Most review sites display one number: the arithmetic mean of every star rating ever submitted. It feels honest. It isn't very useful.
A brand-new shop with three glowing reviews from friends will out-rank a 12-year-old retailer with 3,000 reviews averaging 4.6 — even though only one of them is statistically trustworthy. Worse, a single bad review can swing a young profile by a full star, while having no visible effect on the established one. The number tells you very little about what's actually going on.
What TrustScore measures
TrustScore is a 0–100 composite that bakes three things into one signal:
- Bayesian average. We pull every rating toward the global mean by a virtual "prior" of 10 reviews. New profiles can't pop to 100; established profiles aren't moved by a single outlier.
- Recency. Reviews from the last 6 months count more than ones from 2019. A company that was great then but ignored for years should not coast on old reputation.
- Verified-purchase weight. Verified reviews count for 1.5× of an anonymous one. They're harder to fake, so they should move the needle more.
The result is a number that goes up with sustained quality and goes down when neglect, attrition or a wave of complaints arrives. It's recalculated nightly across the entire catalog.
Reading a TrustScore
For a quick mental model:
- 85–100. Excellent. Hundreds of recent, mostly verified reviews trending positive.
- 70–84. Good. A solid track record, occasional issues.
- 55–69. Mixed. Read the reviews — both ends of the curve.
- Below 55. Caution. Sustained problems or very thin signal.
You can still see the raw star average everywhere we show TrustScore — they're complementary, not a replacement.