
How we rank betting sites and casinos
Every UK operator is scored by the same published 0-10 formula from sourced records. We may earn affiliate commission, but commission is never an input: operators that do not pay us are scored identically and can rank first.
Four pillars
Casinos
Betting sites
Offer fairness
Casino offers are scored mainly on wagering (70%): no wagering scores full marks, and the multiple is doubled when it applies to deposit plus bonus. Expiry and minimum deposit make up the rest. Free bets are scored on value per pound staked, minimum odds, expiry and whether the stake is returned. The headline amount alone never decides the score.
Payments
Breadth of methods, availability of fast e-wallets and open banking, the slowest processing time the operator states, withdrawal fees and minimum withdrawal. Processing times are what the operator publishes; we label them as stated, not measured.
Player ratings
Trustpilot (40%), App Store (30%) and Google Play (30%). Each rating is Bayesian-adjusted toward a neutral prior, so a perfect score from a few dozen reviews cannot beat a strong score from hundreds of thousands.
Product
Casinos: stated game library (scaled to 6,000 games), live casino and app-store quality. Betting sites: cash out, bet builder, live streaming and app-store quality. A product pillar needs at least two verified features.
Rules we do not break
- Commission is not an input. Operators without an affiliate deal are scored identically and can rank first.
- Missing data is never guessed. A pillar we cannot verify shows n/a and is left out of the score. Operators with fewer than four verified pillars are marked provisional and listed after fully verified ones.
- No licence, no rank. An operator whose UK Gambling Commission entry we cannot verify is capped at 5/10.
- Sources and dates on every fact. Records expire and are re-checked; stale records drop out of answers.
How the AI adviser works
Frequent questions are answered instantly by deterministic queries over the records. For anything else, the model (Z.ai GLM) receives only the current records for the chosen category, selects up to six, and writes a short answer with a citation marker after each sentence. Before we show it, the server checks every marker points at a selected record and every number in the answer appears in a cited record. If not, you see the records without the generated prose. The adviser refuses predictions, betting transactions and loss-chasing questions.