Skip to content
Statistical Rigor

Our Measurement Methodology

How AmpliRank collects, calculates, and verifies AI answer evidence without black-box distortion or fabricated projections.

1. Wilson Score 95% Confidence Intervals

In LLM answer measurement, a brand does not have a single static “rank”. If you query an engine 20 times, your brand may appear in 8 answers. Rather than declaring an overconfident “40% visibility metric”, AmpliRank computes the 95% Wilson score confidence interval:

CI = (p̂ + z²/2n ± z√((p̂(1-p̂)/n) + z²/4n²)) / (1 + z²/n)

For 8 mentions out of 20 runs (n=20), the 95% interval is [21.1% – 61.3%]. When your panel expands to 100 runs and 40 mentions (n=100), the interval tightens to [30.9% – 49.8%]. We report both the midpoint and interval bounds on all dashboards.

2. Provider APIs vs Consumer Assistants

AmpliRank queries official provider APIs (OpenAI ChatGPT with Search, Anthropic Claude, Perplexity Sonar, and Google Gemini).

Important distinction: API samples provide clean, unpersonalized, and auditable baselines. Consumer assistant apps (such as the mobile ChatGPT app or Google AI search on personal accounts) include personalized search history, geolocation, device factors, and A/B test variations. Our API sampling is an accurate measure of model knowledge and citation weights, not an exact replica of every user's personal session.

3. Zero Synthetic Revenue Claims

Many marketing platforms show badges claiming “$142,500 Revenue at Risk” or “+24% ROI Guaranteed”. These numbers are almost universally fabricated using arbitrary multipliers.

AmpliRank strictly forbids unobserved causal claims. We track actual observed events, verified web citations, and empirical mention rates. If attribution data is connected via webhook, we report strictly observed events—never simulated financial projections.

4. Cryptographic Provenance & Replay Safety

Every scan stores the full raw LLM response, token consumption, latency, and model version. When an action draft is published, AmpliRank applies compare-and-set (CAS) cryptographic fingerprints across content and cited sources to guarantee published briefs cannot be mutated retroactively without re-verification.