The signals behind the badge
StarSkew is deliberately boring: four rules over numbers Amazon already prints on the page. There is no model you have to trust, so here is every threshold, exactly as the code applies it.
| Signal | Fires when | Risk |
|---|---|---|
| Very high rating at scale | Average of 4.8★ or higher across at least 200 reviews | +30 |
| Implausible perfect score | Average of 4.95★ or higher across at least 50 reviews | +30 |
| Review spike | Review count grew at least 5× compared with a reading from the last 14 days | +30 |
| Skewed distribution | More than 80% five-star with 5% or less at one and two stars combined | +20 |
How the points become a verdict
Points add up and cap at 100. One notable consequence: the distribution signal is worth 20 on its own, so a skewed histogram alone still shows green. It takes a second pattern to move a product out of green — one unusual number is not evidence.
Looks Genuine
0 – 25 points. No signal fired, or only the distribution signal did.
Mixed Signals
26 – 45 points. Exactly one of the 30-point signals fired.
Unusual Patterns
46 – 100 points. Two or more signals fired, or one plus the distribution skew.
Not enough reviews
Not scored. Fewer than 20 reviews. StarSkew declines to judge rather than guess.
Why these thresholds
Very high rating at scale. Holding 4.8 over hundreds of independent buyers is rarer than product pages make it look. Popular, genuinely well-liked products usually settle between 4.3 and 4.7.
Implausible perfect score. A displayed 5.0 means essentially nobody left a 3, 2, or 1. Across 50+ buyers that is close to statistically impossible without curation.
Review spike. Genuine reviews arrive roughly in proportion to sales. A fivefold jump in under two weeks usually means something other than organic demand.
Skewed distribution. Real review populations are J-shaped: mostly high, but with a visible tail of unhappy buyers. A missing tail is the pattern that survives averaging.
The boundaries were tuned against a hand-checked set of real listings, and the goal was to keep well-known genuine products clear of the flags. AirPods 4, for instance, sit at 82% five-star with 6% low reviews — just outside the skew rule. That is deliberate: a false red on an obviously legitimate product costs more trust than a missed fake earns.
What no amount of metadata can tell you
StarSkew never reads review text. That rules out whole categories of manipulation: AI-written reviews, copy-pasted reviews, farmed reviewer accounts, and hijacked listings where old reviews stay attached to a rewritten product page. Slow, patient incentivized reviewing produces an organic-looking curve and passes cleanly.
So a green badge means “none of these four rules fired” and nothing more. A red badge means the pattern is unusual, not that anyone did anything wrong. For the parts a tool cannot do, see how to check reviews yourself.
Where the numbers come from
The extension reads the average, the review count, and the star histogram from the product page you are already looking at, then asks the backend for a verdict. Review counts observed by different users are compared to each other, which is how the spike signal gets its history. Nothing else about your browsing is involved — see the privacy policy.
Free, no account, works on 10 Amazon marketplaces.