ParticleSearch Product Recommendations: Strategies, Placements, Controls, and Evidence

A shopper comparing a sofa needs alternatives. A shopper buying that sofa may need cushions, care products, or coordinated furniture. A shopper with several items in the cart needs a small addition that is not already covered. A shopper opening search without a precise query may need new arrivals or best sellers.

These are different jobs, even when they use the same product-card component. ParticleSearch keeps the strategy and placement explicit so the storefront label, candidate relationship, analytics, and experiment all refer to the same decision.

A store does not need to replace a recommendation surface merely because another product exists. The decision begins with a failed contract: the wrong relationship, insufficient placement control, indistinguishable failure states, or no responsible way to compare strategies.

ParticleSearch is strongest when recommendations should share the same eligible catalogue and product-card truth as storefront search, while giving the merchant a named strategy and an observable placement. It is not a substitute for missing compatibility or unreliable source data.

A matching recommendation block must exist in the storefront theme where the shelf should appear. ParticleSearch does not silently insert a product row into an arbitrary part of the theme. That keeps placement and surrounding design under storefront control.

Each supported product or cart placement can remain theme-managed, or the merchant can make the ParticleSearch dashboard the source of truth for the shown state, strategy, heading, and product count. Keeping ownership explicit avoids a common failure where the theme and dashboard appear to disagree because both are assumed to control the same setting.

The visible workflow has two owners. The theme determines where the recommendation block lives. The placement setting determines whether the theme or dashboard owns the strategy, heading, product count, and shown state. Publishing one side without checking the other creates a configuration that looks saved but does not prove a shopper saw it.

On a product page, the anchor product is removed from the visible recommendation set. For contextual strategies, products already represented by the supplied context are also excluded. This prevents the shelf from recommending the item the shopper is already viewing or repeating a product that is already in the cart.

The remaining candidates still depend on the available catalogue, product state, market, and the quality of relationships the store can support. An empty shelf can therefore be the correct result. ParticleSearch records that state separately instead of pretending a broad fallback was an exact relationship.

ParticleSearch experiments can target the recommendation surface and assign a named recommendation strategy to a variant. A recommendation treatment does not also change search relevance in the same treatment. This keeps the decision interpretable.

Choose one primary metric that matches the job, such as a visible recommendation product action or direct add. Protect empty-module rate, duplicate or incompatible candidates, product-page conversion, cart completion, and verified-order evidence as guardrails. Revenue attached to a verified touchpoint supports interpretation, but it is not automatic proof of incremental revenue.

The ParticleSearch experiments guide explains traffic, variants, decision rules, and evidence thresholds.

ParticleSearch cannot know an undocumented compatibility rule, repair a misleading product attribute by presentation alone, or decide whether a recommendation relationship is acceptable for the store’s return risk. Those facts belong in catalogue data, merchant curation, and the acceptance set.

It also does not treat attributed revenue as guaranteed lift. The product makes the strategy, placement, event state, experiment variant, and verified-order connection visible so the team can make a better decision without exposing the system’s internal implementation.