1. Fixed buyer questions
Each audit creates 15 commercially relevant questions without the target brand name.
2. Three web-grounded AI systems
The current benchmark measures OpenAI, Gemini and Claude through server-side model endpoints.
3. Separate evidence fields
Mentions, official-domain citations and recommendation position are stored separately.
4. Repeat runs
Directional repeats each question three times. Verified repeats each question seven times.
5. Failed providers
Timeouts and provider errors are displayed and excluded from the score.
6. Interpretation
The result is a standardized benchmark for the selected market, language and measurement time—not a universal consumer-app ranking.