WHAT MAKES THE REPORT RELIABLEDecision-grade intelligence requires rigor.
A decision worth real budget consideration demands rigor and diligence. Every AI platform sample uses the current default models of each platform.
95% Confidence Interval
A measured range at every reported level. An initial read is sized to a minimum of 100 observations per reported cell, per platform, which places a 95% Wilson interval around every rate we publish.
Full-Context Scraping
We read what the model reads. We analyze actual cited content across text, schema, and transcripts, not page titles and meta tags.
Crawl & Access
Proof AI can read your content. CDN fingerprinting, bot-access simulation, routing and WAF checks on cited pages.
7-Type Accuracy
Hallucinations, named and scored. From fake citations to staleness, 7 types of accuracy errors scored by severity across every claim.
Citation Provenance
Every source, classified. Traced and marked owned, influenceable, or competitor-controlled.
Temporal Sampling
A baseline plus a variance band. Sampled across a time window, so you see what is stable and what is drifting.
This is what turns AI noise into intelligence. The full procedure, including our sampling design, statistical treatment, and entity-resolution rules, is published on the Method pages. Every finding in the Report can be re-run against it.