September 15, 2026
7 min read
Search ranking: the short answer
Ecommerce search ranking rules give teams direct control over which qualifying products appear first. Machine learning ranking finds patterns in shopper and product data to make those decisions at scale. Most stores benefit from a combined approach: rules protect exact intent, stock, and campaign needs, while learned signals help sort large sets of relevant products.
Rules and machine learning solve different problems
| Method | Strength | Tradeoff |
|---|---|---|
| Manual rules | Clear control over brands, attributes, stock, and campaigns | Takes ongoing review and can miss new patterns |
| Machine learning | Learns from clicks, carts, orders, and product behavior | Needs enough clean data and can be harder to explain |
| Combined ranking | Balances control with scale | Requires testing and sensible guardrails |
Use rules for hard requirements
Rules are useful when a result must satisfy a condition. Keep out-of-stock products down. Protect exact SKU and brand matches. Give a seasonal collection a temporary boost. Respect category-specific needs such as compatibility, size, or country availability.
Write rules around shopper value, not internal preference. A margin-first rule that hides better matches can weaken trust and reduce sales.
Use learned signals for ordering relevant products
Once a result set is relevant, learned ranking can sort products using observed behavior. Signals may include clicks, add-to-cart events, orders, popularity, freshness, and product similarity. The model should learn from enough traffic to avoid treating a small sample as a permanent preference.
New products and seasonal items need care. They have less history, so a pure behavior model may keep showing older products. Use freshness, catalog coverage, and merchandising controls to give new items a fair test.
How to choose an approach
Choose rule-led ranking when your catalog is small, product data is uneven, or teams need immediate control. Add machine learning when the catalog and traffic create enough data to learn from many queries and categories.
Before changing ranking, separate relevance from business priority. First ask whether the product satisfies the query. Then decide how stock, popularity, margin, or campaigns should order the products that qualify.
Clerk.io’s Intelligent Search includes sales-based ranking and tools for synonyms, facets, suggestions, and search analytics. Pair ranking changes with a query test set and a clear rollback rule.
A measurement plan that keeps ranking honest
Track search conversion rate, add-to-cart rate, revenue per search, and exits. Break results down by query class, category, device, and new versus returning shopper. Review high-revenue queries manually each week during a ranking test.
If clicks rise while orders fall, the ranking may be attracting attention without meeting intent. If revenue per search rises but zero-result rate worsens, check whether a small set of valuable queries is masking a broader problem.
Customer results provide useful context. Uniq Perler reported 20% higher average order value and 26% larger baskets using Search and Recommendations. That is a reported customer outcome, not a forecast for every ranking project.
Free ebook: Clerk.io’s Search covers search improvements that can support conversion and average order value.
Start a free Clerk.io trial or book a demo to assess ranking options for your catalog.
TL;DR
Rules provide control. Machine learning provides scale. Use rules for exact intent and hard constraints, learned signals for ordering relevant products, and measure both with revenue per search and search conversion.
