When should your store show popularity cues?

Test popularity information where it helps a shopper interpret a specialist product. Do not assume that the largest storewide seller is the most informative option for every customer.

Tucker and Zhang (2011) changed a wedding directory: 69 bridal shops showed previous click counts and were ranked by popularity; 51 florists retained alphabetical listings without counts. They compared two months before and two months after the change. These were vendor records, not a visitor-randomized purchase test.

For vendors with above-average prior popularity, the estimated extra response to popularity information was 43.93 more clicks for narrow-appeal than broad-appeal vendors over the two-month test. “Above average” here means one standard deviation above that category’s previous click average: a statistical step used to compare vendors with similar established popularity. Narrow appeal meant a smaller-town location, rather than a specialist ecommerce category.

The practical point is the difference in response: roughly 44 additional listing clicks, not 44 orders, for the niche side of this comparison. The difference was statistically clear against the florist control; it was less precisely estimated against caterers that were already ranked by popularity without showing counts.

A popular specialist can be surprising in a way that a popular generalist is not. The authors’ explanation is that customers infer more about quality when a narrow-appeal provider attracts the same attention as a broad-appeal provider. That supports testing relevant popularity context; it does not prove the best wording for a badge.

This was one wedding directory in 2006. Location defined appeal, clicks measured interest, and categories received different treatments. The florist comparison changes both ranking and visible counts; the caterer comparison helps separate those influences. A category-specific sales badge is a transfer to test, not the intervention the researchers ran.

If your store has reliable sales data, test a clearly defined hit list in a specific category before applying it across the catalogue. Review price bands separately.

Yoo, Jeon and Han (2016) studied an apparel seller on Tmall. Their main field experiment covered 290 products. They compared sales changes around selected products appearing on a hit list, using comparison products and matching to address pre-existing differences.

The paper’s abstract reports an estimated average increase of 1.3 units sold per day when a product appeared on the hit list. This is an absolute daily product-sales estimate, not a 1.3% conversion improvement. It also reports a stronger response in one niche subgroup and among middle-price products.

The abstract does not provide the starting daily sales, confidence interval or detailed subgroup estimates. The result shows why a category test is worthwhile; it cannot supply a sales target or be multiplied across your catalogue as a forecast.

A hit list changes which products shoppers notice and gives them information about other customers’ behavior. A middle-price item may be easier to choose than either extreme, but the reported subgroup pattern does not establish that explanation or a universal ideal price band.

This is one apparel seller. The full methods and tables were not accessible for this draft, so the reported average remains an abstract-level estimate with unresolved precision and exact treatment details. Use it to motivate a conditional test, not to promise the same increase in another store.

Keep “customers bought this” separate from “customers liked this.” Do not treat a low purchase share as equivalent to a poor review, or a high share as proof of satisfaction.

Chen, Wang and Xie (2011) studied Amazon’s removal of purchase-percentage information while reviews remained available. Their first comparison followed 90 camera models. Of these, 26 previously had a positive purchase signal, 48 a negative signal, and 16 had no purchase signal and formed the control group.

After removal, the average change in log sales rank was 1.449 for the positive-signal group versus 0.849 for controls. An arithmetic restatement makes that scale easier to understand: the geometric-average rank multiplier was about 4.3 versus 2.3. A larger rank number is a worse position. These are movements in a sales-ranking proxy, not quantities sold or a conversion rate.

The positive-signal group’s deterioration was statistically clearer than the control group’s. For cameras losing a negative purchase signal, the corresponding difference from controls was not statistically clear. The result does not justify treating a weak purchase signal as the same warning as a negative customer review.

A purchase count tells shoppers what others selected; a review tells them what someone experienced or thought. The same paper found different patterns for purchase signals and review content. Treating both as one “social proof” score hides that distinction.

Amazon chose the policy change; the researchers did not randomize shoppers. Controls and statistical adjustments help, but they do not remove every possible difference between products. Camera sales rank is an indirect measure. The recovered authors’ postprint supports these values; a full comparison with the final publisher version remains open.

For shoppers trying to avoid a costly mistake, test whether popularity cues increase uncertainty after choosing. Keep product suitability easy to evaluate.

Ghiassaleh, Kocher and Czellar (2020) report that popularity signs can reduce commitment to a chosen option among prevention-focused consumers: people concentrating on avoiding a mistake. Their explanation links a more varied set of considered options to greater uncertainty.

The accessible abstract reports the direction of this adverse effect, including work with real choices and field data. It does not expose a usable effect size, sample denominator or uncertainty estimate. It cannot tell a store how many purchases, cancellations or returns will change.

A badge can introduce an option the shopper would otherwise ignore. If that option differs from what they need, choosing may become harder to defend even after a selection. Use popularity as one piece of information beside suitability, and test the resulting decision.

The result depends on the consumer’s orientation, not simply on whether the product is expensive. Do not infer a shopper’s psychology from a single click or assume every cautious customer reacts this way. The detailed study measures and estimates remain unverified, so this recommendation is a conditional guardrail.