TopolearnMost disappointing AI output comes from vague prompts; a clear one is the cheapest and biggest improvement you can make.
Imagine handing a task to a capable new colleague who knows nothing about your company or what you have in mind. "Write something about our sale" gets you a guess. A clear brief gets you what you need. Prompts work the same way.
A good prompt usually covers:
- The task, stated directly: "Write a 3-sentence email announcing...".
- Context and purpose: who it is for and why. "Customers are small bakery owners who are busy and skim emails." Knowing the purpose lets the model make sensible choices you did not spell out.
- Constraints: length, tone, words to avoid, what to do when information is missing.
- The output format: bullet list, table, a single word, or a fixed template.
- The material: the text to work on, clearly separated from your instructions, for example inside triple quotes or labelled sections such as
<document>...</document>.
Other tips that work across models:
- Say what to do rather than only what not to do. "Write in plain sentences" beats "don't use bullet points".
- Break a big job into steps, either in one prompt ("First list the issues, then rank them") or as separate prompts one after another.
- Ask for one thing per prompt when quality matters.
- Iterate: read the output, find what is off, and add the missing instruction. Keep the improved prompt for next time.
In automation tools, insert the changing data (the email body, the form answer) into a fixed prompt template using the tool's variable fields. The fixed part holds the instructions; the variable part holds the material.
Example
Vague:
Write about our new opening hours.
Clear:
Write a short notice for our bakery's Instagram.
Audience: regular local customers. Tone: warm, no hype, no emojis.
Facts: from 1 June we open at 7:00 instead of 8:00, Monday to Saturday.
Length: at most 2 sentences. If a detail is missing, leave it out; do not invent one.Remember
- State the task, context, constraints and output format
- Separate instructions from the material with labels or delimiters
- Say what to do, split big tasks into steps, and iterate