A useful product feed contains the correct attributes for the product and market. The target is accurate, applicable data—not thirty columns in a magic sequence.
Google's product data specification, reviewed November 4, 2026, defines required and conditional attributes. Column order is not a demonstrated mechanism for increasing impression share.
Use a field ledger:
| Function | Example fields | Check |
|---|---|---|
Identity | id, gtin, mpn, item_group_id | Stable item identity and valid product/variant relationships |
Content | title, description, product_highlight | Accurate buyer-useful facts that agree with the page |
Destination and media | link, image_link, additional_image_link | Correct destination, licensed media and usable images |
Offer | price, sale_price, availability, condition | Current values consistent with checkout |
Product attributes | brand, color, size, material | Applicable, supported attributes for this product |
Operating labels | custom_label_0 through custom_label_4 | Internally defined grouping, such as margin or seasonality |
Use a legitimate GTIN where one exists; do not invent an identifier for a product without one. Google's identifier-exists guidance covers that distinction.
Custom labels can make margin, seasonality, best-seller, clearance and competitive groupings easier to manage. They are operating inputs, not a guarantee of better bidding.
The earlier 35% to 68% impression-share story did not establish that column count or sequence caused the change. Validate feed eligibility and item-level errors first, then compare a bounded, documented content change using compatible exposure and measurement windows.
Take the idea further
Review the feed requirements