shoppers use an AI tool somewhere in the purchase journey. Adobe found that AI traffic to US retailers converted 42% better than non-AI traffic in Q1 2026. Those visitors arrive late in the decision. A wrong price or a missing size loses a sale you were close to winning.
Other posts here cover the Shopify catalog and ChatGPT's feed spec. This one covers the Google side.
Merchant Center AI shopping starts in the Diagnostics tab
Open Diagnostics and sort by the number of products affected. Disapproved items do not serve at all, so they come first. Warnings come second. Account-level issues come third, because a single missing policy can touch every product you sell.
Product feed errors tend to cluster. One bad column mapping in your export can break hundreds of items at once. Fix the mapping rule in your source data and the whole group clears on the next fetch.
A catalog the size of JCS Fashions' 4,878 products cannot be audited by scrolling. That store is the kind of catalog where you group issues by pattern and fix the pattern.
The six fields that fail most often
| Field | Common fault | Fix |
|---|---|---|
| GTIN, brand, MPN | Blank, or codes copied from a similar product | Use the manufacturer's codes. For custom goods with no GTIN, set identifier_exists to no. |
| Title | Brand name only, or no size and color | Lead with the product type, then brand, color, size and material, in the order a shopper would type them. |
| Price and availability | Feed says in stock, page says sold out | Pull both from the same source as the page and refresh on a schedule. |
| Google product category | Missing or too broad | Pick the most specific category that fits. |
| Images | Placeholders, watermarks, one image for every variant | One clean image per variant on a plain background. |
| Shipping and returns | Not set, or different from the policy page | Set both at account level and copy the policy page wording. |
Policies are product data too
Shopify's catalog requires a standard category, an image per variant, a price above zero and full policies. Google's side asks for the same things in different places. If your return window is 30 days on the policy page and 14 in Merchant Center, a shopping assistant has two answers to pick from, and a shopper who gets the wrong one will remember it.
Check your business information too: store name, contact details, and the claimed and verified website URL. A mismatch between these and your site is an account-level issue that can hold back every product.
Keep the feed and the page saying the same thing
Structured data correlates only weakly with AI citations, but it matters for rich results and feed consistency. The practical use is agreement. The price in your Product schema, the price on the page and the price in your Google Merchant Center feed should all come from one source. Sale prices are a common failure point: the sale ends on the site and the feed still carries the old number for a day.
If you use a feed app, check that it sends variant-level data. A parent product with one price and one image hides the size and color choices that shoppers ask Gemini shopping and other assistants about.
A five-day plan
- Monday: export the Diagnostics issues and list the top five by number of products affected.
- Tuesday: fix identifiers and categories in your source data, then trigger a new fetch.
- Wednesday: rewrite titles for your best 50 sellers and replace placeholder or shared variant images.
- Thursday: compare 20 random products against their live pages for price, stock and variants. Fix the sync if any differ.
- Friday: set shipping, returns and business information, then recheck Diagnostics.
Before you start, write down 10 shopping prompts a buyer would type for your category, and run them in Gemini, ChatGPT and Perplexity. Run the same 10 two weeks after the fixes. Answers change from run to run, so look for products that now appear with the right price and stock, and ignore single flips.
Your category pages, guides and third-party mentions still decide who gets named. Merchant Center AI shopping is the data layer underneath them. If you want the wider picture for stores, the ecommerce page lays it out.
Next step
Fix the top Diagnostics issue today, then run the free AI visibility scan to see where your store is cited now and which products to test first.