Most disappointing AI answers are caused by vague prompts, not a weak model. A prompt is a brief — and the more precisely you write the brief, the better the work you get back. These ten rules will noticeably improve your results, and the builder at the end assembles a solid prompt from your inputs.
1. Give it a role
“You are a Nigerian tax consultant explaining to a small-business owner” primes the tone, depth and assumptions. A role focuses the answer before you have said anything else.
2. State the task in one clear sentence
Exactly what you want produced: “Draft a 200-word LinkedIn post announcing…”, “Compare these three options on cost, risk and speed.” Ambiguity in, ambiguity out.
3. Supply the context it cannot know
Your audience, your constraints, the background, the relevant facts, what you have already tried. AI has no access to your situation unless you put it in the prompt.
4. Specify the format
Bullet list, table, email, 3 paragraphs, headline plus body, a specific word count. If you do not say, you get the model’s default — usually longer and blander than you want.
5. Give an example of “good”
One short example of the style, structure or level you are after does more than a paragraph of description. Show, don’t just tell.
6. Set constraints and boundaries
“No jargon.” “Nigerian context and naira.” “Do not invent statistics — say if you are unsure.” “Keep it under 150 words.” Constraints sharpen the output.
7. Ask it to ask you
End with “If anything is unclear or you are missing information, ask me before answering.” This alone prevents a lot of confidently wrong first drafts.
8. Iterate — do not restart
Refine in follow-ups: “Shorter.” “More direct.” “Rewrite the second paragraph for a sceptical reader.” Treat it as a conversation, not a slot machine.
9. Give it your source material
Paste the notes, the transcript, the data, the draft. AI is far more accurate working from material you provide than from its own memory — and far less likely to hallucinate.
10. Tell it how to check itself
“List any assumptions you made.” “Flag anything you are not confident about.” “Give me the two weakest points in this argument.” Building in self-review catches errors before you do.
Build a prompt
Prompt builder
Fill in what applies. The builder assembles a structured prompt you can copy into any AI tool.
Weak prompt vs strong prompt
| Weak | Strong |
|---|---|
| “Write about saving money” | “You are a Nigerian personal-finance writer. Write a 180-word intro for beginners on why to save before you spend. Plain English, naira examples, no clichés. End with one concrete action.” |
| “Summarise this” | “Summarise the attached contract in 6 bullets: parties, term, payment, termination, penalties, anything unusual. Flag anything I should query.” |
| “Help with my CV” | “Rewrite these 3 bullets to lead with a quantified result, action verb first, one line each. Here are the real numbers: …” |
A worked example
Bature needed a firm reminder email to a vendor who was three weeks late. His first try — “write an email to my vendor about the delay” — produced something generic and slightly rude.
Using the builder: role = “a procurement officer”; task = “draft a firm but professional reminder about a 3-week overdue delivery”; context = “second reminder, PO #4471, we value the relationship but need a firm date”; format = “email, under 120 words”; constraints = “polite, no threats, request a specific commitment date.” The result needed one small edit and went out. Same model, far better brief.
Prompting mistakes
- One vague line and disappointment when the answer is vague too.
- No format specified, so you get a wall of text.
- Not pasting your source material, so it invents details.
- Starting over instead of refining with follow-ups.
- No instruction to flag uncertainty, so errors look confident.
Prompting FAQ
Do I really need a long prompt every time?
Not always — simple questions need little. But for anything you will act on or send, a few extra lines of role, context and format dramatically improve the result and save editing time.
What is the highest-impact single addition?
Context the model cannot know: your audience, your constraints, the background. That plus a specified format fixes most poor answers.
How do I stop it inventing facts and numbers?
Give it the source material to work from, tell it not to invent figures and to flag uncertainty, and verify anything important yourself.
Should I save my good prompts?
Yes. For any task you repeat, keep the prompt as a template. It is the fastest way to get consistent results.
Prompting is one AI skill of several
See 8 AI Skills Every Ambitious Nigerian Needs Now and 8 AI Workflows That Save Time.
Your next action
Use the builder for a task you have right now. Copy the prompt, run it, then refine with one or two follow-ups instead of starting over.
General guidance on using AI tools; outputs still need your review and verification.
Read Next
- 8 AI Skills Every Ambitious Nigerian Needs Now
- AI for Nigerian Professionals: 9 Smart Ways to Work Better
- AI Skills Every Professional Should Learn — the full guide
More in Future Mindset.
