Amazon starts weighting product ratings to favor reviews that are newer, helpful, and verified
Amazon updates customer reviews with new machine-learning platform — The e-commerce giant created a new in-house model to show off more up-to-date and helpful reviews for buyers.
Context & Ripple Effects
This 2015 update is the opening move in a decade-long arc of Amazon rebuilding its review system around machine curation rather than raw crowd averages. The through-line is visible across its own coverage: after this ML re-weighting, Amazon purged over 500K reviews — mostly 'incentivized' ones — under tightened guidelines (the 500K+ review purge), then quietly added one-tap ratings in 2019 to widen the feedback funnel.
By 2023 the same pipeline culminated in generative AI summaries of what customers like and dislike, rolled out to US mobile users (across a broad range of products) with an explicit 'AI-generated' disclaimer. The 2015 weighting matters because it established that Amazon's model, not the star count, decides which customer signal a shopper sees.
First-order effects
- Shoppers immediately see a different default ordering: newer, helpful-voted, and verified-purchase reviews surface ahead of older ones, so products whose quality has drifted up or down get ratings that track current experience rather than legacy averages.
- Sellers who banked on an accumulated stock of early five-star reviews lose that moat overnight; maintaining ratings now requires a continuing stream of genuine verified purchases.
Second-order effects
- Demoting stale or gamed reviews squeezes the paid-review economy — a squeeze Amazon made explicit a year later when its guideline change led to the mass deletion of incentivized reviews.
- Because the weighting rewards volume of recent feedback, it created the incentive structure behind later funnel-widening moves like one-tap ratings, which lower the effort bar for submitting them.
Third-order effects
- If the pattern holds, product discovery on the platform shifts from shoppers reading individual reviews to trusting Amazon's models — weighted rankings, then AI-generated summaries — as the authoritative distillation, concentrating interpretive power over merchant reputation in Amazon's in-house systems.
The trend: Amazon has spent the decade replacing raw crowd-sourced review signals with machine-curated trust layers — weighted rankings, frictionless inputs, and finally AI summaries — making its own models the arbiter of product quality.