Product Feed Optimization for AI Mode Shopping: What to Fix First

Illustration: four stacked bars, gold at the base, two rose and one cream above, with a dashed arrow rising beside them.

A question I hear from e-commerce clients: is our feed “ready for AI Mode”? I understand why. Google’s AI shopping features are loud right now, and a feed is the one thing in an e-commerce account that you fully control.

My answer: nobody outside Google can tell you what decides visibility in AI Mode. What I can tell you is which parts of the feed Google documents, which ones it ties to AI surfaces by name, and the order I fix them in. It’s mostly the same order I use for a plain Shopping campaign, which is a hint in itself.

What Google documents, and what it doesn’t

Documented, as of November 2026:

  • AI Mode in Search “brings together Gemini capabilities with our Shopping Graph”, and Google says the Shopping Graph holds more than 50 billion product listings (Google, May 2025). That post says nothing about Merchant Center or how a merchant’s data gets in.
  • The product highlight page says submitting highlights “can help customers discover information about your products across AI-driven surfaces, like AI Mode in Google Search”. Google recommends 4 to 6 per product, with a minimum of 2 once you use the attribute.
  • Google has a group of “conversational attributes” built for AI systems: question and answer, document link, related product, item group title, variant option and popularity rank. They are optional, they don’t affect the approval of existing products, and Google recommends adding them through a supplemental data source (How to use conversational attributes). The question and answer page says that attribute is primarily intended for conversational experiences such as AI Mode.
  • Merchant Center has an AI performance insights report. It shows share of voice on AI Mode and AI Overviews and an attribute completeness view. It currently covers English-language queries for accounts in Australia, Canada, India, New Zealand and the United States.

Not documented: a ranking formula for AI Mode, a minimum feed score to appear, or any statement that a complete feed guarantees placement. None of Google’s pages I checked describe how AI Mode shopping behaves for the Dutch, German, French or Polish markets. Anything specific you read about that is somebody’s guess.

So the title and identifier advice below comes from the product data specification. My opinion, clearly labeled: a feed that describes products precisely is a safer bet than one that doesn’t, whatever surface it shows on. It costs little, and it helps classic Shopping and Performance Max today.

The audit order

I work from the bottom of the data model up, since each layer depends on the one before it. The table gives the order.

OrderLayerAttributesWhy this position
1Identifiersgtin, brand, mpn, identifier_existsDefines what the product is
2TitlestitleThe most prominent text in an ad or listing
3Classification and factsgoogle_product_category, product_type, color, size, material, product_highlight, product_detailDescribes it in structured form
4Custom labelscustom_label_0 to custom_label_4Only for your bidding, never shown to shoppers

1. Identifiers first

Google calls GTIN, MPN and brand the common unique product identifiers, and says that if a product has an assigned GTIN but you submit it without one, it “may have limited visibility” (About unique product identifiers). GTIN is the number under the barcode: 13 digits for EAN in Europe (Find a GTIN).

I check four rules, all from Google’s pages.

  • Don’t guess. Google says don’t make up identifiers, don’t copy them from similar products, and don’t put your internal SKU in the GTIN field. Wrong identifiers can get a product disapproved.
  • MPN comes from the manufacturer. Unless you are the manufacturer, don’t invent one. The MPN page says to leave it out if you’re unsure.
  • Brand is required for most new products, and when you make an unbranded product yourself, the store name works as the brand, paired with your own MPN (Brand).
  • identifier_exists is for genuinely custom or vintage goods. Setting it to no on a product that does have identifiers triggers a warning (Identifier exists).

You may have read advice to “make up an MPN if you have to”. Google says the opposite. For resellers, the real work is a supplier file mapping barcodes to your product IDs.

2. Titles

The title page gives you 150 characters and says shoppers usually notice only the first 70 or fewer, depending on screen size. Put the defining details first. Product name, brand when it differs between products, specifics like “waterproof” or “maternity”.

Things Google says not to do: promotional text, price, shipping, your company name, all caps for emphasis, and extra spaces. Each variant needs its distinguishing detail in the title, and the same detail belongs in the matching attribute (color, size).

Two details matter here.

  • Google says it may promote key product attributes using content from your landing page, images and other signals in your ads. To opt out of Product Data Customization for Shopping ad titles, the page says to contact your account manager or support team.
  • For titles created with generative AI, Google’s title page says to use structured_title, with the digital_source_type sub-attribute set to trained_algorithmic_media. The page requires that for all AI-generated titles, so it matters if you bulk-rewrite titles with an LLM.

My own rule is to write titles from your product facts and your search terms report, never from a keyword tool alone. Don’t add a feature the product doesn’t have because shoppers search for it.

3. Classification and attributes

google_product_category. Google assigns one automatically and only accepts your override in a few cases, such as category-specific requirements or campaign targeting (Google product category). Check it for products that landed in the wrong category, which also decides which attributes Google demands.

product_type. This is your own taxonomy, up to 750 characters with > between levels. Only the first value is used for bidding and reporting in Shopping campaigns (Product type). Mirror your store navigation here.

Color, size, material. The specification requires color for apparel in several markets, including France and Germany, and for any product sold in different colors. Size is required for clothing and shoes in the same markets. Material is required when it distinguishes variants. For everything else they’re optional, and filling them costs only time.

Highlights and details. Highlights are the one attribute Google ties to AI Mode in its own words, so I add them to the best-selling products first. Short fragments that answer what a buyer would ask: materials, compatibility, what’s in the box. product_detail holds the technical specifications.

Question and answer. Write them for the products where buyers ask the same thing again and again. Google says to give correct, informative answers, to avoid keyword lists and prices, and not to duplicate what’s already in your title, description, highlights or details.

4. Custom labels last

Custom labels are never shown to customers. They exist so you can group products for bidding and reporting in Performance Max, Shopping and Demand Gen, with up to 1,000 unique values per label (Custom label 0–4). They only affect bidding and reporting, so they come after the data that defines the product. If you’re deciding what to put in them, start with margin, as in breakeven ROAS.

Making the changes without editing the store

Your store platform shouldn’t have to change for this. Two Merchant Center tools do the job. Google’s pages say the supplemental sources tab is visible only with the “Advanced data source management” add-on, and the attribute rules page tells you to enable that same add-on, so check Settings > Add-ons first.

Supplemental data sources. A supplemental source adds or updates attributes for products that exist in your primary source. It can’t add or remove products. You match rows by product ID, so the ID has to match exactly, and the feed label has to match the primary source’s (Create a product data source). A Google Sheet with an ID column and a column per attribute is enough for highlights, categories, materials and rewritten titles.

Attribute rules. Rules transform values inside Merchant Center. Google’s own example is setting the title to the brand plus the title. There’s also an “Extract” operation that pulls values out of existing text, and a test-and-preview step before you apply anything (Set up your attribute rules). Rules run in order, so test them on a small set first.

One warning from experience. If a feed app or platform integration sits between your store and Merchant Center, find out which source wins before you rewrite anything. The DataFeedWatch post covers where a feed tool fits in.

Leave id, price and availability alone here. Fix them at the source.

Selling into NL, DE, FR and PL

Four markets means four languages, and Google’s rules on language are strict.

  • One data source can target several countries and the shopping_ads_excluded_country attribute lets you opt individual products out of Shopping ads in specific countries. A German feed can also reach German speakers in Austria, if shipping and Google Ads location targeting are set up for it (Show products in multiple target countries).
  • A new language needs a new data source with translated data. Language is chosen when the source is created and can’t be changed later (Use feed labels).
  • Landing pages and checkout must match the feed language. If most of your data doesn’t match, products can be disapproved (Incorrect language).
  • Feed labels let you target several sources in one campaign, for example one label per language. A label can be up to 20 characters, with no spaces and only uppercase letters, digits, hyphens or _.
  • Google Ads location targeting has to match the target countries in Merchant Center, or the products may not show.

In practice, I translate titles and highlights properly rather than machine-translating the feed wholesale. Polish shoppers don’t search the way a German title reads when translated word for word. Identifiers are the same everywhere, which is a good reason to fix them once.

One caution about measurement. AI performance insights is listed for English-language queries in a handful of non-EU countries. If you sell in the Netherlands, Germany, France or Poland, you can’t use that report today, and I don’t promise anyone an “AI Mode impressions” number for those markets.

How I run it

  1. Export the feed and count blanks. GTIN, brand, MPN, color, size, material, highlights. Rank products by revenue so the top 50 get attention first.
  2. Fix identifiers from supplier data. Leave blank rather than guess.
  3. Rewrite titles for the top sellers using real search terms, then roll out to the rest by pattern with an attribute rule.
  4. Add highlights, material, color and size through a supplemental source.
  5. Add question and answer for a few products where buyers ask the same things.
  6. Set one custom label that serves a bidding decision you actually make.
  7. Check the Needs attention tab after every upload and give changes a few weeks before you judge them.

What I skip: chasing a “feed score”, rewriting 10,000 titles before fixing identifiers, and optimizing for a surface Google hasn’t documented for your country. A bad feed hurts in a way no title fixes, because a suspended Merchant Center account turns off every campaign at once. For how campaign type interacts with feed quality, see Performance Max vs Standard Shopping.

Conclusion

Fix what the product is (identifiers), then what it’s called (titles), then how it’s described (attributes), then how you bid on it (custom labels). Do it through supplemental sources and rules, and keep your store untouched. Google ties highlights and conversational attributes to AI Mode by name, and everything else here is good feed hygiene that pays off on every Shopping surface. If you want a second pair of eyes on a feed, that’s part of my Google Ads for e-commerce work. Schedule a call and bring a feed export.

With Texonica, you get less hype more results.

We are determined to make your business successful. Reserve your free-no-obligation strategy session today.

Schedule a Call