Google Shopping feed specification

The spec is a field-level contract. Each attribute has a name, a format, and a rule about when it applies. Merchant Center validates on ingest, then re-checks the values against your landing page.

The base required set applies to every item you submit:

Attribute	Format	Example value
id	Max 50 chars, stable, unique	HB-38-16-G8-2
title	Max 150 chars	3/8-16 x 2 in. Grade 8 Hex Bolt, Yellow Zinc
description	Max 5,000 chars	SAE J429 Grade 8 hex cap screw, 150,000 psi tensile...
link	Verified domain URL	https://example.com/p/hb-38-16-g8-2
image_link	100x100 px min (250x250 apparel), no watermarks	https://cdn.example.com/hb-38-16-g8-2.jpg
availability	Enum	in_stock
price	Number + ISO currency	12.40 USD

Those seven get you accepted. They do not get you matched, and matching is what determines whether the item ever appears against a query.

Google treats brand, gtin, and mpn as conditionally required: they apply unless the product genuinely has no manufacturer identifier. The logic is strict.

• If the item has a manufacturer-assigned GTIN, send it. Omitting it on a branded product triggers a disapproval, and a disapproved item is gone from Shopping until you fix it.
• If there is no GTIN but there is a part number, send brand and mpn together.
• If neither exists, meaning custom fabrication, cut-to-length wire, or your own private-label line without barcodes, set identifier_exists to no.

The failure mode in B2B is absence. A distributor listing a UL listed 600V wire connector often has the manufacturer's catalog number sitting in the ERP as a free-text string, and no GTIN anywhere. The supplier's PDF has it. Nobody extracted it.

Setting identifier_exists: no to make the error disappear is the common shortcut. It works, briefly. Then the item stops matching to the product cluster and drops out of the comparison listings that cluster feeds. It underperforms quietly, with a green checkmark in Merchant Center.

Reading the spec as a binary pass/fail misses most of its value. Three tiers do different work:

Tier	Examples	What happens if missing
Required	id, title, price, image_link	Item disapproved and never served
Conditionally required	gtin, mpn, brand, condition, color, size, age_group, gender, item_group_id, shipping	Disapproved only when the condition applies; silently weak otherwise
Recommended	google_product_category, product_type, additional_image_link, product_detail, material, pattern	Accepted, but matching and filtering degrade

google_product_category is the clearest case. It is technically optional, but it determines tax treatment, category policy, and how Google interprets the item. Leave it blank and Google infers a category from your title and description. On a catalog of 40,000 fasteners with terse titles, that inference is a coin flip.

Variants are the other quiet trap. A 1/2 in. and a 3/8 in. chuck version of the same 20V cordless drill need a shared item_group_id plus a differentiating attribute. Without the grouping, they compete against each other.

Format validation is only half the bar. Google crawls the landing page and compares what it finds there against what you sent. Three checks account for most of the surprises.

• Price. The price value has to match the price a buyer sees on the PDP, in the same currency, with tax handled the way the destination country expects. A feed that lags a promotional price change by a day gets flagged for price mismatch.
• Availability. If availability says in_stock while the PDP shows a backorder notice or a greyed-out add-to-cart button, the item is disapproved for availability mismatch. Login-gated B2B pricing makes this harder, because the crawler sees the anonymous view of the page.
• Image reachability. image_link has to resolve for Googlebot, with no login wall, robots.txt rule, or hotlink protection in the way. A placeholder graphic counts as a failure even when the URL returns 200.

This is why a feed can satisfy every format rule in the spec and still sit at zero impressions. The spec describes the values. The crawl decides whether they were true.

The spec assumes you have the values. Distributor and manufacturer catalogs usually don't, because the attributes were never captured at item setup.

What that looks like in practice:

• Titles inherited from a 40-character ERP field: HEXBLT 3/8-16X2 G8 YZ
• Descriptions that are a supplier's copy-paste, or empty
• GTINs present on the A-items, absent across the long tail
• color and size living inside the title string instead of in fields
• Images on a shared drive, with no URL Google can crawl

A PIM will store gtin, color, and material the moment someone fills them in, and flag the cell while it is empty. Flagging an empty cell is not the same as filling it.

That gap is a data-completion problem. Someone has to read the manufacturer's spec sheet, the UL file, and the supplier PDF, extract the thread pitch and the voltage rating and the GTIN, normalize the units, and write the values back. The PIM stores your product data; completing it is separate work.

The practical test: run a fill-rate report against the required and conditionally required fields for your top revenue categories. If gtin, color, and size come back mostly empty, the spec is not your problem. The empty cells are.