Merchant Center and Product Feeds: From Ad Plumbing to AI Source Data
Google Merchant Center product feeds, from setup basics through the AI context. Google states no special markup is needed for generative AI, yet recommends Merchant Center feeds for product information. This guide covers the three conditions that matter — freshness, granularity and attribute coverage — plus consistency with schema and how ChatGPT's feed differs. In our study, AI named official campaigns 43.3% of the time.
Key takeaways
- Product feeds have moved from ad plumbing to source data for AI answers. The same data now supplies free listings and AI shopping answers
- Google states that no special markup is needed for generative AI, yet recommends Merchant Center feeds for product information. That asymmetry is the point
- When we measured how well AI caught limited-time campaigns, official campaigns were named in 39 of 90 opportunities (43.3%). Publishing on your own site is not enough to reach them
Product feeds became AI source data
Merchant Center was understood for years as the machinery behind Shopping ads — something the paid media team touched, largely invisible to SEO and content owners.
That position has changed. With free listings added, and product information now feeding AI shopping answers, the feed has become an exposure route independent of ad spend. Retailers running no advertising at all now have a reason to maintain one.
What Google says
Two statements from the official guidance, side by side. The article turns on the gap between them.
In its guidance for AI features on Search, Google states that there is no special schema.org markup to add for generative AI search. In the same guidance, it names Merchant Center feeds and Google Business Profile as ways to improve visibility for product listings and local information.
Structured data unnecessary, feeds recommended. What does the asymmetry mean? AI wants settled product data, not information it has to infer from a page. HTML allows enormous freedom of expression — whether a price includes tax, what "in stock" refers to, is up to whoever wrote the page. A feed has a fixed format, leaving no room for interpretation.
That does not make structured data pointless; the purpose is simply different. Our structured data guide covers that distinction, and our overview of LLMO gives the full map.
Merchant Center basics
Before the AI part, the setup.
Start with opening an account and verifying site ownership. If the site is already verified in Search Console, the same Google account carries over.
Then register the feed. Four methods: direct file upload, scheduled fetch from a URL, automated integration through the Content API, and cart platform apps. A few dozen products can live on a file; several hundred with moving stock realistically need scheduled fetch or the API.
The required attributes are id, title, description, link, image_link, availability, price, brand, GTIN and condition. Among these, GTIN is the identifier that uniquely specifies a product and forms the basis for matching against other listings of the same item. An error or omission can leave you outside the comparison set entirely. If the product has a barcode, include it.
On free listings and ads: the same feed serves both. Even without running ads, registering a feed makes you eligible for the free slots in the Shopping tab, and that data reaches the AI side as well.
What makes a feed work for AI answers
A feed built for advertising and a feed built for AI emphasize different things. Three of them.
First, freshness. Update frequency always mattered for Shopping ads, and it matters more in AI answers. Asked "which one should I buy right now" in a conversation, stale data does not get used.
We have a measurement here. In our study putting nine official campaigns from three electronics retail brands to five major AI services, the campaign was named in 39 of 90 opportunities, or 43.3%. By brand, the strongest reached 70.0% while the weakest sat at 30.0%. Information that rotates is not reaching AI, even when it is published on the official site.
Second, granularity. Titles written for search ads are built around short keywords and rarely carry enough to answer a conversational question. Asked for "a washing machine under 60cm wide for a single-person apartment", a product without size and installation attributes does not survive the filter.
Third, attribute coverage. Size, material, compatibility, warranty length. Whether you hold, as attributes, the axes customers actually raise. Listing the questions your support team receives shows you which attributes to fill.
Feed specifications keep moving, so check the current attribute list in the Merchant Center help at the time you start.
When schema and feed disagree, both sink
This is the most common failure in practice.
The same product carries a tax-inclusive price in the feed and a tax-exclusive one in the page's JSON-LD. Or the feed says in stock while the page shows sold out. To a machine, these are states where neither source can be adjudicated. With no basis for trusting one over the other, confidence in both falls.
The fix is simple. Generate both from the same source. Emit the feed and the JSON-LD from the product master. Once nobody is hand-editing one side, this class of failure stops occurring structurally.
When the feed belongs to the ads team and the JSON-LD to engineering, a sale updates one of them. If ownership must be split, at least make the underlying data single-sourced.
ChatGPT has a feed too
This is not only a Google story. ChatGPT also accepts product feeds through its merchant program.
The two feeds differ in purpose and requirements, so here they are together.
| Aspect | Google Merchant Center | ChatGPT (OpenAI) |
|---|---|---|
| Primary purpose | Shopping ads, free listings, and supplying product data to AI answers | Product recommendation inside ChatGPT (discovery) |
| Submission method | File upload, scheduled fetch, Content API, cart integration | Submitted through the merchant program. Check the official documentation for terms |
| Required attributes | id, title, description, link, image_link, availability, price, brand, GTIN, condition | An identifier plus price and stock at the core. Check the official documentation for details |
| Update frequency | Scheduled fetch or API as needed. Once a day is not enough where price and stock move | Speed of reflecting changes is emphasized. Specifications change, so check the official documentation |
| Review process | Per-product approval, visible in the diagnostics screen | Check participation terms and whether review applies in the official documentation |
| Where purchase completes | On the merchant's site | On the merchant's site, since Instant Checkout ended in March 2026 |
The ChatGPT side moves fast. Submission format, update frequency, whether a review applies, which merchants are eligible — check OpenAI's official documentation at the time you start. Feed mechanics, including what is written here, is the fastest-aging material in this article.
One more thing worth noting: after Instant Checkout ended in March 2026, the feed remains in place as material for discovery. Purchase no longer completes inside ChatGPT, but the feed's role in supplying recommendations has not changed. The full set of ChatGPT routes is covered in getting your products recommended in ChatGPT.
Nobody is measuring the result
In our audit of nine major electronics retail brands, no published information about product feeds, API integration, or measurement of AI-referred traffic was confirmable for any of the nine.
That is not proof they are doing nothing. It is entirely possible the work exists without being published. The audit covers public information, so what is measured internally is out of view.
Something can still be said. Whether you publish a feed and whether you measure its effect are separate questions. A feed becomes an investment only once you have checked whether your products started appearing in AI answers after it went live.
How to work through it
Check disapproval and warning counts in Merchant Center diagnostics. Done when disapprovals are at zero
id, title, description, link, image_link, availability, price, brand, condition. Done when every product carries all of them
Include the barcode wherever one exists. Done when no product that should have a GTIN is missing one
Move from manual upload to scheduled fetch or API integration. Done when price and stock changes appear the same day
Put a feed row and the same product's JSON-LD side by side. Done when price, stock and GTIN all agree
Frequently asked questions
Yes. Free listings apply, and that data reaches AI shopping answers. It is an exposure route independent of ad budget.
Basic attributes sync automatically, but GTIN and detailed attributes such as size are frequently sent empty. Check Merchant Center diagnostics to see what is actually being transmitted.
For own-manufactured or handmade goods where no GTIN exists, declare that explicitly with identifier_exists. That is handled differently from simply leaving the field blank.
Where price and stock move, once a day is not enough. Schedule multiple fetches, or push changes through the API as they happen.
Eligibility is defined by the program and changes. Check OpenAI's official documentation at the time you start.
Marketplace listings are managed by the marketplace. If you also run your own store, that feed is yours to maintain. When AI answers name the marketplace but not you, work on your own site is what changes it.
Summary
The meaning of the product feed has changed. What was machinery for advertising is now the route by which settled product data reaches AI.
Google calls special markup unnecessary while recommending feeds because it wants settled data rather than inference. That is exactly why freshness, granularity and attribute coverage translate directly into results.
In our measurements, AI named official limited-time campaigns 43.3% of the time. Publishing on your own site does not get you there. Once the feed is in place, check whether your products actually started appearing. We offer a free AI visibility assessment for retail and e-commerce businesses.
