Skip to main content

​
Steps

  1. Go to Sort Orders and click Create.
  2. Add simple rules (like newest or price) or combine a few signals.
  3. Give the sort order a clear name so your team knows when to use it.
  4. Save your sort order.

​
Sorting attribute types

The Add Expression menu offers simple attributes (Metric and Product Attribute) plus several specialized expression types: Promote and Demote (priority rules), Weighted Group, Soft Boost, Product Diversity, Sequence Grouping, and Shopper Response (Beta). On relevance sort orders, the relevance expression is labeled Search Relevance.

​
Simple attributes

Sort by product attributes like price, title, or sales metrics. Simple attributes are the most common sort expressions.
You can attach an Only when… condition to any value sort term so products that don’t qualify fall through to the next expression instead of ranking on an unreliable signal. See Conditional sort expressions for the full guide.

​
Date bucketing

When the sort attribute is a date (for example, created_at, published_at, or updated_at), a Bucket dropdown appears next to the direction control. Select a bucket to round each product’s date down to the start of that period before Layers sorts on it.
  • Options: No bucket (default), Hour, Day, Week, Month, Quarter, Year.
  • When to use it: If your store launches many products in the same short window, raw timestamps make every product a unique value. A date attribute placed after another sort then acts as a fine-grained tiebreaker and can shuffle the primary order. Rounding to Day, Week, or Month groups products that shipped in the same period so the earlier expression stays in control and the date only breaks true ties.
  • How it applies: Layers applies the bucket wherever the date attribute contributes to sorting: as a value sort, as a tiebreaker, as a feature inside a weighted group, or as the base attribute a soft boost modifies.
Leave Bucket on No bucket to keep the previous behavior of sorting on the exact timestamp.

​
Priority rules

Boost or demote products matching specific conditions. Priority rules cluster matching products at the top or bottom of results based on their direction setting:
  • Descending direction: Promotes matching products to the top
  • Ascending direction: Demotes matching products to the bottom
Layers applies priority rules before your other sort expressions as hard overrides, so their position in the sort group usually doesn’t matter: promoted products land at the top and demoted products at the bottom, and the remaining expressions order products within each group. One exception: a promoting priority rule placed below a soft boost stays in its listed position, so it doesn’t split up the boost’s spaced-out matches. Demoting priority rules always apply first. See Priority rules.

​
Soft demotion

When demoting products in a search context, enable Soft Demotion to apply threshold-based relevance demotion. Items above the relevance threshold are immune to demotion, while items below receive proportional demotion. When available:
  • Only appears for priority rules on relevance sort orders
  • Only applies in search context (not browse)
  • Only available for demotion direction (ASC)
How it works:
  • Items with relevance score >= threshold → immune to demotion
  • Items with relevance score < threshold → proportional demotion based on distance below threshold
Relevance threshold:
  • Default: 0.5 (range: 0-1)
  • Higher threshold (e.g., 0.7): Only highly relevant items are immune; more products receive demotion
  • Lower threshold (e.g., 0.3): More items are protected; only weak matches receive demotion
UI control: When creating a demotion priority rule on a relevance sort order, a Soft Demotion toggle appears. Enable it to reveal a Relevance Threshold slider that controls which products are immune to demotion:
  1. Enable the Soft Demotion toggle
  2. Adjust the Relevance Threshold slider (0-100, representing 0.0-1.0)
  3. The slider displays the current threshold value
  4. Items with relevance at or above this threshold are immune to demotion
Falls back to hard demotion in browse context. Available operators: Text operators:
  • equals / does not equal - Exact match
  • contains / does not contain - Partial text matching
  • begins with / does not begin with - Prefix matching
  • ends with / does not end with - Suffix matching
  • in / not in - Match any value in a list (enables multi-select)
  • is null / is not null - Check if attribute has a value (hides value field)
Numeric operators:
  • equals / does not equal - Exact numeric match
  • greater than / greater than or equal - Value comparisons
  • less than / less than or equal - Value comparisons
  • between / not between - Range matching (requires two values)
  • in / not in - Match any value in a list (enables multi-select)
  • is null / is not null - Check if attribute has a value (hides value field)
Date operators:
  • equals / does not equal - Exact date match
  • after / before - Date comparisons
  • between / not between - Date range matching (requires two values)
  • in the last (days) - Matches dates within the last N days of now. The window stays current as time passes, so it always tracks a rolling period. Use it for “new arrivals” style rules that should follow the calendar without manual updates.
  • is null / is not null - Check if attribute has a value (hides value field)
Tag operators:
  • contains / does not contain - Check if tag exists
  • in / not in - Match any tag in a list (enables multi-select)
Examples: Promote scenario - Priority rule with descending direction, placed first:
Result: All Nike products appear first, then remaining products sorted by sales. The priority rule is the primary sort because it is first in the list. Demote scenario - Priority rule with ascending direction, placed after metric:
Result: Products are primarily sorted by title. Nike products are demoted to the bottom as a tiebreaker since the priority rule is after the primary sort. Soft demotion scenario - Threshold-based demotion in search:
Result: Clearance items above the 0.6 threshold are immune to demotion. Clearance items below the 0.6 threshold receive proportional demotion based on how far below the threshold they are. Only works in search context. Multi-value selection - Using in or notIn operators: When you select the in or notIn operator in the UI, the UI automatically enables multi-select, so you can select multiple values in a single rule.
Result: All products from Nike, Adidas, or Puma appear first, then other brands sorted by sales.
Result: Products with any of the specified tags appear first, sorted by price, then remaining products. Null operators - Check if attributes are set: When you select is null or is not null, the UI automatically hides the value field since these operators check for presence/absence of data.
Result: Products on sale (with compare-at price set) appear first, sorted by discount.
For a stricter definition that only counts products with a compare-at price greater than the current price, use the on_sale system attribute: on_sale equals true. See On-sale attributes.
Result: Products with inventory tracking appear first, products without tracking at the bottom. Range exclusion - Using not between:
Result: Products priced outside $100-$200 appear first, products in that range at the bottom.

​
Weighted groups

Combine multiple attributes and metrics into a single sort score. Each feature contributes a configurable share of the blend, and you choose whether higher or lower values are better on a per-attribute basis. See Weighted groups for the full guide. Configuration:
  • Expressions — the attributes and metrics to blend. Use the picker on each row to select a feature. A group can contain a single feature, though a blend is most useful with two or more. The metric picker groups metrics by section (with your three most recently used metrics floated to the top) and offers a Create metric… action to build a new metric inline without leaving the sort order. See Weighted groups for details.
  • Preference (attribute rows only) — Higher is better or Lower is better. Sets the direction each attribute contributes in. Metric rows default to higher is better.
  • Weight — a share of 100, controlled with a slider. Layers automatically normalizes weights so they always sum to 100; dragging one slider rebalances the others proportionally.
  • Final Sort Direction — set in the group settings sheet. Controls how products are ordered by the final blended score (Descending by default).
  • Conditions — optional contextual conditions in the group settings sheet that gate the entire group. When the conditions don’t match a request, the whole weighted group is skipped.
A Blend preview appears on hover in the group header, showing how the configured weights divide up the final score. Compose with other expressions: A weighted group is a first-class base in the sort pipeline. You can drop a soft boost row directly below the weighted group, and the modifier attaches to the blended score the same way it would attach to a single attribute or metric.

​
Soft boost

Lift products matching conditions toward a target score while maintaining natural ranking. Unlike priority rules, soft boost creates interleaving rather than clustering. Each soft boost appears with a Modifier badge and is visually grouped beneath the metric or attribute it modifies, so you can see at a glance which expression the boost targets. Key differences from priority rules:
  • Natural interleaving: Boosted products are distributed throughout results based on their base score.
  • Capped lift: Products are only lifted as far as needed to reach the target band.
  • Requires a target expression: Soft boost modifies the next metric or attribute in the list.
Configuration:
  • Attribute: The product attribute to check (for example, tags, vendor, options).
  • Operator: Condition operator (see available operators below).
  • Value: The value to match (multi-select available for in and notIn operators, hidden for is null and is not null).
  • Boost Mode: Select Nudge matches up (multiplicative, default) or Lift matches into view (additive).
    • Nudge matches up: Gives matches a small multiplier that’s strongest on low scores and fades on high ones. Products with no metric data don’t move.
    • Lift matches into view: Best for products that may not have metric data yet, like new arrivals or restocks. Places matched products at the chosen spot in the results, including ones that would rank higher on their own.
  • Boost Strength (Nudge matches up only): Amplitude of the multiplier applied to the base score (default: 0.25, range: 0-10; the editor slider covers 0-2, and negative values aren’t allowed). Lift matches into view does not use this value.
  • Percentile Target (Lift matches into view only): Where boosted products land (default: 50, range: 0-100). With spacing on, 75 starts matches about a quarter of the way down the results. With spacing off, matches are lifted to the base metric’s value at that percentile.
  • Min positions between (both modes): Number of non-matched products between two consecutive boosted matches in the final results (default: 10, range: 1–50). Prevents clustered matches from landing adjacent. Layers treats the value as a maximum and tightens the gap automatically when a smaller result set cannot fit the full spacing. In Nudge matches up, a single match stays put and only clustered matches are spread. In Lift matches into view, matches are laid out from the percentile anchor.
Available operators for soft boost: Soft boost supports the same operators as priority rules, including:
  • Text operators: equals, does not equal, contains, does not contain, begins with, does not begin with, ends with, does not end with, in, not in, is null, is not null
  • Numeric operators: equals, does not equal, greater than, greater than or equal, less than, less than or equal, between, not between, in, not in, is null, is not null
  • Date operators: equals, does not equal, after, before, between, not between, is null, is not null
  • Tag operators: contains, does not contain, in, not in
Example (Nudge matches up): Boost products tagged “featured” while sorting by sales:
Example (Lift matches into view): Sprinkle new arrivals with zero sales into best sellers:
Example (Multi-value): Boost products with any of multiple tags using in operator: When you select the in or notIn operator in the UI, multi-select is automatically enabled.
When to use Lift matches into view:
  • Boosting products with zero or near-zero base values (new arrivals with no sales)
  • Creating a specific target position in the sort order
  • “Sprinkling” new products into established rankings
Soft boost requires a metric or product attribute to modify. If no sortable expression exists yet, an error dialog asks you to add one first. When you add a soft boost, Layers automatically inserts it above the last metric or attribute so that it targets the correct expression.
Learn more about soft boost in the Soft Boost Sorting documentation.

​
Product Diversity

Spread products from the same group across results so a single group doesn’t dominate the top of a collection. Add one or more diversity axes. Each axis caps repeats on a different dimension (product family, product type, vendor, or any categorical attribute). See Diversity expressions for the full set of options.

​
Sequence Grouping

Keep related products grouped together according to a configured sequence while the rest of the sort order ranks around them. Adding a Sequence Grouping expression enables sequences for a sort order. Once you save the expression:
  • The sort order appears in the sort order dropdown when you create a product sequence.
  • Selecting the sort order in a merchandising rule shows the Enable Sequences toggle, which turns on automatically. Sort orders without a Sequence Grouping expression don’t show that toggle.
See Product sequences for how sequences behave across collections.

​
Shopper Response (Beta)

Let shopper behavior fine-tune the top of your collection pages. A Shopper Response expression takes the order produced by the expressions above it and reorders the first 48 products based on how shoppers respond to them in that collection. Products that get more clicks, add-to-carts, and revenue than their position would predict move up, and products that underperform move down. How it works:
  • Layers looks at the last three weeks of shopper activity in each collection, with recent days counting more.
  • Every day, Layers tests different ways of blending these signals with your order against a held-out week of activity. It keeps a blend only when it would have improved shopper engagement, mainly add-to-carts, over your order. If no blend helps, your order is kept as is.
  • Layers moves each product only a few positions from where your sort order placed it.
  • Pinned products and soft-boosted products keep their positions. Other products are reordered around them.
  • Products with too little activity to judge are treated as average, so the shopper signals don’t push them up or down.
Settings:
  • Top-24 churn cap: Choose how many products may leave the top 24 because of the reordering: up to 2, 4, or 6. The default is 4. A lower cap keeps your first page closer to your original order.
Below the cap, a comparison table shows how each cap (2, 4, and 6) performed in the latest tune, with your selected cap highlighted:
  • Add-to-cart vs random reshuffle, last week: the estimated add-to-cart gain, in percentage points, measured on last week’s traffic. A cap reads no gain when no blend beat your order, in which case your order is kept.
  • Products leaving the top 24 (median): how many products typically leave the top 24 at that cap.
A cap reads Waiting for the first nightly tune until Layers has results for it. When two or more caps tuned to the same order, a note such as Caps 4 and 6 tuned to the same order tells you that switching between them won’t change what shoppers see. A freshness line shows when the results were last updated and how many collections have enough shopper data. When you first add the expression, Layers starts tuning in the background, then refreshes the results nightly. To see how the expression changes a specific collection, open the preview. See Shopper Response in the preview. Where it applies:
  • Collection pages only. Search results and recommendation blocks that use the same sort order are not reordered.
  • When product sequences are enabled for a collection, the sequence order is kept and the expression has no effect there.
Requirements:
  • Shopper Response must be the last expression in the sort order, below at least one expression that orders products.
  • You can add only one Shopper Response expression per sort order.
  • Shopper Response doesn’t support Only when… conditions.
Shopper Response is in beta. Consider a sort order experiment to compare a sort order with and without it before you roll it out widely.

​
AI-suggested expression names

When you add or edit a sort expression, Layers automatically suggests a short, plain-language label for it based on the attribute, operator, and values you configured. The suggested name appears as the card title in the editor. How it works:
  • As soon as you finish editing an expression, Layers sends the configuration to a small language model and replaces the placeholder title with the generated name.
  • Names are deterministic for the same configuration and cached, so identical expressions across different sort orders share the same label.
  • You can always rename an expression manually by clicking the title. Your edit is preserved and won’t be overwritten.
Examples:
Suggested names make sort orders easier to scan when you have many expressions. They are especially helpful for priority rules and soft boosts, where the configuration alone can be hard to read at a glance.

​
Tips

  • Start simple. You can refine weights over time.
  • Use soft boost for subtle promotion that maintains organic ranking.