Why two 5-star reviews don't make a 5.0
A 5.0 rating from two reviews tells you almost nothing. On Vouch Rate it can't happen. Here is the full method, every number included, because a rating you can't check is just a claim.
A 5.0 rating from two reviews tells you almost nothing. One more review could move it a whole star. On Vouch Rate, it can't happen.
Two 5-star reviews show 3.8. Twenty purchase-verified 5-star reviews show 4.6. A perfect 5.0 takes roughly 220 purchase-verified 5-star reviews.
This post is the full method, every number included, because a rating you can't check is just a claim.
The formula
rating = ( Σ stars × weight + 3.7 × 8 ) ÷ ( Σ weight + 8 )
weight = verification × recency × authenticityEvery review has a weight. The rating is a weighted average of the stars, blended with a starting point. The rest of this post explains each part.
1. Every business starts at 3.7, as if it had 8 average reviews
The 3.7 × 8 in the formula is a starting point. Before a business has any reviews, it is treated as though it already had eight reviews averaging 3.7. Real reviews gradually outweigh that starting point, and once a business has plenty of them it barely matters.
This is what stops a small number of reviews from producing an extreme number:
- One glowing review can't make a business. A single purchase-verified 5-star review shows 3.8.
- One angry review can't sink it. A single 1-star review shows about 3.5.
- Two 5-star reviews from email-confirmed reviewers show 3.8, not 5.0: (2 × 5 × 0.48 + 29.6) ÷ (0.96 + 8) = 3.84.
- Twenty purchase-verified 5-star reviews show 4.6.
- A displayed 5.0 needs the underlying figure to reach 4.95, which takes roughly 220 purchase-verified 5-star reviews.
Statisticians call this a Bayesian average, or shrinkage. The idea is simple: the less evidence there is, the less a rating should move away from what's typical.
2. Verification: how sure we are the dealing happened
Each review's weight starts with how confident we are that the reviewer actually dealt with the business:
- Anyone wrote it: ×0.60
- Email confirmed: ×0.80
- Dealing confirmed, from the business's records or proof a moderator checked: ×1.00
- Specific purchase confirmed, by a connected sales system or a receipt a moderator checked: ×1.15
Two higher levels, identity checks (×1.20) and several independent signals agreeing (×1.25), are published but not in use yet.
The gaps are small on purpose. Unverified reviews still count. Plenty of genuine customers paid cash or lost the receipt, and a review site that drops unverified reviews ends up burying genuine complaints. Verification makes a review count for more; its absence doesn't make a review disappear.
You can see exactly what each level means on our verification page.
3. Recency: a review's weight halves every 18 months
A review's weight halves every 18 months, and never drops below 35% of what it started at:
- a review written today counts in full;
- a review from 18 months ago counts half;
- from about 27 months onwards, a review stays at 35%.
That does two jobs. A business that has improved isn't stuck with its past, because recent reviews carry more weight. And a business that has got worse can't bury its history, because old reviews never fade to nothing. Every profile also shows its 12-month and 30-day averages separately, so you can see the direction of travel for yourself.
4. Authenticity: every review is checked for signs it isn't genuine
Every review gets an authenticity score out of 100 from automated checks. Signs that lower it include:
- brand-new accounts;
- bursts of reviews arriving together;
- several accounts on the same network;
- near-identical text;
- throwaway email addresses.
No single sign decides anything, because each has an innocent explanation. A shared office network is not fraud. The score turns into a multiplier between 0.25 and 1.
Reviews that score low don't publish straight away. They wait for a human moderator, and they don't count towards the rating until a person has approved them. Reviews removed for breaking our guidelines are left out of the calculation entirely.
Putting it together: what the chart shows
The chart above plots the displayed rating against the number of 5-star reviews, for three kinds of new review. Each curve uses a typical authenticity score for a new review of that kind, which gives these effective weights:
- Unverified: about ×0.33
- Email-confirmed: about ×0.48
- Purchase-verified: about ×0.91
The better evidenced the reviews, the faster a business earns a high rating. But no amount of evidence lets a handful of reviews produce a perfect score.
What never goes into the rating
- ✕ The plan a business pays for
- ✕ Whether it has claimed its profile
- ✕ Anything a business can switch on, switch off or buy
The rating is calculated from reviews and nothing else.
Why publish all of this?
Because every other way of doing it asks you to take the number on trust. Every profile shows the weighted rating and, next to it, the plain unweighted average, so you can always see what the simple arithmetic says. If the two differ a lot, the method above explains why.
The full methodology, with every constant, is on our rating page. If you think a number on it is wrong, tell us.