Talking to AI (Prompts & Tools)

Weak request: “Teach me photosynthesis.” Better request: “I am a Form 2 learner who understands that plants need sunlight but I do not understand how food is made. Explain photosynthesis in four short stages, define any new science word when it first appears, use a maize plant as the example, and finish with three questions without answers.” The improved prompt states prior knowledge, the gap, audience, structure, example, and a way to practise.

A manager wants a weekly project update. Instead of pasting a confidential meeting transcript into a public chatbot, the manager uses an approved enterprise tool or replaces names and sensitive figures with placeholders. The prompt asks for decisions, owners, deadlines, and unresolved risks. The manager compares the draft with original notes, corrects ownership and dates, and approves it before distribution.

A model may repeat the same false claim because both answers come from similar learned patterns. Verification means comparing the claim with independent evidence: the official policy, original research, a textbook, system records, or a qualified expert. Multiple generated answers can help reveal uncertainty, but agreement between them does not turn them into a source.

A useful request tells the AI who the audience is, what task to do, what information matters, and how to present the result. For example: “Explain photosynthesis to a Form 2 learner in Zimbabwe, using five bullets and one local farming example.” If the first answer is too hard or too long, change one instruction and try again rather than accepting it unchanged.

A class essay may contain a learner's name, family situation, health details, or contact information even if it does not show a password. Putting that text into a public tool can expose more than you intended. Use a made-up sample or replace identifying details with neutral labels such as “Learner A.”

First attempt: 'Tell me about small business marketing.' The answer is generic and could apply anywhere in the world. Second attempt: 'I run a small tailoring business in Bulawayo. My customers find me through word of mouth. Suggest five low-cost ways to reach new customers, assuming most people use WhatsApp and Facebook rather than email. Format as a numbered list with one action step each.' The second answer is specific, actionable, and easy to check. Nothing about the AI changed between the two attempts — only the quality of the brief. The information you provide is the biggest lever you control.

Nobody verifies everything, so build a proportional habit. For low-stakes uses like brainstorming a party theme, a quick sense-read is enough. For schoolwork, check names, dates, and definitions against your textbook or syllabus before you rely on them. For anything involving money, health, law, or another person's reputation, treat the AI answer as a starting hypothesis only, and confirm it with an authoritative source or qualified person before acting. Writing these three levels down — and deciding which level a task belongs to before you prompt — takes seconds and prevents the most common AI mistakes.

Never paste ID numbers, passwords, medical details, or identifiable student stories into public tools. Treat a prompt like a message that could be stored.

State the role, task, audience, and format. Then revise after you read the first answer — that is prompt practice, not a secret formula.