ai content fact audit

The problem

AI-generated content often includes "hallucinated" citations or real statistics misattributed to the wrong sources. One fabricated number or mis-scoped metric can destroy your brand's authority and technical credibility.

What it does

• Extracts every checkable factual proposition, statistic, and regulatory claim from a document.
• Traces citations upstream to the originating primary source rather than relying on secondary blog posts.
• Assigns standardized provenance-strength grades including PRIMARY-TRACED, SECONDARY-SOURCED, and UNCORROBORATED.
• Applies composable flags for MIS-ATTRIBUTED, MIS-SCOPED, or CONTRADICTED claims.
• Provides actionable recommendations to cut, re-source, or re-scope weak claims for transparency.

Why this beats prompting it yourself

Standard prompts often ask an LLM if a statement is "true." This skill ignores the concept of "truth" in favor of checkable provenance. It uses a specific multi-axis grading rubric that separates the strength of the source from the scope of the claim, catching "population drift" errors that typical prompts miss.

Use cases

• Auditing a technical whitepaper or pitch deck for statistical accuracy before publication.
• Verifying that AI-generated citations actually exist and support the claims made.
• Reviewing marketing copy to ensure single-vendor sample data isn't being framed as an industry standard.
• Validating regulatory or compliance assertions against actual standards or agency reports.

Known limitations

This skill grades provenance, not "truth." It cannot verify claims where the primary source is a matter of contested expert opinion or is behind an inaccessible paywall.