How we weight reviews
Every score on this site shows the weighted figure, the raw average and the count. This page is the weighting. It is designed to lower scores as readily as it raises them.
What a review's weight is made of
Each published review gets a weight between roughly 0.2 and 3. The product's weighted score is the weight-averaged rating. Four factors:
| Factor | Why | How |
|---|---|---|
| Recency | Lifetime-deal quality genuinely changes. A review from two years ago describes a different product. | Exponential decay with a 270-day half-life from the review date. |
| Verified redemption click | Someone who clicked through to redeem has a basis for judgment. | ×1.5 when the click-through was recorded. Reviews without it still count: someone who could not redeem is exactly whose review matters. |
| Elapsed time since purchase | A review at day 180 says more about durability than one at day 3, and it dampens launch-week enthusiasm. | Multiplier from ×0.6 (same day) rising to ×1.6 at 180 days. |
| Helpfulness, normalised | Critical reviews attract fewer votes. A well-regarded critical review must not be buried. | Helpful votes are compared with the average for reviews of similar rating on the same product, not the overall average. Capped at ×1.5. |
What we do not do
- No downvotes. A downvote button on a critical review is a suppression mechanism.
- No weighting by reviewer history. Filtering by who tends to rate low is sentiment filtering by the back door.
- No gate on redemption. You can review a product you never managed to redeem.
- Credited reviews are badged, and the credit is the same for one star and five. Nothing is withheld for a negative.
Worked example
A product with three reviews: 5★ on day 2 with no redemption click (weight ≈ 0.73), 4★ on day 90 with redemption (≈ 2.55), 2★ on day 200 with redemption and above-average helpfulness (≈ 2.81). Raw average 3.67. Weighted: 3.20. The weighting lowered it, because the longest-owned, best-evidenced review was the critical one.
You can see every product where this happens on the benchmarks page.
Tester ratings are published too
Paid testers rate on a five-point scale and we publish their rating distribution. A tester whose ratings cluster at 5 is not using the scale. Reports name what was broken and whether it was fixed before launch.