METHODOLOGY
How we measure trust.
No black box. No paid placements. Every formula is public — and calculated live.
The formula
In simple terms: the more backed-up, recent reviews a provider has, the higher the score. A prior of 10 neutral reviews prevents a single 5-star review from immediately producing a perfect score — the score has to be earned.
Display convention:
The Bayesian average produces a value between 0 and 5 (star scale). For better readability, this value is displayed on a 0–10 scale:
Trust Score = Bayesian result × 2Example: Bayesian avg. 4.2 → Trust Score 8.4 / 10
The 10-step pipeline
Every review automatically goes through:
- ✓1. Auth & email verification
- ✓2. Daily user limit (max. 5/day)
- ✓3. Daily provider limit (max. 10/day) + Z-score
- ✓4. Anti-cluster (IP, time, stylometry = automated writing-pattern analysis, semantics)
- ✓5. Sentiment consistency (rating vs. text)
- ✓6. Incentive language (53 words DE/FR/EN)
- ✓7. OCR receipt recognition (Tesseract.js, Swiss server)
- ✓8. Merchant matching (OCR vs. provider)
- ✓9. Image-hash duplicate detection (pHash)
- ✓10. IP hash + user ID enforced server-side
Proof weighting
| Proof type | Weight | Meaning |
|---|---|---|
| Purchase receipt (high OCR) | 1.0 | Date, merchant, and amount verified |
| QR code (partner) | 1.0 | Issued directly by the provider |
| API-confirmed | 1.0 | Shop system confirmed the purchase |
Time-factor curve
The half-life of ~138 days (~4.5 months) is a deliberately chosen starting value. It will be calibrated once enough real review data is available — that's part of being transparent.
What Verix doesn't do
✗ Paid visibility for businesses
✗ "Featured reviews" or bought placements
✗ Automatically deleting inconvenient criticism
✗ Rewards or discounts for reviews
✗ Manual score interventions
✗ Sharing user data with third parties
Comparison
Full criteria list. A more compact selection of the key points is also shown on the homepage. Google/Trustpilot figures reflect our own research (as of July 2026), not their own statements.
| Feature | Trustpilot | Verix | |
|---|---|---|---|
| Mandatory purchase receipt | ✗ | ✗ | ✓ Always required — reviews without proof are rejected, not just down-weighted |
| Incentive detection | ✗ | ✗ Rarely | ✓ Automatic NLP |
| Anti-cluster / spike detection | ✗ Reactive | partial | ✓ Real-time alert |
| revFADP compliance | ✗ GDPR | ✗ GDPR | ✓ Native revFADP compliance |
| 100% Swiss hosting | ✗ US / EU | ✗ US / EU | ✓ 100% Switzerland |
| Transparent Trust Score | ✗ Algorithm secret | ✗ Algorithm secret | ✓ Formula public |
| OCR receipt extraction | ✗ | ✗ | ✓ |
| Bayesian ranking | ✗ | ✗ | ✓ |
| Time weighting | partial | ✗ | ✓ |
| Reputation system | ✗ | partial | ✓ |