For two decades the product feed sat on the paid side of the wall. Shopping ran on it, the PPC team maintained it, and SEO’s involvement ended at a glance in Search Console. That split made sense when the feed only fed ads. It does not make sense now.

An ecommerce product feed is the data layer behind paid Shopping, free listings, organic rich results, the Shopping Graph, and the AI surfaces built on top of them. Google names it directly in its guidance on product visibility in generative search. Yet plenty of teams are debating whether to publish a markdown copy of their site for language models while the file that decides whether their products are eligible at all quietly rots.

Three copies of your catalog, and none of them agree

Google does not read your products once. It reads three parallel representations, each with its own rules and, in most companies, its own owner.

• The feeds. Google Merchant Center (titles, GTINs, prices, availability) and, for richer specification data, the Manufacturer Center. A complete data structure living outside your website.
• On-page structured data. Usually JSON-LD, used partly to verify the feed and partly to power ecommerce rich results. Schema.org is not the only vocabulary in play; Google has pointed at GS1, UN/CEFACT and other ontologies.
• The website. The rendered page a shopper sees, and the machine-readable version an agent parses and checks against the other two.

Google has floated unifying schema markup and feed submission to reduce the duplication. Until that lands, you maintain three sources for one truth, and when they disagree Google decides which to believe. That call rarely favours the merchant.

What breaks when nobody sees all three layers

The usual failure is not incompetence, it is scope. A paid manager seeing a disapproval spike investigates what sits in their remit: feed attributes, the live page, bid strategy, policy flags. They fix what they can reach and escalate the rest to development. Nobody calls SEO, because structurally there is no reason to.

The product with four different prices

Price is not optional, and Google has been tightening enforcement on mismatches. In one case, an office furniture retailer’s catalog started being disapproved en masse. The website said £34.80. The primary feed said £34.80 GBP. Merchant Center had decided the price was £33.54. The JSON-LD reported a fourth figure, £29, because the template was outputting the ex-VAT price alongside a priceValidUntil field.

Google uses structured data to verify feed data and, through automatic item updates, can overwrite feed prices with it. That is why the wrong number reached Merchant Center. The fix took minutes; finding it required someone able to open the feed and the markup in the same session.

Availability and variants speak different languages

Some mismatches are harder than a wrong number, because the systems do not share a vocabulary. Merchant Center accepts four availability values, in_stock, out_of_stock, preorder and backorder, with the last two requiring an availability_date. Schema.org expresses the same idea through the availability property on an Offer, with its own value set. Maintained separately, they drift.

Variants are worse. In a feed they are a flat list stitched together by a shared item_group_id; in structured data they are a nested parent-child relationship built with ProductGroup, hasVariant and variesBy. When one team owns the flat file and another owns the nested markup, variant mapping fails in ways neither can see alone. Sites already fighting near-duplicate listings feel it first; our notes on fixing thin content across similar ecommerce product pages cover the on-site half of that problem.

When everything is correct and products still vanish

Data conflicts are at least auditable. Infrastructure failures are not. In one case a retailer’s products flipped from approved to disapproved overnight with a clean feed, clean schema and a healthy site. A change to the CDN’s bot protection rules had begun treating Googlebot as a threat. With the website layer unreachable, Google could not verify it against the data it already held, and the listings were pulled.

Diagnosing that needs someone reasoning about crawl behaviour, feed health and infrastructure at once, which is the shape of an SEO skill set. If crawl access is a live concern, it helps to understand how Google treats pages it cannot reliably crawl or index.

Why the feed is now a search asset

Exported feeds are written for databases, not shoppers

Left alone, a feed is whatever the platform’s export plugin produces: generic titles, approximate categories, thin attributes. It passes validation and stays commercially invisible. Paid teams patch around it with feed rules and supplemental feeds, neither of which was designed to compensate for a primary feed built without search intent in mind. Treating it as a search asset means front-loading titles with the terms people actually use, tightening taxonomy so products are not buried under catch-all categories, and matching attribute depth to how customers filter.

Feed health is a trust signal

Merchant Center account health, including disapproval rates, missing attribute warnings and policy compliance, feeds how Google assesses you as a data source. The Shop Quality program formalises this, benchmarking merchants on approval rates, shipping data completeness and return policy clarity, then surfacing the outcome as a Top Quality Store badge across paid and organic placements. The Shopping Graph raises the stakes again: with more than 50 billion product listings feeding AI Overviews, AI Mode and Gemini, how confidently Google can verify your data sets your standing inside it. The direction is visible in Google’s wider product feed strategy.

The results page is turning into a product page

Brodie Clark has described how Google’s results increasingly behave like a product detail page, with visual grids taking prominent space and cannibalising branded queries, particularly for brands stocked by large third-party retailers. On mobile it can take several scrolls before a brand’s own category pages appear. That makes the feed the source behind a growing share of commercial SERP real estate, which is why merchant listing structured data deserves more than a plugin default.

Agents have to read your page, then buy from it

AI surfaces draw on Merchant Center data to answer commercial queries, so thin attributes are a handicap before discovery even happens. Google’s UCP guidance tells merchants to use their existing Merchant Center feeds to reach high-intent customers, with UCP unlocking surfaces including AI Mode and Gemini, and Google is adding conversational commerce attributes such as compatibility, substitutes and related products specifically to inform AI responses.

Visibility is only half of it. To complete a purchase, an agent works from a machine-readable version of your site: raw HTML, the accessibility tree, rendered screenshots. The accessibility tree matters most, since it distils the page into the roles, names and states of interactive elements. Non-semantic markup, layout instability and controls buried behind overlays degrade that map, and the transaction fails before it starts. The checklist sits in our guide to agentic commerce optimization and Google’s UCP. A third feed matters here too: comparing several offers for one item requires authoritative specification data beyond price and stock, which is what the Manufacturer Center holds. Gianluca Fiorelli calls this the product truth layer, and the label fits.

What shared ownership actually looks like

1. Route alerts to both teams automatically. Monitoring that fires on disapproval spikes should notify SEO and paid at the same moment, not whoever opened Merchant Center that morning. The cross-team conversation should start when the data moves, not after someone notices.
2. Put feed health in the recurring review. Brooke Osmundson has argued for reviewing feed health alongside spend, ROAS and CPA, which is the right weekly cadence. Layer a monthly audit on top: are availability values consistent across all three layers, do prices match, are required attributes present in both feed and schema for priority product types.
3. Declare a source of truth per attribute. Product data lives in the CMS, the ERP, the feed platform and the schema template, and usually nobody has decided which wins. For most teams the feed should. Then generate or validate schema against it rather than letting a plugin emit its own version.
4. Bring SEO into feed architecture decisions. Not to veto tooling, but so attribute choices, site implementation and markup are decided with a shared understanding of how Google reconciles them. Custom labels are a good test: five slots exist, paid usually claims all of them for bidding, and some could carry search priority instead. Retrofitting that later is painful.

That last point has a prerequisite: SEOs cannot hold a seat at that table without a working knowledge of Merchant Center attributes and where they do and do not map to structured data vocabularies. Google’s feed documentation is detailed, readable and cross-references schema wherever an equivalent exists. That is the baseline, not an advanced topic.

Frequently asked questions

Should SEO take the feed away from the PPC team?

No. Shared ownership is additive. Paid keeps bidding, campaign structure and commercial strategy; SEO brings search intent, taxonomy and cross-layer diagnostics. The benefit for paid is fewer disapprovals to firefight and cleaner attribute data to optimise against.

Which layer wins when feed, schema and website disagree?

It varies by attribute, and that unpredictability is the point. Google uses structured data to verify feed data and can overwrite feed prices through automatic item updates, so a wrong figure in markup can propagate into Merchant Center. Treat disagreement as risk, not a hierarchy you can rely on.

Does feed quality affect organic results, or only ads?

Both. Merchant Center powers free listings as well as paid, and the same structured data drives organic rich results such as price, availability and review counts. Rich result errors are the dangerous ones, because no disapproval mechanism flags them; incorrect information simply surfaces.

Where should a neglected feed programme start?

Audit price and availability consistency across all three layers first, since those cause the most disapprovals. Then confirm the site is genuinely crawlable, including CDN and bot protection rules. Title and taxonomy work comes after the eligibility problems are fixed.

The takeaway

The question was never whether an ecommerce product feed counts as an SEO asset. It plainly does, across organic, paid, free listings and the agentic layer forming above all three. The open question is whether SEO teams will claim a share of the responsibility for it.