Methodology version v3-bayesian-geo-weighted

How GEO Company Rankings works

Rankings are calculated from published data using the rules below. Companies cannot buy a position, choose a position, pay to enter a city, or improve their score merely by claiming a profile.

Ranking positions are calculated using a multi-factor scoring system. The GEO Ranking Score is not a company's average star rating. The current implementation considers qualifying verified-review volume, overall customer ratings, structured GEO experience dimensions when available, and business-verification status. Category and local rankings additionally restrict which reviews qualify by category association or geographic eligibility.

A company with a higher average customer rating will not necessarily rank above a company with a larger body of consistently strong qualifying reviews. Review quantity alone does not guarantee a high position, and a perfect average rating alone does not guarantee the top position.

GEO Company Rankings evaluates the total body of qualifying evidence rather than relying on one metric. Rankings reflect the implemented combination of customer ratings, structured experience quality, qualifying volume, business verification, scope relevance, and the eligibility rules documented below.

1. How companies and reviews enter the database

Company records may be submitted by a company, added through editorial research, or imported from documented public sources. A listing is not a ranking. Claiming a profile permits corrections and responses but does not create local eligibility.

Customers submit reviews through the review form. Reviews pass moderation before publication. GEO Index may check the reviewer's business website, project website, visible implementation, or supporting project information. See the Review Standards and Verification Policy.

2. What “Verified Review” means

A verified review is a review that GEO Index manually vetted. We review the submission and check the reviewer's business website, project website, publicly visible implementation, or other available evidence to determine whether the work described appears to have been performed.

Verification means we found reasonable evidence of a genuine business engagement and the described work. It does not mean every revenue, traffic, ranking, conversion, or visibility result was independently audited. Results remain reviewer-reported unless specifically labeled “Supporting result evidence reviewed.”

3. The current Bayesian score

The application uses v3-bayesian-geo-weighted. Its weighted-rating equation is:

S = (v / (v + m)) × R + (m / (v + m)) × C

S: ranking score before any documented business-verification bonus
v: qualifying verified-review count used by that ranking
m: prior weight, currently 20
C: prior mean, currently 3.5 out of 5
R: review-quality signal. When both inputs exist, it is 35% overall stars and 65% weighted GEO review dimensions.

Bayesian weighting prevents one perfect review from automatically outranking a company with many consistently strong reviews. Review volume gradually moves the score away from the 3.5 prior and toward the company's observed quality.

Scores may contain several decimal places because the application combines weighted dimension averages with Bayesian fractions and then rounds the result to four decimal places. The displayed value is always labeled GEO Ranking Score; customer stars are displayed separately as Average Customer Rating.

Public ranking pools require qualifying verified reviews. Editorial research can create or enrich a company profile, but research-only records do not enter a ranking snapshot. Business verification adds 0.05 to the score; merely claiming a profile does not. This business-verification signal is separate from review verification and is disclosed in ranking evidence.

4. Score breakdown: factors currently implemented

Qualifying review evidence and volume: only reviews eligible for the ranking scope enter its count. Volume controls how strongly observed review quality moves away from the 3.5 prior.
Overall customer rating: the mean overall star rating is 35% of the review-quality signal when structured GEO dimensions are also available.
Structured GEO experience quality: 65% of the quality signal when available. Its implemented dimensions are:
- AI visibility lift: 30% of the structured GEO-dimension score
- Citation & mention quality: 22% of the structured GEO-dimension score
- GEO strategy quality: 18% of the structured GEO-dimension score
- Execution & delivery: 14% of the structured GEO-dimension score
- Communication: 10% of the structured GEO-dimension score
- Value for investment: 6% of the structured GEO-dimension score
Missing dimensions are omitted and the remaining dimension weights are normalized. The current national and live city/state pipelines use this signal. Current service/platform/industry snapshots use their scoped Average Customer Rating instead; they do not borrow company-wide GEO dimensions from unrelated reviews.
Business verification: a verified business record adds 0.05 to the final score. This is separate from review verification and profile claiming.
Category relevance: service, platform, and industry pools use only verified reviews explicitly associated with that category. This determines eligibility; it is not an undisclosed percentage multiplier.
Geographic eligibility: local pools use only verified review locations qualifying for that market. This determines which reviews enter the calculation; it is not a separate score bonus.

5. How reported issues currently affect rankings

The current scoring code does not run a hidden complaint classifier over review text. Reported concerns affect a ranking only through the review's numeric overall rating and structured experience-dimension scores after the review qualifies and is verified.

An isolated low score affects the aggregate as one qualifying review; Bayesian weighting limits the ability of any single review to determine the position.
Repeated concerns can have more aggregate effect because multiple qualifying reviews contribute multiple numeric scores. There is no separate repeat-complaint multiplier.
Serious and minor issue text do not currently receive different numeric weights.
Resolution status and company responses do not currently change the score.
Unverified reviews and removed reviews are excluded from ranking pools.
Older reviews do not currently receive a separate time-decay adjustment.
Positive evidence offsets lower numeric feedback through the same weighted averages and Bayesian calculation; there is no manual offset.

Administrator clarification is still required before issue severity, resolution, response quality, consistency analysis, or recency can become independent score factors. Those factors must not be advertised as implemented until a documented algorithm and auditable data model are approved.

6. Ranking scopes

National

National rankings score eligible published companies using verified-review totals and review-quality data. Unverified and removed reviews do not contribute. Removed reviews are excluded after ranking snapshots are recalculated.

Current stored snapshot jobs also require the company record to be marked indexable, while live city/state queries require publication but do not use that SEO flag. This is an eligibility-rule inconsistency requiring administrator clarification; indexability is not a numeric score weight.

State and city

State rankings use qualifying verified reviews associated with that state. City rankings use qualifying verified review locations within 300 miles of the city's stored geographic reference point. Office addresses and service-area claims do not create local rank. Read the complete local methodology.

Service, AI platform, and industry

A scoped ranking uses only verified reviews explicitly associated with that service, platform, or industry. A general review is not substituted as support for an unrelated category.

7. Ties, recalculation, and history

Scores are sorted from highest to lowest. Local ties are resolved by qualifying review count and then company name. Stored ranking snapshots use score order and a stable company identifier as the final deterministic tie-break. Every published ranking should show its actual snapshot timestamp and methodology version rather than implying a fixed schedule that is not configured.

Recalculation creates a new snapshot and marks the previous snapshot non-current. Removed reviews stop contributing on the next completed recalculation; historical snapshots remain available for audit and ranking history.

8. Manipulation prevention and corrections

Reviews may be held for manual review based on duplicate, spam, conflict, burst, or abuse signals. Signals support moderation; they are not automatic proof of fraud. Companies cannot submit self-reviews or silently assign themselves a position. Administrative changes must be recorded in the audit log.

Report incorrect profile data through the profile correction tools. Ranking or review disputes may be filed under Ranking Appeals. Editorial corrections follow the Editorial Standards.