Weak answers usually come from vague prompts. The model isn’t ignoring you; it is filling the gaps with guesses. Every detail you leave out is a guess it has to make, and the average guess produces an average answer.
The good news is that three small habits fix most of the problem.
1. Give it a role
Instead of “write an email”, say “as a sales manager, write a short email to an unhappy customer”.
A role tells the model which vocabulary, priorities and level of expertise to use. A few more you can borrow:
- “As a patient math tutor for a 12-year-old…”
- “As an editor who removes filler words…”
- “As a travel planner working with a tight budget…”
2. Give it context
The model can’t see your situation unless you describe it. Before you press send, check that you’ve covered:
- Audience: who will read this?
- Goal: what should happen after they read it?
- Tone: formal, friendly, direct?
- Length: three sentences or three paragraphs?
- Limits: anything to include or avoid?
3. Give it an example
Paste a sample of the output you like and say “match this style, but write it about X”. Examples are the fastest way to communicate taste, because they show what words struggle to describe.
You can also show what you don’t want: “Avoid sounding like this: …”
Before and after
Weak prompt
Write a post about our new app.
Better prompt
You are a friendly social media manager. Write a 3-sentence Instagram caption announcing our new language-learning app to busy adults. Keep the tone warm, not salesy, and end with a question. Here is a caption I like: “[paste example]”.
The second version takes thirty seconds longer to write and saves several rounds of rewrites.
A reusable template
Copy this and fill in what applies:
Role: You are a …
Task: Write / explain / compare …
Context: The reader is … The goal is …
Format: … (length, bullets, table, tone)
Example: Here is something I like: …Don’t stop at the first answer
Treat the first reply as a draft. Follow up with short instructions:
- “Make it shorter and more direct.”
- “Give me three different versions.”
- “What assumptions did you make?”
Common mistakes
- Asking for five things in one prompt. Split big tasks into steps.
- Hiding the real goal. Say what the output is for.
- Accepting the first draft without checking facts.
- Pasting private or confidential data into an online tool.
Try it today
Pick one prompt you already use. Rewrite it with a role, some context and an example, then compare the two answers side by side. The difference is usually obvious.