How the Trust Score works
Last updated: November 10, 2026
The Trust Score is a number from 0 to 100, computed by a rules-based scoring engine from the evidence each review cites. The same evidence always produces the same score. No AI model generates it, we don’t test products hands-on, and we don’t use Amazon customer reviews or star ratings.
1. What evidence counts
Each piece of evidence is recorded with its source, its type and a confidence level. The type decides how much a number can move a score. Qualitative statements and reader feedback never become a score on their own; they add context and can mark a factor as disputed.
| Evidence type | Weight as a numeric input |
|---|---|
| Objective measurement (for example, a measured battery runtime) | 100% |
| Reviewer score (for example, 8.7/10 from an independent review) | 75% |
| Manufacturer specification (for example, rated battery life) | 40% |
| Qualitative review statement (no number) | Not used as a number (context only) |
| Owner or reader feedback (for example, a reader poll) | Not used as a number (context only) |
That weight is multiplied by the source’s confidence: high 100%, medium 60%, low 30%.
2. From evidence to factor scores
- Ratings such as 8.7/10 or 4.8/5 are converted to 0–100.
- Measured or rated runtimes are converted with a diminishing-returns curve, 100 × (1 − e^(−hours ÷ reference)), where the reference depends on the product category.
- A factor’s score is the weighted average of its numeric inputs. A factor with no numeric evidence is marked “Not assessed” and left out — never guessed.
- Each factor is labelled corroborated (two or more sources), single-source, or disputed (sources disagree). A disputed factor keeps its best-evidence score but its confidence drops one level.
3. Factor weights by category
The Trust Score is the weighted average of the assessed factors. Factors that aren’t assessed are excluded and the remaining weights are scaled up, so a missing factor neither helps nor hurts.
| Factor | Weight | Notes |
|---|---|---|
| Noise cancellation | 20% | |
| Sound quality | 20% | |
| Comfort | 15% | |
| Battery | 15% | Runtime reference: 17 hours |
| Call quality | 10% | |
| Features | 10% | |
| Value | 10% |
4. Evidence confidence
Confidence is reported next to the score, not mixed into it. “Coverage” is the share of the category’s weight that is assessed; “backing” is the share of assessed weight supported at high or medium confidence.
- High: coverage at least 80% and backing at least 70%.
- Medium: coverage at least 60% and backing at least 40%.
- Limited: anything below that.
5. From score to verdict
| Trust Score | Verdict |
|---|---|
| 85–100 | Buy with confidence |
| 70–84 | Buy with caution |
| 50–69 | Consider alternatives |
| 0–49 | Not recommended |
A review is published to search engines only when it has a score, at least 60% coverage and cited sources.
6. What we don’t do, and our limits
- We don’t test products hands-on. Reviews synthesize the tests, measurements and specifications they cite.
- No AI model produces the Trust Score or the verdict. Both come from the fixed rules above.
- We don’t use Amazon customer reviews, star ratings or prices.
- Scores reflect the sources as of each review’s “Last updated” date. New firmware, price changes or newer tests may not be reflected yet.
- Weights are editorial choices about what matters most in a category. Your priorities may differ, so every review shows the factor scores, not just the total.
7. Independence and corrections
Affiliate commissions are not an input to any score (see our affiliate disclosure). Reviews are maintained by Vinayak Madar (about). Spotted an error or a newer source? Email hello@shopnexusapp.com.