Spotting AI-Washing (and What “Proof” Looks Like)
Common AI-washing patterns
- Vague claims: “AI-powered” with no technical detail
- No metrics: can’t explain accuracy, precision/recall, or validation
- Refuses a real demo with your data
- Unrealistic promises: “100% accuracy” or “no human review needed”
What good evidence looks like
- Benchmarks by task (e.g., clause extraction accuracy by document type)
- Validation methodology (holdout sets, sampling, error analysis)
- Clear limits and known failure modes
Use this visual during demos
Use the quick check visual to keep conversations grounded in evidence.