How we collect, verify and score

We don’t store “cashback = 3%”. We store the value, its unit, conditions, the exact source page, source type, retrieval date, verification date and a confidence level. Values we cannot find are shown as Not disclosed (the source was checked) or No data yet (not researched) — never guessed.

When sources disagree, the higher-priority source wins (terms beat pricing pages, which beat FAQs and marketing). Community sources (tier 4) can never publish a fact. If marketing says “up to 4%” and the terms say “3% base”, we show 3% as base and 4% as a conditional maximum.

• Source pages are fetched and fingerprinted automatically, usually every day (longest-unchecked cards first). If a page is unchanged since its values were read, those values are confirmed as still current.
• When a page changes (or has not been read yet), its values are re-read in a later research run. Research runs are started by our editor at irregular intervals and use AI-assisted extraction: every value must quote the page word for word, and each quote is checked against our stored copy of the page before the value is accepted.
• Until a changed page has been re-read, the card keeps showing the date it was last verified — a page that changed is not counted as a verification.
• Material changes (cashback, FX, fees, country availability, credit terms, discontinuation) are never published automatically — an editor approves them with old value, new value, source and confidence side by side.
• Only low-severity changes from official sources with ≥ 90% confidence are auto-published.
• If none of a card’s source pages can be fetched or read, the card shows “Verification failed · last successful X days ago” instead of presenting old data as current.

Our initial dataset was collected from official issuer pages via a search index. A card is labelled “Initial research · not yet verified” until every readable source page of it has been read in a research run. Its score stays provisional as long as any of its values is still unconfirmed initial research — for example a value from a page that cannot be read automatically.

• Each criterion is scored 0–100 from published facts only, with documented formulas (e.g. FX: 100 − 40 × fee%; base cashback counts more than “up to” rates; token-paid cashback −10%; staking-gated top rates −30%).
• Missing criteria are excluded; their weight is filled with a neutral 50/100 so a card with little data can neither win nor lose by omission. Below 50% data coverage we show “Not enough data to score”.
• Transparency rewards issuers that disclose key terms and penalises contradictions between their own pages.
• Closing, paused and discontinued products score 0 on availability and are excluded from all rankings.
• overall = round(raw_0–100 / 10, 1).

• Best Overall — Default weighting across all nine criteria.
• Best for Cashback — Rewards dominate; fees still matter because they eat into rewards.
• Best for Travel — FX costs, ATM access and wallet support abroad.
• Best for Self-Custody — Only cards where you keep control of your keys.
• Best for DeFi Users — Crypto-native cards with on-chain custody and borrowing.
• Best for Stablecoins — Cards that spend stablecoins directly.
• Best for High Spenders — Reward rates and limits that scale with spend.
• Best for Low Fees — Lowest subscription, FX and ATM costs.
• Best for Beginners — Simple, transparent cards with wallet support and broad availability.
• Best Crypto Credit Card — Cards with a real credit line, including crypto-backed credit.
• Best E-Money Card — Fiat-first FinTech and e-money cards.

Country scores replace the availability criterion with the status for that country: available, not available, not listed, or unknown. Cards are only ranked for a country if the issuer lists it. The Card Finder computes a fit score = overall × country availability × preference match × cost efficiency.

Net annual value = base cashback (respecting caps) − annual/monthly fees − FX fee × foreign spend − conversion fee × crypto-funded spend − ATM fees beyond the free allowance. Paid tiers, staking boosts and promotions are not included. Staking opportunity cost, where shown, is an explicitly labelled estimate using a yield you choose. If any applicable fee is unknown, the result is marked “Not enough data”.

We earn money through affiliate links. Affiliate data is stored separately and the scoring and ranking code has no access to it. Sponsored listings are labelled as such and never change a score. We never claim a card is “the best” for everyone, “risk-free” or that cashback is guaranteed.