AI Output Verification Kit
Quick Reference Card + 27-point Hallucination Detection Checklist for verifying AI-generated content before publishing.
No email required. No account needed.
Quick Reference Card
The 4-Phase Process
EXTRACT → VERIFY → SCORE → REPORT
EXTRACT
- Break AI output into individual verifiable claims
- Tag each claim by type: statistic, attribution, quote, or inference
- Prioritize high-stakes claims for verification first
VERIFY
- Check each claim against reliable, independent sources
- Apply the source reliability hierarchy to weight evidence
- Document verification status for every claim
SCORE
- Assign a confidence score to the overall document
- Map the score to confidence bands: HIGH, MODERATE, LOW, or INSUFFICIENT
- Flag claims that cannot be verified and assess their risk
REPORT
- Document findings, corrections, and unresolved claims
- Note the risk level of any unverified content
- Produce a publish, revise, or rewrite recommendation
Source Reliability Hierarchy
- 1Primary sources greater than
- 2Peer-reviewed greater than
- 3Institutional greater than
- 4Industry greater than
- 5AI-generated
Confidence Bands
Hallucination Detection Checklist
27 systematic checks across 6 fabrication categories
1. Fabricated Sources
- Cited paper or report does not exist in scholarly databasesCritical
- Author name does not match any real publication recordCritical
- Journal or publisher name is invented or misspelled to appear legitimateCritical
- DOI or URL provided does not resolve to the claimed sourceCritical
- Citation attributes a real paper to the wrong author or institutionHigh
2. Statistical Fabrication
- Statistic cannot be found in the cited original sourceCritical
- Number is close but not exact to the source (e.g., 32% vs 35%)High
- Implausible precision without cited methodology (e.g., "exactly 47.3%")High
- Sample size or study scope misrepresentedMedium
- Statistic applied to wrong industry, region, or demographicHigh
3. Temporal Errors
- Data from prior years presented as current without date qualificationHigh
- Laws, regulations, or policies cited that have been supersededHigh
- Product features, pricing, or leadership roles that have changedMedium
- Anachronistic claims (events referenced before they occurred)Critical
4. Attribution Errors
- Direct quote cannot be found in attributed person's published workHigh
- Paraphrased content presented as a direct quotationHigh
- Quote attributed to wrong person or organizationHigh
- Title or role of quoted individual is incorrectMedium
5. Logical Fabrication
- Correlation presented as causation without supporting evidenceHigh
- Cherry-picked examples presented as established trendsHigh
- Circular reasoning (conclusion used as premise)Critical
- Missing counterpoints or contradictory evidence omittedMedium
- Conclusion does not follow from the evidence presentedHigh
6. Confidence Miscalibration
- Hedging language ("may", "suggests") removed to state speculation as factHigh
- Uncertain claims presented with unwarranted certaintyHigh
- Confidence level exceeds what sources actually supportHigh
- Limitations and caveats from sources stripped from summaryMedium
Download the complete kit as a printable PDF
Quick Reference Card and full 27-point checklist — ready to print or share with your team. No email required.
Download PDFWant the editable version?
Enter your email for Google Sheets + team sharing guide
When you need more
This checklist tells you what to check. The Verification System shows you how to check systematically — with claim extraction worksheets, source verification matrices, confidence scoring, and report templates.
Methodology reviewed: August 2026. Next review: November 2026.
Version 1.0 — Based on AI Research Verification System methodology
License: Free for personal and professional use. Attribution appreciated.