Recommendation filters constrain the candidate pool that an algorithm draws from. Without filters, Personyze considers your entire catalog; with filters, it only considers items that match your criteria. The result is sharper, more contextually relevant recommendations — and protection against awkward outputs (out-of-stock items, off-brand products, items outside the visitor’s price range).
The same walkthrough twice — once for a product catalog, once for content:
Filters against a fixed value, and filters that follow the product on the page.
Filters by author, type or date, and by the topic of the article being read.
Where filters live
In the algorithm step of any recommendation wizard, scroll past the algorithm choice to the Filters area. You’ll find common filters as toggle switches and a button to add custom filters using any product or content attribute.
Common filter examples
How to build a filter
- Click Add Filter.
- In the first dropdown, choose the product attribute (e.g., brand, category, price).
- In the second dropdown, choose the filter value — either a specific value (e.g., “iPhone”) or a dynamic match (e.g., “same as currently viewed product”).
- Repeat for additional filters. Multiple filters combine with AND logic — items must match all of them.
For OR logic, you’d typically use a single filter with multiple values selected (e.g., brand IN [Apple, Samsung, Google]) rather than multiple filter rows.
Dynamic filters (context-aware)
A filter compares each candidate item against a value. The value can be fixed — in stock, a price above 100, one category — or it can come from what the visitor is doing right now. The value menu offers the item being viewed, the last one viewed, the one in the cart, the last one purchased (or that item’s category), and fields from the visitor’s profile. For example:
- Brand equals the viewing product’s brand — the widget only recommends from the brand of the product on the page.
- Price lower than the viewing product’s price — a “cheaper alternatives” row.
- Brand equals the cart item’s brand — more of what they are already buying.
These follow the visitor as they browse — one widget, no per-product campaign. A filter this narrow will sometimes leave the row short; that is what a fallback is for, and the fallback keeps the filter.
Content recommendation filters
Content recommendations use the same filter mechanics but with content-specific attributes (article category, author, publish date, content type, tags). The interface looks slightly different but the logic is the same — pick an attribute, pick a value, combine with AND.
Best practices
- Start broad, narrow gradually. Heavy filtering can produce empty recommendations. Add filters one at a time and verify in the QA step that you still get good results.
- Always include in-stock + active. These are the bare minimum for any production recommendation.
- Pair filters with a fallback. If your filters are restrictive, set a fallback algorithm with looser filters so visitors never see an empty widget.
- Test on real visitor data. Use the QA step with several real subscriber emails to see what filters produce in practice — especially for dynamic filters that depend on browsing context.