Reviews, Ratings and Their Effect

Reviews are the most discussed lever in marketplace selling and the relationship to rank is more indirect than the discussion suggests.

The actual mechanism

Reviews do not enter the sales rank calculation. Sales rank is sales.

Reviews affect conversion. A shopper who arrives at a listing is more likely to buy with substantial positive social proof.

Conversion affects search ranking, which affects traffic, which affects sales, which affects sales rank.

So the chain is real and it has three links, each of which can break. This is why "get more reviews" produces inconsistent results.

What the evidence supports

Count matters more than rating, up to a point. Moving from zero reviews to a modest number produces a large conversion effect. Moving from many to more produces little.

Rating matters within a band. Below a threshold — commonly cited around four stars — conversion suffers noticeably. Above it, differences are smaller than sellers assume, and a perfect score can read as implausible.

Recency matters. A listing whose reviews stopped two years ago reads as stale.

Review content matters for specific concerns. In categories where fit, durability or compatibility are the buying question, the text does work the star rating does not.

These are general patterns and category-dependent. In some categories reviews are decisive; in commodity purchases they are close to irrelevant next to price and delivery.

Measuring the effect

Record review count and average rating with every rank observation. This is one of the context fields that makes a dataset answer questions later.

Look for step changes. A listing that gains a large number of reviews in a short period and shows a rank improvement afterwards is suggestive.

Beware the reverse causation, which is the dominant effect: sales produce reviews. A product selling well accumulates reviews, so review count and rank correlate strongly without reviews causing anything.

This is the single most common analytical error in this area. Almost every observed correlation between reviews and rank is sales causing both.

Establishing causation requires an intervention — a review campaign with a control, or a natural experiment such as a review purge — and observational data cannot do it.

Review velocity as a signal

More useful for competitive intelligence than review count.

Reviews per week is a proxy for sales, with a conversion rate from purchase to review that is low, category-dependent and reasonably stable for a given product.

A competitor's review velocity tracks their sales velocity, which makes it a second observable alongside rank and a useful cross-check.

A sudden change in review velocity without a corresponding rank change is worth investigating, and frequently indicates review solicitation rather than a sales change.

Review velocity is less noisy than rank for slow products, because it is a count rather than a position, which makes it the better measure in the long tail.

Manipulation and what it looks like

Review manipulation is against every marketplace's terms and is widespread enough to affect analysis.

Patterns worth noticing: a burst of reviews in a short window, reviews clustered at the extremes with nothing in between, review text of similar length and structure, reviewers with narrow histories, and a review count inconsistent with the rank.

That last one is the strongest signal available from outside: a product with a thousand reviews and a rank implying very low sales has an inconsistency that needs explaining.

For analysis, the implication is to weight review data lower than rank, and to treat a review-count outlier as a flag for investigation rather than a finding.

For sellers, the implication is that purchasing reviews carries real enforcement risk — listing suppression and account action — and is covered separately.

The practical position

Track review count and rating as context, not as a driver.

Use review velocity as a secondary sales proxy, particularly for slow products where rank is noisy.

Do not claim a causal effect of reviews on rank from observational data. The correlation is real and it is mostly sales causing both.

Treat a large mismatch between review count and rank as a question about the listing rather than as data.

Using review velocity as a proxy

For slow products where rank is too noisy, review accumulation is the steadier measure.

Count reviews per week rather than total. The total is cumulative and reflects history; the rate reflects current sales.

The conversion from purchases to reviews is low and varies by category, but is reasonably stable for a given product over time, which is what makes the rate usable as a relative measure.

Compare rates between products in the same category, not across categories.

Watch for solicitation. A sudden rate change with no rank change usually means the seller started asking, not that sales moved.

Combine with rank. Where both move together, the signal is strong. Where they diverge, one of them has an explanation worth finding — a stock-out, a review purge, a solicitation campaign, or manipulation.

Record the count at every observation, which costs nothing and provides this whole capability retrospectively.