Start with the outcome
Begin by explaining what you need the AI to help with, who the result is for and what a good answer would look like. Avoid relying on a vague command such as “make this better”.
Five ingredients of a useful prompt
- Role: the perspective or expertise to apply.
- Objective: the action and result you want.
- Context: facts, background, audience and source material.
- Output: format, length, tone and structure.
- Checks: constraints, assumptions, verification and clarifying questions.
Compare a vague and a useful prompt
Vague: “Write a project plan.”
Act as an experienced delivery manager. Create a four-week plan for introducing a new internal helpdesk process to a team of 25 people. Include milestones, owners by role, dependencies, risks, communications and acceptance criteria. Use UK English and a concise table. Flag any assumptions.
Use examples and source material wisely
Include only information you are permitted to share with your chosen AI service. Treat pasted documents as reference material; review the generated response for accidental assumptions and omissions.
Test, challenge and iterate
Try the prompt, inspect the result, identify what did not meet the brief, and improve the wording. Ask for sources only when the model can actually access them, and independently verify important claims.
Common prompting mistakes
- Leaving out the audience or success criteria.
- Asking for incompatible output requirements.
- Overloading the prompt with irrelevant material.
- Assuming a role instruction guarantees expertise or accuracy.
- Sharing sensitive or confidential information without approval.
Frequently asked questions
Do I need a different prompt for each AI assistant?
Usually not. Plain-text prompts are portable, although different models may behave differently. Review the result and adapt as necessary.
Does a longer prompt always work better?
No. Relevant information and precise instructions matter more than length.
