> ## Documentation Index > Fetch the complete documentation index at: https://docs.uselayers.com/llms.txt > Use this file to discover all available pages before exploring further. # User Affinity > Personalize collection browsing in Layers by lifting products that match each shopper's learned tastes from their own cart and purchase history. ## Overview User Affinity is a sort expression that personalizes browse results for each individual shopper. It appears in the **Add Expression** menu as **User Affinity** with a **New** badge. Like [Soft Boost](/platform/sorting/soft-boost), User Affinity does **not** sort on its own — it is a modifier that lifts products matching the current shopper's learned tastes *within* the next sort expression below it. Because it always modifies a following sort attribute, **User Affinity cannot be the last expression in a sort order**: it requires a sort attribute beneath it to modify. User Affinity only ever **promotes** matching products. It never demotes — products that don't match a shopper's affinities keep their organic position. User Affinity is **browse personalization**. It applies on collection and browse pages, not on search results. ## How User Affinity works As a shopper browses, Layers builds a lightweight picture of their tastes from their own activity. When that shopper loads a collection that uses a User Affinity expression, products that match their inferred preferences are lifted up within the underlying sort, while everything else stays in its organic order. If a shopper has **no learned affinities yet** — for example, a brand-new visitor — the expression is automatically skipped and the base sort applies unchanged. Shoppers always see a complete, sensibly ordered collection. ### Signals Affinities are learned from each shopper's own **cart adds** and **purchases** within their session context: * **Purchases weigh more heavily than cart adds.** A completed purchase is a stronger signal of taste than an item a shopper added to their cart. * Signals are interpreted in the shopper's session context, so the personalization reflects that shopper's recent behavior. ### Eligible attributes Affinities are powered only by attributes in the **Categorical** and **Feature** [attribute classes](/platform/attributes/attribute-classes). These are the attributes that describe what a product *is* — brand, color, material, product type, size, and similar characteristics — so matching reflects the kinds of products a shopper gravitates toward. If a store has no categorical or feature attributes, User Affinity is unavailable. When you add the expression, the form shows how many categorical and feature attributes currently power affinities. ### Affinity weights The **Affinity Weights** picker lets you choose which categorical and feature attributes drive matching and tune a per-attribute weight. Raise the weight on the attributes that matter most for your catalog (for example, brand for a multi-brand store, or material for an apparel store) and lower or disable attributes that are less meaningful. ## Configuration parameters User Affinity reuses the same boost-mode model as [Soft Boost](/platform/sorting/soft-boost). ### Boost mode Determines how matching products are lifted within the underlying sort. * **Amplify strong performers** (multiplicative) — pushes matching products that *already* rank well on the underlying metric even higher. Use this when matches have real metric data, so personalization reinforces proven performers. * **Lift matches into view** (additive) — surfaces matching products regardless of their metric value. Use this for new arrivals, restocks, or products with little performance data, where you want affinity matches to appear even without a strong base score. The UI defaults to **Amplify strong performers**. ### Boost strength A global multiplier on the affinity lift. Higher values produce a stronger personalization effect; lower values keep the lift subtle so organic ranking remains dominant. ### Percentile target *Additive mode only.* The percentile that matching products are lifted toward, expressed against the underlying metric. A higher percentile lifts affinity matches further up the results. ### Min positions between / decay rate Anti-clumping spacing controls. These keep affinity matches from bunching together so personalized results still feel varied across the page. ### Direction Controls the direction of the affinity lift relative to the underlying sort. ## Preview as a shopper You don't have to wait for live traffic to see how User Affinity behaves. The **Preview profile** shopper picker lets you simulate a shopper's cart and purchase products, then preview how User Affinity reorders a collection for that profile. When you select or build a profile, the preview shows: * The reordered collection as that shopper would see it, with affinity matches lifted into place. * A plain-language summary of the shopper's inferred preferences, so you can sanity-check that the affinities make sense before publishing. Use this to compare how the same collection looks for different kinds of shoppers and to confirm your **Affinity Weights** and boost settings produce the experience you intend. ## Best practices * **Always leave a sort attribute beneath it.** User Affinity modifies the expression below it, so it can never be the last expression in the sort order. * **Pick the right boost mode.** Use **Amplify strong performers** when affinity matches already have metric data, and **Lift matches into view** for new arrivals or low-data products you still want personalized into view. * **Tune Affinity Weights to your catalog.** Emphasize the categorical and feature attributes that best capture taste in your store and de-emphasize the rest. * **Preview across profiles.** Use **Preview profile** to simulate several shopper types before publishing, and confirm the inferred-preference summary matches your intent. * **Start subtle.** Begin with a modest boost strength so personalization complements your organic ranking rather than overwhelming it. ## See also * [Soft Boost](/platform/sorting/soft-boost) — the same boost-mode model, condition-based rather than personalized * [Attribute classes](/platform/attributes/attribute-classes) — the categorical and feature classes that power affinities * [Preview & annotations](/platform/sorting/preview) — preview sort orders and simulate shopper profiles * [Sort orders](/platform/sorting) — full guide to creating and managing sort orders