How ReviewRank Scores work

The ReviewRank Score is a weighted composite of four signals. Each is scored on a 0–100 scale before weighting.

Evidence blending. Because sentiment and consistency come from a sample of at most 5 reviews, they are blended toward priors in proportion to how much of the business the sample actually covers: weight = min(0.5, √(sampleSize / totalReviews)). The sentiment prior is the overall rating mapped to the same 0–100 scale; the consistency prior is 75. For a 412-review business the sample weight is ≈ 0.11, so the sampled signals contribute lightly. The 0.5 cap (v1.2) means that even for a business whose sample covers every review it has, the relevance-selected sample never outweighs the priors — this bounds how far one unlucky sampled review can ever move a score.

The weighted composite is clamped to 0–100 and rounded to one decimal. (A reserved fraud-signal multiplier exists in the code and is currently fixed at 1.0 — it does not affect any score today.) Component scores shown on business pages are the post-blending values, individually rounded to one decimal; the final score is computed from unrounded values, so recomputing from displayed components can differ by up to ±0.1.

A landscaping company with a 4.7★ rating, 412 reviews, and sampled review ratings of 5, 5, 4, 5, 5. Every number below comes from running the production scorer on these inputs when this page is built:

A nearby competitor with 4.9★ but only 18 reviews scores 72.5 (Highly Trusted) on the same formula. The Bayesian shrinkage and the volume component are doing the work: we need evidence before we declare someone the best.

The score maps to a band label shown with the number:

Scores under 35 are labelled Limited Reputation. Businesses under 4.0★ are not displayed in ranked results at all, regardless of other signals.

Separately, internal evidence tiers apply additional gates on top of the score (e.g. the top tier requires score ≥ 65, rating ≥ 4.5, and ≥ 150 reviews). These tiers no longer render as a badge — the displayed classification is the score band plus the Confidence level, which communicate the same information without a third overlapping label — but they still gate conservative features such as the quote button and the narrative language.

• ·Ratings and review counts come from the Google Places API.
• ·Google's API returns at most 5 reviews per business, selected and ordered by RELEVANCE as determined by Google. It offers no way to request the most recent reviews, so sentiment and consistency are computed from this relevance-ranked sample — treat them as a spot-check, not a trend.
• ·For hotels, restaurants, and attractions we also display Tripadvisor data (rating, review count, ranking, awards) in a supplemental panel. Tripadvisor data does NOT change the ReviewRank Score — it is shown side-by-side so you can compare platforms yourself.
• ·The "review sample above/below average" indicator compares the sampled reviews against the long-run rating. It is a proxy, not a measured trend — we do not yet have historical data.

Next to every score we show a Confidence level — how much review evidence backs that score. It never changes the score itself; it tells you how seriously to take it. Thresholds come from the review-count distribution of businesses we actually evaluate:

• ·Very High — 300+ reviews and a full 5-review sample.
• ·High — 100 to 299 reviews.
• ·Moderate — 25 to 99 reviews.
• ·Low — fewer than 25 reviews. A 90-point score here is a promising early signal, not an established fact.

Rating-stability inputs will join this indicator once we have our own longitudinal data.

Honest limitations. Any reputation score is only as good as the data it sees. Here is what the current ReviewRank Score does not incorporate:

• ·Full review history. We see the aggregate rating, the review count, and a 5-review relevance-ranked sample — nothing more.
• ·True recency. Because the sample is relevance-ranked, nothing in the score is genuinely "recent". Historical tracking is planned; until it exists we avoid recency claims.
• ·Owner response rate. Planned once we build our own review-tracking backend.
• ·Cross-platform scoring (Yelp, BBB, Facebook, Nextdoor). Tripadvisor data is displayed for hospitality businesses but does not alter the score.
• ·Review velocity. Whether a business earned 400 reviews in 10 years or in 6 months looks identical today.
• ·Individual reviewer credibility. We treat every public reviewer's contribution equally.
• ·Paid or sponsored signals. We reject them by design — they would compromise the score.

Google and Yelp rank businesses primarily on raw rating plus proximity, with paid placements mixed in. A 5.0★ business with 2 reviews can outrank a 4.7★ business with 400 reviews on those platforms.

The ReviewRank Score is designed to answer a different question: "If I walked in today, how likely is this to be a good experience?" That requires evidence. The Bayesian prior and volume signal exist to prevent small-sample businesses from outranking well-established competitors purely on statistical luck.

We are not a discovery engine. We are a trust layer on top of the data the discovery engines already show you.