NEW: Metafield rules for recommendations
Your fine-tuned versions of EliteAI™ Ultra can now build metafield rules for recommendations, so a shopper with a lamp in the cart gets offered the bulb that actually fits it.
The Update
A lighting retailer asked for something simple: when someone has a lamp in the cart, recommend the bulb that fits it. Tags could express that, but only by hand-maintaining a matching tag on both sides of every pairing. Add one product and you add two tags. Forget one and the pairing quietly stops working.
Most stores already hold that relationship as proper product data: a fitting code, a compatibility group, a model number, stored in a product metafield. Metafield rules read that data directly, so the association you already maintain becomes the recommendation rule.
In the strategy editor, the lighting example reads as one rule:
If the trigger item metafield
custom.bulb_fithas a value, then recommend products that have the same metafield value as the trigger item and are product type LED bulbs.
What's New
Match on the trigger item's own value: Have the same metafield value as the trigger item is the relational option. Any product sharing a value with the item in the cart is a match, with no list of values to maintain.
Or match against values you choose: Have one of these metafield values, do not have one of these metafield values, and have a metafield value starting with a prefix, for prefix schemes like gu10-.
Scope the rule: Metafield is also an If the trigger item condition, including has a value and has no value, so a rule only fires for products that carry the association.
Lists work too: Single-line text metafields and lists of them are both supported. With a list, sharing any one value counts as a match, which is how a product that fits several models still pairs correctly.
Pick it from a dropdown: You choose from your own product metafield definitions, the ones you created under Settings → Custom data. Reserved Shopify and app namespaces stay hidden.
Metafield Rules Work Without Purchase History
This is the part that matters most. A lamp and its bulb that have never been bought together still get paired, because the engine fills recommendations from your catalog using your rule rather than waiting for co-purchase data to build up.
That makes them a good fit for new products, long-tail catalogs, and compatibility that is true from day one.
Impact on Your Store
- Replace hand-maintained pairing tags with data you already keep in Shopify.
- New products inherit correct recommendations the moment their metafield is filled in.
- Combine a metafield rule with your price, collection, and product type rules in the same version.
Configuration
Upsells → AI upsells → Fine-tune EliteAI™ Ultra → your version → Filters. Add a filter, pick Metafield, choose the field and the match type, and save.
Good to know
- Supported types are single-line text and lists of single-line text. Numbers, dates, references, and rich text are not available.
- The metafield needs a definition in your shop. Values set without a definition cannot be selected.
- You can use up to 5 different metafields across all your fine-tuned versions.
- Rule changes take effect on the next training run, not instantly.
Why This Matters
The relationship between a product and its accessory is usually something you already know and already store. Metafield rules put that knowledge to work on every recommendation.
For every rule type with examples, see our Filters and Boosts help article, or start with the Fine-Tuning EliteAI™ Ultra overview.