Similar Products vs Frequently Bought Together
Choose recommendation logic for comparison, accessories, baskets and repeat purchase. See where similar products and frequently bought together each belong.
Similar products vs frequently bought together: which belongs where?
A recommendation can be relevant to a product and still be wrong for the moment. Similar items help shoppers choose between options. Complementary items can help complete a purchase. Start with the decision the shopper is making, then choose the relationship, eligibility rules and placement that support it.
Start with the shopper's next decision
A visitor viewing a sofa may still be comparing styles. A shopper who has added one to the basket may be deciding on delivery or a useful accessory. Recommending another sofa in both places treats different decisions as the same task.
Define the placement's purpose in one sentence. “Help the shopper choose an alternative” calls for similarity or substitution. “Help complete this purchase” calls for complements or compatible accessories. “Help replenish a known item” calls for repeat-purchase evidence and timing. The label above the carousel should match that purpose.
Similarity is a comparison relationship
Similar items can share product type, appearance, attributes or price band. Decide which similarities matter for the query or current product. A visually similar alternative may be useful for fashion comparison, but a replacement part needs stronger compatibility evidence.
When an item is unavailable, explain that suggestions are alternatives. Do not imply identical specifications merely because a model assigns a high similarity score. The underlying relationship should be understandable to a category reviewer.
Frequently bought together is behavioural evidence
Co-purchase can suggest a complement, but the relationship should still be checked against the placement. A household buying two alternative brands in one order does not prove every shopper needs both. A promotion can also affect what appears together in a basket.
Use product constraints and basket context to remove irrelevant additions. Avoid recommending something already selected unless extra quantity is a deliberate part of the use case. For technical accessories, check compatibility independently of co-purchase frequency.
Match the method to the surface
On a product page, test separate comparison and complement modules when both decisions matter. In the basket, prioritise useful additions and avoid restarting a choice the shopper has made. At checkout, protect purchase completion and consider whether the extra decision is justified. These are testable design choices rather than universal placement rules.
PA's Recommendations API documents current-product and cart context inputs, providing concrete integration context for these decisions. API parameters.
Judge the whole purchase outcome
Measure the intended result: discovery, accessory attachment, basket contribution or repeat purchase. Keep conversion and returns or cancellations in view where the retailer can measure them. Clicks alone cannot tell whether a recommendation helped or distracted.
Use the placement worksheet to record the shopper's task, relationship, required evidence, exclusions and primary metric. Then test a small set of representative products with a category specialist before launching a wider experiment.
Downloads
- placement-strategy-worksheet.csv (illustrative template)
Sources
- Recommendations API Parameters, checked 2026-10-25
Frequently asked questions
Are similar products and frequently bought together interchangeable?
No. Similarity usually supports comparison or substitution, while co-purchase suggests a possible complement. Choose the relationship that fits the shopper’s next decision and validate it with product constraints, placement context and appropriate outcome metrics.
Should the basket contain alternative products?
Sometimes, but treat that as a deliberate experiment. Alternatives may help when an item becomes unavailable or the shopper wants a better fit. They can also restart a completed decision. Label the purpose and measure purchase completion as a guardrail.
Does bought together prove compatibility?
No. Purchase records can suggest a relationship but cannot certify that products fit or function together. Use authoritative compatibility data where required, and exclude uncertain pairs from placements that explicitly promise a compatible accessory.
Related guides
- Compatible accessory recommendations for electronics retailers: Build accessory recommendations around verified product compatibility, clear evidence and useful basket context. Explore a data model and practical test fixtures.