Most disappointing AI output comes from vague prompts, not a limitation of the tool itself. A few habits consistently improve results.
Give it a role and an audience
"Write a summary" produces something generic. "Summarize this for a board member with no technical background, in three sentences" produces something usable.
Vague
Write a summary of this.
Nothing here says who it’s for or how long it should be, so the model picks for you — and usually picks generic.
Specific
Act as a communications lead, writing for a board member with no technical background, summarize the attached incident report, in three sentences, plain language, no jargon.
- Role
- Audience
- Task
- Format
Be specific about format
Asking for a numbered list, a specific word count, or a particular tone (formal, conversational) up front saves a round of back-and-forth editing.
Provide the actual context
Pasting in the relevant email thread, document, or data — rather than describing it from memory — consistently produces more accurate, more useful output.
Iterate instead of starting over
If the first response isn't right, ask for a specific change ("shorter," "less formal," "focus more on the budget impact") rather than rewriting the prompt from scratch.
Treat it as a draft, not a final answer
Even a well-written prompt can produce a confidently wrong answer. The prompt gets you a strong starting point — review is still the human's job.
A little more specificity up front consistently outperforms a vague request followed by several rounds of correction.