How the AI Companion Trust Score Works
The score is a decision aid, not a certification. It measures how many practical questions you can answer with evidence before paying; it does not inspect a service's internal systems.
Evidence, not brand reputation
The tool asks about facts a visitor can verify: operator identity, policy consistency, support, checkout clarity, renewal, credits, cancellation, deletion, chat use, media visibility and receipt naming. It contains no hidden list of approved or rejected brands.
How answers affect the number
A clear positive answer adds no risk points. An unknown answer adds one unit because uncertainty needs follow-up. A negative answer adds two units. Higher-impact questions carry more weight. The total is converted to a 0–100 evidence-gap score, where a higher number means more unresolved or negative answers.
Why unknown is different from no
Not finding a policy is not proof that the worst outcome applies. The tool preserves that distinction while still warning that you lack information needed for a confident payment decision. The action list tells you which unknowns to investigate first.
What the tool cannot prove
It cannot confirm security controls, legal compliance, model behavior, future billing conduct or the accuracy of every policy statement. A low score means you found clearer evidence, not that no risk exists. A high score is a reason to pause, not an accusation of fraud.
Keeping the method honest
Scores are calculated in your browser and are not sent to us by this static tool. We do not save the selected answers. Named services shown elsewhere on the site are optional starting points and never receive preset scores.
Reproduce the calculation yourself
Each yes receives zero points, unknown receives one times the question weight, and no receives two times the weight. The maximum weighted total converts the result to a percentage. Operator identity, checkout, renewal, cancellation, deletion and private-content use carry more weight because mistakes there are harder to unwind. Receipt naming and sales pressure carry less weight but still add context. You can ignore the displayed band and inspect the action list directly; the same inputs always produce the same result. Because the questions and weights are public, another person can reproduce the number without trusting a hidden model. The method measures evidence gaps only and is not a technical security audit.