Mistakes
Common Mistakes
Most bad prompts fail for ordinary reasons: unclear task, missing context, missing source material, or no output target.
Confidence isn't reliability
OpenAI Help Center
Key Facts
- Most weak prompts fail because the task, context, source, or output format is missing.
- A confident AI answer can still contain unsupported claims, fabricated details, or wrong assumptions.
- The safest beginner habit is asking the model to request missing information before answering.
A bad AI answer does not always mean the tool is useless. Sometimes it means the request gave the model too much room to guess. The mistakes below are common because they feel natural when you are just starting. Each one has a simple repair.
1. Asking a vague question
"Help me with marketing" gives no job. Use a verb: draft, compare, critique, summarize, plan, or explain.
2. Burying the real request
If you paste a long story and ask the question at the end, the model may optimize for the wrong part. Put the task first.
3. Leaving out source material
When the answer depends on a document, policy, transcript, rubric, or email thread, include the relevant material.
4. Asking for too many jobs at once
"Research, analyze, write, design, and plan" can blur quality. Break complex work into focused steps.
5. Skipping output format
If you need a table, checklist, short draft, JSON, or bullets, say so before the model chooses for you.
6. Trusting confident claims
AI can sound sure while being wrong. Ask for sources, then check important claims yourself.
7. Treating the first answer as final
The first answer is often a draft. Use follow-ups to correct tone, detail, assumptions, and format.
8. Using personas instead of details
"Act as an expert" is weaker than naming the audience, source material, goal, and criteria for success.
Repair prompts for the most common misses
- Too vague: "Restate my request as a clearer prompt before answering. Ask questions if the task is underspecified."
- Too generic: "Revise with examples specific to [audience/context]. Avoid advice that could apply to any situation."
- Unsupported: "Mark each factual claim as sourced, inferred, or needs verification."
- Wrong tone: "Make this sound [direct/warm/formal] for [relationship]. Keep the facts unchanged."
- Too long: "Compress to [exact shape]. Preserve only decisions, risks, and next actions."
The safest beginner habit
Add one sentence to important prompts: "If the answer depends on information I have not provided, ask me questions before answering." This reduces guessing and teaches you what context the task actually needs.
The second safest habit
Separate drafting from fact-checking. It is fine to use AI to make a rough draft, outline options, or explain a concept. It is not fine to publish a claim, send a legal-sounding answer, make a purchase, or change medical behavior just because the answer sounded polished. Prompting can improve usefulness. It does not remove your responsibility to verify.