geo-hallucination-checker_skill

Overview

This skill detects and annotates hallucinations, unsupported claims, fabricated studies, and incorrect conclusions in text so AI only cites verifiable content. It flags risky language, suggests evidence-friendly rewrites, and delivers a structured claim-level analysis to make content safe to publish or cite.

How this skill works

I parse the provided text into atomic factual claims, check available evidence (prioritizing user-provided sources), and classify each claim by status and risk level. The output includes a short risk summary, a claim-level table with reasons and suggested fixes, and—if requested—a hallucination-safe rewrite. I never invent sources and always mark unverifiable claims as unsupported or problematic.

When to use it

• When asked to fact-check, verify, or validate any written content.
• For medical, legal, financial, or technical claims that could cause harm if wrong.
• When text contains numbers, percentages, dates, study mentions, or strong superlatives.
• Before publishing GEO-optimized content that AI models may cite.
• When you want a safer, citation-ready rewrite of marketing or product copy.

Best practices

• Provide any source links or documents up front—these get priority in verification.
• Ask for an analysis-only pass first, then request a hallucination-safe rewrite if needed.
• Treat vague references (e.g., “a 2026 study”) as unsupported unless a precise citation is given.
• Prefer cautious language over absolutes: use ‘may’, ‘suggests’, or ‘early evidence’ when unsure.
• Mark high-risk medical/legal claims as ‘requires expert review’ if not clearly supported.

Example use cases

• Reviewing a landing page that claims specific clinical effects without citations.
• Checking a blog post that quotes percentages, sample sizes, or unnamed studies.
• Auditing FAQ content for overconfident guarantees or absolute statements.
• Validating GEO-targeted drafts to ensure they don’t invent institution names or reports.
• Turning unsupported marketing claims into citation-safe, compliant copy.