Merchant Center Feed Rules: Steps, Examples and Fixes

Merchant Center feed rules are transformations Google Merchant Center applies to your product data during processing, without touching your source file. Google now calls them attribute rules.

I use them to make correct data readable to Google, and that is their limit: a rule fixes how a value is written, not whether it is true. Every operation and limit here comes from Google’s Merchant Center Help, read on 5 November 2026.

What are Merchant Center feed rules?

Merchant Center feed rules, renamed attribute rules, are ordered instructions that change product attributes during processing. They map unrecognised headers, clean up values, fill gaps under a condition and assign custom labels.

You turn them on with the Advanced data source management add-on (Settings, Add-ons, Discover). Then open Data sources, pick a primary product source and go to its Attribute rules tab.

Older guides say “feeds” and “feed rules” for the same screens.

Google’s attribute rules documentation splits the operations into two groups: ones that choose where a value comes from, and ones that modify it.

What can a feed rule not fix?

A feed rule cannot make a wrong fact right. Mapping brand_name to brand tells Google where the brand lives, but it can’t correct a wrong name in that column.

Keep core product facts in the source: identity, description, images and manufacturer data. A rule that hides a source error only adds a dependency for the next person auditing the feed.

Who owns a feed rule?

The harder problem in most accounts isn’t a missing attribute; it’s ownership.

A product you discounted yesterday still sits in a high-margin campaign, and nobody knows whether the shop export, a supplemental feed or an old rule wrote that label. Treat rules as maintained production logic with an owner, not fixes you save and forget.

How do you create a feed rule in Merchant Center?

You create a feed rule on a primary data source’s Attribute rules tab: add a rule for one attribute, save it as a draft, test it, apply it and reprocess the source.

Prepare first, because testing only shows the products you think to check.

Why does rule order matter?

Rule order matters because Google runs attribute rules in cascade: the first rule runs, then the second reads its result, and so on. The safe order is map headers first, normalise values second, apply scoped defaults last.

The reason is the gap. A default placed first fills condition before you learn the source never sent it.

A stock rule that runs before availability mapping reads the raw text, not the cleaned value. When a label looks wrong, trace the value at the point where that rule runs, not only in the final product record.

The test that matters most is on the products a rule must leave alone, not only the ones it should repair.

A costly slip is saving a rule against a supplemental source when the input belongs to the primary source: confirm which source supplies the attribute before you debug the condition.

Feed rule examples you can copy

These patterns are starting points, not configurations to paste unchecked. Each one assumes you’ve looked at the raw values first.

Normalise codes with a condition, not a default

Say a fashion shop receives F, M and Female in its gender column. Map F and Female to female and handle M explicitly. A product-wide female default would erase valid differences instead of cleaning them.

Extract only from a dependable pattern

Extract works when every title follows the same pattern. Test it on the messiest titles in the feed: punctuation, missing size markers, bundles. If the formats differ much, a structured size column beats ever more complicated title parsing.

Take latest compares sources, not column names

Take latest picks whichever data source updated price or availability most recently; Google limits it to those two attributes.

It can’t tell that a price_2026 column is newer than price_2025 in the same file. If your export carries both, point Set to at the right column, or fix the export.

Don’t manufacture missing product facts

Google’s local product-data specification requires brand for new products, except movies, books and musical recordings, and GTIN for new products that have one assigned by the manufacturer.

The same page warns against using your store name as the brand unless you make the product. Scope brand-fill rules to that, and get assigned GTINs from the manufacturer rather than inventing them.

condition is required for used and refurbished products, so a blanket new default on a mixed range describes some products wrongly. Scope it to a verified ID list.

For enrichment at the source, the guide to Shopify catalog improvements with ECORN covers the upstream work, and Google’s product categories need the same care: they’re more than another text clean-up.

How do custom labels work with feed rules?

Custom labels are five optional attributes, custom_label_0 to custom_label_4, that Google never shows to shoppers and that you use to split products in Performance Max, Shopping and Demand Gen campaigns.

Google’s custom label specification allows one value per product per field, 1–100 characters, and up to 1,000 unique values per field across the account. Values beyond that are ignored for reporting and bidding.

Spend a field only on a split someone will use to set campaign priorities. Fix its vocabulary before anyone writes to it: high_margin, low_stock, star_candidate.

Google treats label values as not case sensitive, but high-margin and high_margin are still two values in two product groups.

Why do commercial labels go stale?

A margin or stock label is only as current as the data behind it. Feed rules reapply on every processing run, but margin, returns and stock cover usually live outside the feed.

At Product Metrics, Product Rules run after segmentation and write labels to a Merchant Center supplemental feed.

For example, the Promote Star candidates rule sets custom_label_0 to star_candidate when POAS is at least 1.4, margin at least 40% and stock cover over 8 weeks. That rule is only right while its inputs are.

A discount changes the margin, a delivery changes stock cover and ad results change POAS, so reapplying a saved condition to stale inputs doesn’t help.

When a product’s segment changes, its label changes and Merchant Center merges it on the next fetch (06:30 in our setup). The product then moves to whichever campaign or asset group your team split on that label.

Check the campaign impact before you publish

Tightening a threshold can empty the campaign it feeds. Preview the split before changing the logic, and decide what happens to a product that stops qualifying.

Doing these checks by hand works for a short product list. Across a full range, the labels that drive your priorities need to come from data that updates itself.

Why do feed rules break, and how do you fix them?

A feed rule that worked yesterday and fails today usually means its input changed, not the rule: a renamed header, new stock wording or a supplemental source that stopped updating.

Start from the symptom and inspect the raw input before you edit the rule again.

Diagnose output that’s valid but commercially wrong

Some failures never produce an error. A feed can process cleanly while its labels describe yesterday’s stock or last week’s price.

Compare the freshness of the supplemental and primary sources first. If a supplemental source overwrites current availability with an older value, the stock label is valid and wrong.

Google’s guide to supplemental feeds and feed rules shows how each supplemental feed, such as one for price and one for availability, can carry its own rule set, and notes that rules can only be created in Merchant Center, not through the Content API.

Check your product feed sync schedules against your fetch and processing history.

Keep a regression sample and stable IDs

Keep a fixed sample of products in a spreadsheet: out-of-stock items, regional differences and variants. After every material edit, compare expected and processed output, then read the item-level diagnostics once processing finishes.

Keep IDs identical across sources. A supplemental row with a different ID never matches its product, so the rule’s input was never there.

For the errors Merchant Center does report, the Merchant Center errors guide traces each one back to its source.

Which rules should stay manual, and which should you automate?

Keep stable formatting rules in Merchant Center and automate labels built on changing commercial data. Header mapping rarely changes; labels built on margin, returns and stock need fresh inputs every day.

Feed rules already run automatically, so automation isn’t about execution. It’s about who keeps the inputs fresh and who checks that a label still supports the campaign decision.

With Product Segmentation, labels reach Merchant Center through one supplemental feed, with no feed rules to maintain, while your team still chooses the logic and the campaign priorities.

Hand over one label at a time

List every custom label you maintain by hand and mark those driven by margin or stock. Pick one, write the automated version to a spare field and compare the two before you remove the manual one.

The spare field keeps the two approaches from overwriting each other. Investigate disagreements, confirm the campaign routing and name an owner before you switch. Automation removes the rule maintenance, not the responsibility for the decision.

Final thoughts

The most useful habit with feed rules is boring: write down what each rule reads, what it writes and when its assumptions expire. Formatting rules belong close to the feed; commercial labels belong close to the data that decides them.

Once the feed is clean, the work moves to campaigns and the shop.

If you’re also building acquisition campaigns, SimpleLab Digital’s guide on how to create Google search campaigns covers that separate workflow; just don’t use campaign changes to compensate for product data you haven’t validated.

And a clean feed only gets shoppers to the product page. Magic Genie’s guide to turning store traffic into more sales covers the merchandising and conversion work a better feed supports but can’t replace.