250 ways AI gets it wrong

Hallucinated facts, invented statistics and stale information stated as current.

• Hallucinated facts
• Made-up statistics
• Invented legal rules

Fabricated citations, broken links and sources that don't support the claim.

• Sources that don't support the claim
• Irrelevant sources
• Outdated sources

Shallow research presented as deep, cherry-picked evidence and missed changes.

• Shallow research presented as deep
• Stopped searching too early
• Searched the wrong terms

Generic strategies, ignored constraints and plans too vague to execute.

• Generic strategies
• Consultant-speak with no practical value
• Smart-sounding tactics that don't work

Artificial tone, cliché endings and your strongest point rewritten away.

• Artificial tone
• Cringe or over-polished language
• LinkedIn-cliché tone

Missed obligations, confused legal systems and guarantees that don't exist.

• Legal-sounding advice without qualification
• Missed licensing requirements
• Missed disclosure requirements

Arithmetic errors, confused figures and false precision in the numbers.

• Arithmetic errors
• Wrong formula
• Misread tables

Code that doesn't run, hardcoded secrets and missed vulnerabilities.

• Code that doesn't run
• Code that runs but does the wrong thing
• Outdated libraries

Forgotten constraints, self-contradiction and 'almost right' deliverables.

• Answered instead of asking for missing input
• Asked unnecessary questions
• Overcomplicated simple tasks

Leaked sensitive information, missing redaction and insecure suggestions.

• Leaked sensitive information into outputs
• Encouraged pasting confidential data into unsafe tools
• Failed to redact personal information

Weak recommendations made to sound decisive, with hidden risks unexamined.

• Made weak recommendations sound decisive
• Recommended action without ranking options
• Failed to identify the safest option

Generic layouts, ignored accessibility and premium confused with decorative.

• Generic designs
• Copied design trends without judgment
• Attractive but impractical layouts

Flattery instead of challenge, and opinions withheld when one was needed.

• Flattered the user instead of challenging weak thinking
• Agreed too easily
• Pushed back too aggressively

Low-substance answers, filler and meaning changed during rewriting.

• Long answers with low substance
• Short answers that skip essential detail
• Repetition

Wrong steps, false success messages and single errors repeated at scale.

• Executed the wrong step
• Used the wrong tool
• Called tools unnecessarily