Back to prompts
AI Masterybeginner
0.0

The BS Detector — Catch AI Hallucinations Before They Catch You

A systematic prompt that forces AI to flag its own uncertain claims. Trust but verify — automatically.

Copy & Paste this prompt
You are a fact-checking specialist and AI reliability analyst.

I'm about to use AI-generated content for something important. Before I do, audit it for hallucinations, fabrications, and overconfident claims.

The content to audit:
[PASTE AI-GENERATED CONTENT]

The topic/domain: [WHAT IS THIS ABOUT?]

Perform this audit:
1. CONFIDENCE MAP — Go through every factual claim and rate your confidence:
   🟢 HIGH (>90%) — Well-established fact, easily verifiable
   🟡 MEDIUM (50-90%) — Likely true but should be verified
   🔴 LOW (<50%) — Possibly hallucinated, unverifiable, or suspicious
   ⚫ FABRICATED — This is definitely made up (fake statistics, non-existent studies, invented quotes)

2. RED FLAG PATTERNS — Identify common hallucination tells:
   - Suspiciously specific statistics without sources
   - Named studies or papers (often fabricated)
   - Quotes attributed to specific people
   - Historical dates and events
   - Technical specifications or benchmarks

3. VERIFICATION CHECKLIST — For every 🟡 and 🔴 claim, tell me exactly what to Google to verify it

4. CLEANED VERSION — Rewrite the content with hallucinations removed and uncertain claims properly hedged

5. TRUST SCORE — Overall reliability rating (0-100%) with justification
Use as Template
#hallucination#fact-checking#reliability#ai-safety

Works with

chatgptclaudegemini

💡 Pro Tips

  • •Always run this on AI content before publishing, presenting, or making decisions
  • •Use a different AI model than the one that created the content
  • •The more specific a claim is, the more likely it's hallucinated

✨ Example Output

CONFIDENCE MAP:
🟢 'Python is the most popular language for data science' — TRUE, verified by Stack Overflow surveys
🔴 'A 2023 Stanford study found that 67% of developers...' — SUSPICIOUS. Cannot verify this specific study exists.
⚫ 'As Jeff Bezos said in his 2019 keynote...' — FABRICATED. No such keynote quote exists.

TRUST SCORE: 62% — Content has 3 likely hallucinations that need verification before publishing.

🧠 Why This Works

This prompt exploits AI's ability to self-audit by forcing explicit uncertainty quantification. When instructed to flag claims by confidence level and identify unsupported assertions, AI becomes dramatically more honest about what it actually knows versus what it's pattern-matching.

📅 When to Use This Prompt

Use before publishing any AI-generated factual content — research summaries, technical explanations, historical claims, or statistical assertions. Critical for journalists, researchers, students, and anyone whose credibility depends on accuracy.

🎯 What You'll Get

You get a confidence-rated breakdown of every factual claim in the output, with high-confidence statements separated from uncertain ones. Specific claims are flagged for manual verification with suggested fact-checking approaches.

🔗 Related Prompts

Decision MakingPremium

Cognitive Bias Detector

Identify cognitive biases affecting your decisions with debiasing techniques.

cognitive-biaspsychologycritical-thinking
★4.6
intermediate