A review sort decides which part of the customer record becomes the reading experience. “Most helpful” and “Most recent” answer different questions, even when they draw from the same reviews.
The choice is how to make that selection useful and understandable. Audit the evidence each view exposes, then test whether shoppers can find a suitable product.
🔎 What research says
Make the active sort clear, test topic grouping by category, and measure the onward shopping journey. A model’s preferred visibility assumption is not a tested ranking policy.
🗺️ In this guide
1. Make the sort explicit and measure its exposure
📈 Recommendation
Show shoppers which sort is active and let them switch. When comparing “Most helpful” with “Most recent”, record the reviews actually displayed and read rather than treating the entire database as equally visible.
For a cosmetics product, log the active sort, visible review IDs and expansions. Compare the issues surfaced by each view.
Audit recent defects and useful older accounts under both sorts. A ranking that looks successful can simply expose a different evidence set.
Use a randomized store test for a rollout decision. Measure product understanding, suitable orders and returns; do not select a sort from model correlations alone.
🎓 Findings
Alzate, Arce-Urriza and Cebollada (2021) modeled a nine-week panel of 119 blush products, using inverse sales rank as a sales proxy. Under helpful-ranked visibility, a 1% increase in review volume was associated with about a 0.064% increase in inverse rank; under recency-ranked visibility, the association was 0.134%. The same input had roughly twice the estimated association under a different exposure assumption.
These are small model associations with a ranking proxy, not gains caused by switching the sort. Some text-style associations even changed sign between models. The useful lesson is that the assumed visible reviews change the analysis; the study did not observe a randomized sort rollout or actual sales units.
🧠 Why it works
A review system stores more information than any shopper reads. Explicit sorts and exposure logs tell the operator which subset informed the choice, making a local comparison interpretable.
✋ Limitations
The study assumed visibility rather than tracking each shopper’s sort choice. Its two-step model standard errors may be too small, as the authors note. The volume coefficients do not establish a causal benefit of more reviews or identify a winning sort.
2. Test grouping by the product question
📈 Recommendation
Test grouping factual attribute reviews and overall experience accounts when an undifferentiated feed is hard to follow. Choose the order from the shopper’s task, then validate it by category.
For a printer, let shoppers start with setup, connectivity and print performance. For a fragrance, make overall use experience easy to explore. These are proposed applications.
Keep negative and positive evidence accessible within each group. Group by information type rather than quietly grouping complaints out of sight.
Compare comprehension and question completion, then paid outcomes. Check whether automatic grouping misclassifies mixed reviews.
🎓 Findings
Huang and colleagues (2014) compared grouped and randomly mixed attribute-based and experience-based reviews. Attribute-first displays were more helpful for products evaluated through specifications; experience-first displays worked better for products evaluated through use.
The outcomes were perceived helpfulness and ease of comprehension. Verified group means and confidence intervals were not recovered from the accessible primary preview. The result supports testing category-specific grouping, not a numerical sales prediction.
🧠 Why it works
Grouping lets a shopper stay with one kind of question before changing to another. The best starting point can depend on whether specifications or lived experience resolve the main uncertainty.
✋ Limitations
The exact store labels and illustrated layout were not tested. Reviews often combine attributes and overall impressions. Avoid treating every category as purely specification-led or experience-led, and retain access to the full feed.
3. Count useful onward search, not only focal conversion
📈 Recommendation
When testing a review sort, follow what shoppers do after reading a problem. A lower conversion rate for one item may accompany useful discovery of a better-fitting substitute.
For a printer defect, track visits to alternative printers and their reviews, then the eventual basket and returns.
Keep the original complaint accessible and route verified recurring faults to merchandising. Do not make a sorting experiment a complaint-suppression experiment.
Decide on the category’s suitable purchases and margin after returns, alongside customer understanding and access to current problems.
🎓 Findings
Varga and Albuquerque (2024) used retailer clickstream records and review movement under recency sorting. Negative reviews were linked to reduced focal purchases and increased visits to substitutes and their reviews. These effects differed by product type and whether the complaint concerned function, service or taste.
The primary abstract establishes the direction and boundaries; final fulltext estimates and uncertainty were not recovered. Earlier working-paper numbers differ from the later author explanation, so no numerical cross-product lift is used here.
🧠 Why it works
A shopper who learns an item is unsuitable may rationally continue elsewhere. An ordering decision should be judged across that journey instead of treating every departure from one product as a failure.
✋ Limitations
The evidence is a quasi-experiment with particular visibility assumptions, not a randomized comparison of all available sorts. Category-wide tracking and returns are our proposed evaluation. It does not follow that complaints should be buried to preserve focal conversion.
💡 The takeaway
Make review sorts visible, group information by task, and follow shoppers when credible problems send them to alternatives.
Instrument the two existing sorts in one category, inspect the review questions each exposes, and test task grouping only where it solves a reading problem. Follow substitute purchases and returns before choosing a default.
📚 Sources and study types
Alzate, M., Arce-Urriza, M., & Cebollada, J. (2021). Online Reviews and Product Sales: The Role of Review Visibility. Journal of Theoretical and Applied Electronic Commerce Research,16,638–669. DOI. Observational panel with assumed exposure models and sales-rank proxy,not a randomized sort test.
Huang, L., Tan, C.-H., Ke, W., & Wei, K.-K. (2014). Do we order product review information display? How? Information & Management,51(7),883–894. DOI. Controlled review-order experiment;perceived helpfulness/comprehension;fulltext extraction incomplete.
Varga, M., & Albuquerque, P. (2024;online2023). The Impact of Negative Reviews on Online Search and Purchase Decisions. Journal of Marketing Research,61(5),803–820. DOI. Retailer clickstream quasi-experiment;final primary-table magnitudes not used.
