Product feeds for shopping channels

The useful mental model is that a shopping feed is a rendering of your catalogue into one channel’s schema, the same way a storefront template is a rendering into HTML. Google wants an availability value drawn from a fixed vocabulary; Shopify has inventory quantities and a policy about overselling. Something has to translate. That translation is the feed.

Once you see it that way, the governing rule follows: a feed should transform, not author. Anything the feed invents — a title nobody can find in Shopify, a brand that exists only in the feed tool, a category assigned in a third-party dashboard — is product data living outside the system where product data is managed. It will be wrong within a year, because the person who changes the product in Shopify has no idea the other copy exists.

This is the single most common way feed projects go bad, and it does not look like a mistake at the time. It looks like moving quickly.

Google publishes a product data specification, and it is worth reading rather than summarising from an app’s onboarding flow. The required set is small: id, title, description, link, image_link, availability and price. The limits attached to them are concrete — id at a maximum of 50 characters, title at 150, description at 5000.

Identifiers are where accounts get suspended. Brand is required for new products except movies, books and musical recordings. GTIN is strongly recommended where one exists, and MPN is required only when a product does not have a manufacturer-assigned GTIN. The identifier_exists attribute is for genuinely unidentified products — a product with no brand, GTIN or MPN — and setting it to no on products that do have identifiers is not a shortcut, it is a misrepresentation. Shopify’s own requirements page is blunt about the stakes: "If these identifiers aren't correct, then Google doesn't publish your products and your account could be suspended."

Some categories are stricter. Shopify notes that products in Music, Movies or Video games are required to have a UPC, EAN or JAN, and that Books require an ISBN. Apparel and accessories can require age group, gender, colour and size depending on the product.

Images have a floor. Google specifies a minimum of 500 × 500 pixels, with enforcement beginning 31 January 2027, and prohibits promotional text, watermarks and borders. That last rule catches more merchants than the dimension does, because the "Free delivery" flash somebody added to the hero shot is exactly what the rule is about.

Price and availability must agree with the page. Availability has to be one of in_stock, out_of_stock, preorder or backorder, and it must match what the landing page says. Price is required in ISO 4217 format. Both of these sound trivial and both are where multi-currency and app-driven pricing quietly break things — a market with converted prices or a discount rendered by an app can put the page and the feed out of step without anyone touching the feed.

A meaningful share of feed problems are store-level and get diagnosed as product-level, which sends teams off editing a catalogue for a week.

Shopify’s requirements list three that are worth checking before anything else: the store must display contact information — email, phone, mailing address or a contact form; refund policy and terms of service must be visible in the footer menu; and in countries that require it, "If your offers don't include shipping information, then the offer is disapproved".

All three are fifteen minutes of work and all three will disapprove your entire catalogue while a team is busy rewriting titles. Rule them out first.

The native channel, with data fixed in Shopify. The Google & YouTube channel syncs your products to Merchant Center without an additional licence. For a single-channel merchant with a catalogue under a few thousand SKUs, this plus disciplined product data is the correct answer, and the discipline is the whole job: a Google product category on every product, a barcode where one exists, a vendor that is genuinely the brand, real descriptions.

Metafields as deliberate overrides. The middle option, and the one we reach for most. Define metafields for the handful of attributes where the channel genuinely wants something different from the storefront — most commonly a feed title — and let the sync read them. The override lives in Shopify, in a defined metafield with a documented purpose, so it is visible to whoever edits the product and it is backed up with everything else. Treat these definitions as a contract: named, described, and owned by someone.

A feed management platform. These are good tools and they solve a real problem, which is not the one most merchants buy them for. They earn their cost when you are running several channels with contradictory requirements — a marketplace that wants flattened variants while Google wants them grouped, a comparison engine with its own taxonomy, a market that needs different identifiers — or when the transformation logic is genuinely complex and rules-based. If you are on one channel and the products are wrong, the platform will faithfully transform wrong data into a different shape.

The question we ask before anyone buys one: name the two channels whose requirements conflict. If you cannot, the problem is upstream, and connecting the catalogue correctly is the cheaper project. That is the same argument we make about most [integration work](/services/integrate) — a transformation layer over bad data is a way of paying monthly to avoid a one-off piece of governance.

Your product page title is written for someone who has already arrived and knows roughly what they are looking at. A shopping feed title is matched against a query typed by someone who has not arrived. Those are different jobs, and this is the one place where the feed legitimately wants something the storefront does not.

The pattern that works is dull: brand, product, the distinguishing attributes a shopper would actually type — material, size, colour, capacity — in the order someone would say them. "Aurora" is a product name; "Aurora merino wool jumper, women's, navy" is a title that can be matched. You have 150 characters and you should not need most of them.

Where this transformation lives is the decision that matters. In a metafield in Shopify it is visible, reviewable and survives you changing feed tools. In a rule inside a feed platform it is invisible to everyone who works on products, and it is the first thing that gets lost in a migration. Same output, very different half-life.

Google supports supplemental feeds that add or override attributes on top of a primary feed, and they are genuinely useful for things that have no home in Shopify — a channel-specific promotion identifier, a temporary exclusion list, an attribute you are testing.

They are also the easiest way to end up with product data nobody can trace. Our rule is that a supplemental feed is allowed to be temporary or channel-specific, but not both permanent and substantive. If an override has been in place for six months and changes what the product *is*, it belongs in Shopify as a metafield.

The test is simple and worth applying annually: can you reconstruct exactly what is being sent to each channel from data held in Shopify, plus a documented transformation? If the answer requires logging into a fourth tool and reading rules somebody left behind, you have a governance problem rather than a feed problem.

Feeds fail quietly. There is no error page, no alert in the store, and the symptom is a gradual decline in impressions that looks like the market being tough. So somebody has to own the Merchant Center diagnostics tab the way somebody owns the fulfilment queue.

A weekly ten-minute check covers most of it: the count of active versus disapproved products against last week, any new disapproval reason, and a spot check of three products comparing the feed values against the live page — price, availability, title, image. The spot check is the part people skip and the part that catches the interesting failures, because the aggregate counts look fine when a hundred products have the wrong price.

Then the seasonal one: before a peak trading period, confirm that whatever changes prices or availability during a sale — an app, a market price list, a discount mechanism — produces a feed that still matches the page. A feed that disagrees with the landing page during your biggest week is the version of this problem that actually costs money.

• Shopify integrations
• Product data governance: who is allowed to change what
• Marketplace listings and the inventory they will oversell