Methodology

Each audit runs a fixed, vertical-specific prompt panel against the answer engines through their official APIs with web search / grounding enabled, plus licensed SERP data for Google AI Overviews, with the location set to the prospect's metro. Because AI answers are non-deterministic, every prompt is sampled several times per engine (default 3×; 8× for baselines) and rates are reported with 95% confidence intervals computed by response-level bootstrap.

Answers are read by a structured extractor that records which businesses are named, in what order, with what sentiment, and which sources are cited. Nothing is scored by hand.

The AI Visibility pillar (0–10) is a weighted composite of five components — Mention Rate, Share of Voice, Answer Accuracy, Citation Ownership and Foundation Readiness — computed by the frozen rubric engine (scoring.py, regression-tested). Weights and bands are fixed; reps and reviewers cannot adjust a score. A change at re-baseline is only claimed as real when it clears the confidence interval.

• Consumer-interface responses may differ from API responses.
• This is a point-in-time baseline, not a ranking; AI answers shift week to week.
• No provider offers ranking control. Improvement is directional and measured as mention probability over a rolling window.
• Attribution from AI answers to revenue remains immature and is not projected.
• Every report lands in review before it is sendable; the reviewer attests entity match and finding accuracy.