CogniPrep audits AI-generated test reports to ensure factual accuracy

CogniPrep's test prep app generates prose reports from user metrics, treating each sentence as a factual claim. The company audited every possible sentence the app could produce to identify and fix misleading statements. The audit uncovered three main defect classes: awkward phrasing that erodes trust, incorrect grammatical details like article usage, and invalid statistical claims from single data points. The core fix involved implementing logic gates to prevent claims from being generated unless the underlying data supports them.