Merchant Center and Product Feeds for AI Shopping

For AI shopping, your product feed is the shop window. Google says it uses Merchant Center data "to match your products to the right queries" and as "a foundational input" for "our AI powered formats and experiences". Get the required attributes right, make price and availability match your page and structured data exactly, then add descriptive attributes AI answers can draw on. ChatGPT takes a separate feed.

Method: we followed Google's Merchant Center product data specification, Google's January 2026 retail announcement and OpenAI's commerce docs, read on 31 October 2026 (CET). Quotes are verbatim. Opinions are marked "Our view"; examples are marked "Illustrative".

On this page

• Why do feeds matter more in AI shopping?
• Step 1: Get the required attributes right
• Step 2: Make feed, page and structured data agree
• Step 3: Add descriptive attributes AI answers can use
• Step 4: Set up the ChatGPT feed
• Step 5: Monitor and fix
• From MentionX data: how often AI answers name brands for shopping-style questions
• Worked example: Ledgerline Scan's feed clean-up
• How do you write titles and descriptions AI can use?
• Which products should you fix first?
• How do variants affect AI answers?
• How do you test what AI answers show?
• What do people get wrong with product feeds?
• Common questions
• Feed audit checklist
• Sources

Checked on 31 October 2026. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.

Why do feeds matter more in AI shopping?

Google's product data specification says: "Google uses this data to match your products to the right queries, and as a foundational input to help optimize and enhance the content of the ads being served in our AI powered formats and experiences" (Google Merchant Center).

In January 2026 Google also announced "dozens of new data attributes in Merchant Center designed for easy discovery in the conversational commerce era, on surfaces like AI Mode, Gemini and Business Agent". These "go beyond traditional keywords to include things like answers to common product questions, compatible accessories or substitutes" (Google). Google said it would roll them out "with a small group of retailers soon".

Our view: the direction is clear. Conversational shopping needs facts, not keywords, and the feed is where you supply them.

Step 1: Get the required attributes right

Google labels attributes Required, It depends or Optional. For Required: "Submit this attribute. If you don't, your product won't be able to serve in ads and free listings" (Google Merchant Center).

All from (Google Merchant Center).

Fictional example: a title like "Trail Runner 2 Women's Waterproof Running Shoe, Size 8, Navy" tells an AI system the product type, audience, key feature, size and colour. "TR2-W-NVY" tells it nothing.

Step 2: Make feed, page and structured data agree

Google lists "conflicting data between your data source and website" among common problems that "can prevent your ads and free listings from showing on Google" (Google Merchant Center).

For availability, Google asks you to "match the availability from your landing page, checkout pages, and structured data". For price, "Ensure that the product can be purchased online for the submitted price" (Google Merchant Center).

Do this:

1. Add merchant listing structured data to product pages (Google). See also Google's product structured data guide.
2. Generate price and availability for the feed, the page and the JSON-LD from one source system.
3. Spot-check ten products a week in each of the three places.

Step 3: Add descriptive attributes AI answers can use

Two optional attributes carry the kind of detail AI answers quote:

• `product_highlight`: "The most relevant highlights of your products." Google says "Use between 2 and 100 product highlights. Describe only the product itself. Don't list keywords or search terms" (Google Merchant Center).
• `product_detail`: "Technical specifications or additional details of your product." You can submit "up to 100 product details" (Google Merchant Center).

Also fill variant attributes. Google lists "missing or incorrect variant attributes (such as item group id [item_group_id], color [color] or size [size])" as common problems (Google Merchant Center).

Fictional example: for the running shoe, highlights might be "Waterproof membrane rated for wet trails", "8 mm heel-to-toe drop" and "Weighs 260 g in size 8". Each is a specific, checkable fact an AI answer could quote.

What about AI-written titles and descriptions?

Google's spec has a structured_title attribute: "For titles created using generative AI, use the structured title [structured_title] attribute, otherwise use the title [title]". Its digital_source_type sub-attribute value trained_algorithmic_media marks content "created using" generative AI (Google Merchant Center). The same applies to structured_description.

Our view: if you generate copy with AI, use these fields. It's honest, and it's what Google asks for.

Step 4: Set up the ChatGPT feed

OpenAI runs a separate feed. "Onboarding product feeds in ChatGPT is currently available to approved partners" (OpenAI). The integration path is:

1. "Review the file upload guide and product file schema."
2. "Validate the required fields for every record."
3. "Upload your catalog over SFTP."
4. "Keep the feed current by overwriting the same file or shard set with your latest snapshot. We recommend uploading a full snapshot at least once a day."

Source: (OpenAI, Get started). OpenAI says required fields "ensure correct display of price and availability", while recommended attributes "improve ranking, relevance, and user trust" (OpenAI, Key concepts).

Check OpenAI's prohibited list first. It includes "age-restricted products (for example, alcohol, nicotine, gambling)" and "prescription-only medications" (OpenAI).

If you're on Shopify, it says products reach AI channels "through Shopify Catalog or through the Google & YouTube sales channel or Google Merchant Center for Google AI Mode and Gemini" (Shopify).

Step 5: Monitor and fix

1. Check Merchant Center for disapprovals and "limited eligibility" weekly. Google warns that "Incorrect, inaccurate, or missing product information can cause disapprovals, limited eligibility, incorrect displays" (Google Merchant Center).
2. Ask AI Mode, Gemini and ChatGPT shopping-style prompts for your category and record whether your products appear, and with what price.
3. When an answer shows a wrong price or stock status, trace it to the feed, the page or the schema, and fix the source.

From MentionX data: how often AI answers name brands for shopping-style questions

Method. Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

Product feeds matter most when an answer recommends or compares products. In our sample, the kind of question changed how often brands were named:

Comparison and brand questions name brands most often, and those are close to how people shop: "which is better", "best X for Y". Engines also differ in how often they show sources:

Our view: our sample covers general answers rather than dedicated shopping surfaces, so treat it as context on brand naming, not as a measure of product listings. Measure your own products with shopping-style prompts.

Limits. A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

Worked example: Ledgerline Scan's feed clean-up

Ledgerline is a fictional accounts-payable tool. Figures are illustrative.

Ledgerline mostly sells software, but it also sells Ledgerline Scan, a receipt-scanning hardware bundle, through its own store. Its Merchant Center feed was set up quickly at launch: the title is "LL-SCAN-01", the description repeats the title and there are no highlights.

The team rewrites the title as "Ledgerline Scan Receipt Scanner Bundle with 12-Month Software Plan, Black", writes a factual description and adds highlights such as the scan speed, supported paper sizes and what the bundle includes. It submits the GTIN and adds merchant listing structured data to the product page.

Then it finds a mismatch: the feed shows the bundle in stock while the page says "ships in two weeks". The team switches the feed, page and JSON-LD to read price and availability from the same inventory system. A shopping-style prompt run a few weeks later shows the bundle with the correct price.

How do you write titles and descriptions AI can use?

Titles should say what the product is, who it's for and which variant it is, in plain words. Google limits titles to 150 characters of plain text (Google Merchant Center). Put the most important words first, because shorter displays may cut the end. Avoid promotional text such as "best price" or "free shipping"; it isn't a fact about the product.

Descriptions should add the details that don't fit in the title: materials, dimensions, compatibility, what's in the box and care instructions. Google's own example describes the material, fit, logo placement and washing instructions in two sentences (Google Merchant Center). Our view: write as if answering a shopper's question in a shop, not as if writing an advert. Those sentences are the kind an AI answer can repeat accurately.

Which products should you fix first?

Start with the products that matter most to revenue and that people compare. For most catalogues that's a short list of hero products and best-sellers. Fix their required attributes, titles, highlights and variants, then check them in shopping-style prompts. Once the process works, apply it to the rest of the catalogue through templates and rules rather than product by product.

How do variants affect AI answers?

Shoppers ask about specific sizes, colours and configurations. If variants share one product with no item_group_id, color or size, an answer can't say whether the version someone wants exists. Google lists missing or incorrect variant attributes among common problems (Google Merchant Center). Give each variant its own ID, link them with item_group_id and fill the attributes that distinguish them.

How do you test what AI answers show?

Write a small set of shopping prompts for your category: "best waterproof trail shoe under 150", "compare X and Y", "is X in stock in size 8". Run them on the AI shopping surfaces you care about and record whether your products appear, the price and availability shown and any wrong details. Trace each error to its source: the feed, the page or the structured data. Re-run the same prompts after fixes. Our view: answers vary between runs, so run each prompt more than once before drawing conclusions.

What do people get wrong with product feeds?

• Treating the feed as a one-off setup. Prices, stock and products change; the feed must change with them.
• Keyword-stuffing titles and highlights. Google says highlights should describe only the product and not list keywords or search terms (Google Merchant Center).
• Letting feed and page disagree. Conflicting data can prevent listings from showing (Google Merchant Center).
• Hiding AI-generated copy. Google asks you to use structured_title and structured_description for AI-generated text (Google Merchant Center).
• Assuming one feed covers every assistant. ChatGPT has its own feed and onboarding for approved partners (OpenAI).

Common questions

Is GTIN required? Google lists it as 'It depends', strongly recommended if available.

How often should the ChatGPT feed update? OpenAI recommends uploading a full snapshot at least once a day.

Can I use AI to write product titles? Yes, but Google asks you to put AI-generated titles in structured_title.

Feed audit checklist

• Every product has all Required attributes
• GTINs submitted where they exist
• Titles say what the product is, for whom and in which variant
• 2+ product highlights per hero product
• Price and availability match across feed, page and JSON-LD
• Images are at least 500 x 500 px before 31 Jan 2027
• AI-written copy uses structured_title or structured_description
• ChatGPT feed applied for or live, refreshed daily

Next: agentic checkout protocols.

Sources

All accessed 31 Oct 2026.

• Google, Product data specification (Merchant Center Help)
• Google, Merchant listing structured data; Product structured data
• Google, New tech and tools for retailers (11 Jan 2026)
• OpenAI, Get started; Key concepts
• Shopify, Agentic storefronts