WooCommerce stores product information in several different places: core product fields, global attributes, custom attributes, taxonomies, brand plugins, and post metadata. Amazon Personalize does not automatically understand that structure. An agency must create a deliberate mapping from WooCommerce product data to the item metadata used by Amazon Personalize.
The quality of this mapping affects more than product filtering. Item metadata helps Amazon Personalize identify relationships between products, interpret product categories and brands, and produce recommendations when interaction history is limited.
Separate the item identifier from descriptive metadata
Every product sent to an Amazon Personalize item dataset needs a stable ITEM_ID. In WooCommerce, this is usually the product ID, provided that the integration consistently uses the same level of the catalog.
For example, an integration can use parent product IDs for recommendations such as “similar products,” while using variation IDs when size, color, or other variation-level choices need to be recommended independently. Mixing parent and variation IDs in the same item dataset creates ambiguous relationships and should be avoided.
- ITEM_ID: A stable identifier that is also used when recording interactions.
- ITEM_NAME: A human-readable title, if useful for the selected recipe or dataset.
- BRAND: A categorical field containing a normalized brand value.
- CATEGORY: A categorical field representing the product taxonomy.
- ATTRIBUTES: Additional categorical, textual, or numerical fields selected for recommendation use.
The item identifier should not be based on a product title, SKU that may change, or a URL slug that can be edited. If the WooCommerce identifier is used, the same identifier must be used in the catalog export, event payloads, recommendation requests, and any subsequent update files.
Model brands as first-class categorical data
Brand data is often stored by a plugin rather than by WooCommerce core. Depending on the site, it may be a custom taxonomy such as product_brand, a plugin-specific taxonomy, a custom field, or a relationship to a separate brand post type.
Before creating an export, identify the actual source of truth and make the output consistent. A brand value should generally be normalized to a stable, readable label such as Acme, rather than a display string that changes according to the current language, template, or marketing campaign.
For a product with one brand, an item record might contain:
ITEM_ID,BRAND,CATEGORY,PRODUCT_TYPE
1842,Acme,