Site Search for Fashion: Style, Size and Availability
Build fashion site search around style, size and availability. Learn where semantic retrieval, exact filters and visual alternatives help, and how to test them.
Site search for fashion ecommerce
Fashion site search must connect descriptive intent with products the shopper can actually buy. Combine semantic and lexical retrieval with reliable size, colour and availability data, then offer clearly labelled visual alternatives when the exact choice is unavailable. Evaluate each step with real fashion queries and variant-level checks.
Start with the shopper’s requirements
Consider a hypothetical search for a floral midi dress in UK size 12 for a summer wedding. The shopper has described a product type, pattern, length, size and occasion. Treat these as separate requirements. A visually similar maxi dress may be an interesting alternative, but it does not satisfy the original request.
Build a query panel that includes brand and product names, style descriptions, misspellings, size systems and combinations of attributes. Include products with seasonal or creative names so the exercise tests more than title matching.
Combine retrieval with trustworthy attributes
Semantic retrieval can connect a description with differently worded catalogue content. Lexical retrieval remains useful for precise names and identifiers. Neither should replace explicit checks for required size, price or availability. Microsoft’s hybrid-search documentation explains why both retrieval methods can contribute.
Keep product-level descriptions separate from variant-level facts. A dress may exist in several sizes while the requested size is unavailable. Confirm what a result card, selected variant and destination page will show. For PA’s Shopify app, field configuration distinguishes product and variant mappings; custom metafields need explicit configuration.
Give unavailable choices a useful next step
Offer a shopper-initiated route to similar items without presenting them as exact matches. Carry forward the selected size and other mandatory requirements when the implementation can enforce them. If no eligible alternative exists, explain that clearly and offer a relevant category or availability notification where supported.
Particular Audience’s visual discovery product includes See Similar experiences on search and listing surfaces. Ask for a demonstration using your imagery and size availability so visual resemblance and purchase eligibility can be assessed separately.
Judge discovery and commercial impact separately
Review whether retrieved products satisfy the query, whether available variants are selectable, and whether alternative clicks lead to useful product pages. Then test effects on completed purchases, revenue per assigned visitor and returns over an appropriate observation window. A higher alternative click rate alone does not establish additional sales.
PA’s published fashion example reports a 60.1% click-through rate for See Similar. The public case does not disclose the denominator, period or control group, so treat it as a deployment example rather than an expected uplift.
Bring difficult queries, a sample of variant data and current recovery screens to a review of your fashion discovery journey.
Carry intent across the fashion discovery journey
Map three handoffs: search to product page, unavailable variant to alternative, and alternative to basket. At each point, record which shopper selections remain active and which would change. The destination should make the selected variant and any price or style difference clear.
For a hypothetical shopper looking for a black blazer in a selected size, a See Similar action may offer a differently cut blazer. Preserve the size requirement where it can be enforced, and identify any changed style preference. If nothing suitable is available, provide an honest next step rather than a false match.
Measure feature exposure, use and purchase separately. A similarity-button click rate uses the people or impressions eligible to see that button, while a purchase rate needs its own population and window. Define those denominators before comparing periods. Assess overall revenue and returns through the experiment’s assigned population when judging commercial impact.
This makes a fashion discovery review actionable: the team can see whether failure comes from retrieval, variant data, the handoff or the alternative itself.
Illustrative example
No numerical illustration needed; qualitative examples in the body are hypothetical unless explicitly attributed.
Evidence
- PA’s published anonymised fashion case reports 60.1% CTR for See Similar; denominator, period and comparator are not disclosed. (Vendor-published engagement rate, not a conversion lift, causal effect or performance forecast.) Source
Sources
- Hybrid search using vectors and full-text search in Azure AI Search, checked 2026-10-25
- Field Configuration, checked 2026-10-25
- Visual Search & Discovery for eCommerce, checked 2026-10-25
How Particular Audience does it
Frequently asked questions
Does visual similarity prove that an item fits?
No. Image resemblance does not establish garment measurements, fit or availability in the selected size. Use verified variant and sizing information alongside visual similarity.
Should fashion search keep synonym rules?
Keep rules that resolve tested catalogue language or brand ambiguities. Remove or change a rule only after checking representative queries; semantic retrieval does not make every rule unnecessary.
How should we test an unavailable-size journey?
Select an unavailable variant, preserve the shopper’s mandatory requirements, inspect every recommended destination, and verify that unavailable or mismatched alternatives are clearly identified. Record recovery engagement separately from completed purchases.
Which selections should survive a See Similar action?
Preserve mandatory requirements such as selected size when the implementation can enforce them. Show which optional preferences changed and let the shopper refine the alternatives.
Can a high See Similar CTR be compared directly with site conversion?
No. They usually have different event definitions, populations and windows. Report each denominator explicitly and use a suitable experiment to estimate the overall commercial effect.
What should a cross-surface fashion test record?
Capture the starting query and selections, source item/variant, alternative offered, destination state and basket outcome. Keep availability context so a later catalogue change does not obscure the original issue.
Related guides
- How to evaluate ecommerce site search: Evaluate ecommerce search with representative queries, calibrated relevance labels and clearly defined metrics. Validate commercial impact with a planned live test.
- Vector search vs keyword search for ecommerce: Compare vector, keyword and hybrid ecommerce search. Understand exact matches, semantic recall, filters and ranking, then evaluate them on your own catalogue.
- How to reduce zero-result searches: Diagnose zero-result searches using genuine query events, catalogue checks and constraint-aware recovery. Measure useful results without hiding failed searches.