How to Write Prompts That Actually Work
Master role, task, context, output format, and evaluation criteria to turn AI from 'able to talk' into 'actually usable.'
What you'll learn
- Understand the 5 core building blocks of a strong prompt.
- Learn how to give AI context and constrain its output.
- Know when to forbid AI from making its own assumptions.
Step-by-step guide
Why do prompts matter so much?
Two users can get very different results from the same AI model. The difference usually isn't the model — it's how clearly you briefed it. Treat AI like a consultant on their first day: you wouldn't hand a person a one-line request like "make me a marketing plan," and AI needs the same context and goals.
The 5 core building blocks of a good prompt
1. Role: Tell the AI who it is right now (e.g., a senior B2B website strategy consultant). 2. Task goal: What should this produce, and who will use it? 3. Background: Company, industry, size, audience, constraints. 4. Steps or reasoning path: How many stages to complete this in. 5. Output format and evaluation criteria: Word count, sections, tables, things to avoid.
A common fix: what if some information is uncertain?
AI doesn't have your company's data, so it will easily "make things up." Add a line like: "If any of the variables below are missing or insufficient, list the missing fields and ask the user — do not assume or fabricate data." This significantly reduces hallucination.
When should you restrict AI's autonomy?
Whenever numbers, regulations, client names, or financial data are involved, explicitly instruct the AI "do not fabricate." Also clearly mark fields that "require human confirmation" so drafts don't get mistaken for finished work.
Good example / Bad example
You are a senior B2B website consultant. Task goal: produce a site-architecture proposal for [Company Name]. Company background: ... Steps: 1. Summarize positioning; 2. Recommend 6–10 pages... Output format: bullet points + tables. No fabrication: if any variables are missing, list them.
Write me a website structure.
Common mistakes
- Expecting a tailored recommendation from AI without giving any company background.
- Stuffing too many tasks into one conversation, causing the AI to lose focus.
- Copying AI output straight into a final draft without review.
Cautions
- AI output is only a draft; final decisions still require human judgment.
- For sensitive data, confirm your company's policy on AI tool data usage.
Related prompts
Related articles
Need help customizing this for your company and industry?
AEGIS provides enterprise AI adoption, process assessment, and systems integration consulting to help you turn prompts and AI tips into reliable internal workflows.