Most bad AI output is caused by underspecified prompts. Here are the mistakes that produce generic, risky, or incorrect results.
1) No deliverable
“Help me with marketing” is not a deliverable. Ask for a specific asset: ad angles, landing page hero, email sequence, SEO outline.
2) Missing audience and objections
Without audience pains and objections, the model defaults to vague benefits.
3) No constraints
If you don’t specify format and length, outputs drift. Add structure and boundaries.
4) Asking for facts without sources
Models can hallucinate. Ask it to label assumptions and to avoid making up numbers.
5) Treating v1 as final
Expect a draft. Then iterate with stronger constraints and examples.
Fix pattern
After any output, ask: “What assumptions did you make? What could be wrong? Give a checklist to validate.”