Our Methodology
Promptalyze is built on transparency, not black-box scores. Here's how we measure and evaluate.
Visibility detection
Promptalyze detects brand presence through deterministic string matching: your brand name and all configured aliases are checked case-insensitively against the normalized answer text. The result is binary and traceable — either the brand is mentioned or it isn't.
Citation detection
Citations are detected via domain matching: every URL contained in an answer snapshot is checked against your configured owned domains and competitor domains. Subdomain matching is included. The result shows exactly which domains are cited.
Accuracy assessment
Accuracy assessment is based on the critical facts framework: you define key facts, keywords, and patterns. The system detects when configured keywords are missing or risky patterns appear. AI-assisted analysis supplements but does not replace deterministic detection.
Diagnostic generation
Diagnostics are derived from a combination of findings and historical comparisons. Severity follows a transparent matrix: issue type × question priority = severity. Every diagnostic references concrete evidence (snapshots, findings, diffs).
Recommendations
Recommendations are evidence-based: each references the triggering finding and includes a rationale with trigger, evidence, expected outcome, and confidence. The taxonomy includes 16 recommendation types for precise, actionable guidance.
What we don't do
We don't claim semantic truth verification we can't deliver. We don't generate scores without inspectable foundations. We don't bypass anti-bot measures. Our methodology is deterministic and explainable, supplemented by honest AI assistance.