Use a five-part prompt structure
A useful starting framework is: goal, context, audience, output format and constraints. State the job first, provide the information the model needs, describe who the result is for, tell it how to organise the answer, and name anything it should avoid. Add an example when tone or structure matters.
For instance, “Write a YouTube description” leaves many details unresolved. A stronger prompt might specify the topic, intended audience, video length, tone, key facts that must appear, approximate length and whether the result should include a call to action. The model then has fewer important assumptions to invent.
Goal: [what I need]. Context: [relevant facts]. Audience: [who will use it]. Format: [bullets/table/steps]. Constraints: [length, tone, exclusions]. Quality check: list assumptions or uncertain claims before finishing.
Give context, not private information
Context improves relevance when it describes the task, reader, materials and desired outcome. But a prompt does not need your password, school ID, private messages or complete personal record. Replace identifiable details with placeholders, and only provide material you have permission to share.
If the task depends on a document or syllabus, give the necessary excerpt or upload only the relevant section. Tell the model which source to prioritise. A clear source boundary can reduce unsupported additions, though you still need to compare the response with the source.
Ask for a useful format
Output format matters because the answer should be easy to check and use. Ask for numbered steps when doing a task, a comparison table when choosing options, a short summary before the detail, or a checklist when you need to execute a plan. For code, specify the language, runtime and expected behaviour.
You can also set length and reading level. “Explain this in simple language for a beginner, then give one worked example and three practice questions” is clearer than “explain simply”. If the answer is too long, ask for a concise version while preserving the important steps.
Iterate instead of starting over
Treat the first response as a draft. Identify one concrete weakness at a time: missing examples, incorrect audience, too formal, too much detail or unsupported claims. Then request a targeted revision. This makes it easier to understand whether the change improved the response.
For a complex task, split the work. Ask for an outline, review it, then develop one section at a time. Ask the model to state assumptions and list points that need verification. It is safer to verify important facts yourself than to ask the model whether its own answer is correct.
Check accuracy and originality
AI models can produce incorrect dates, invented references, flawed arithmetic and confident wording that hides uncertainty. Verify important claims with original sources, particularly for exams, laws, medical information, finance and technical instructions. If a citation is supplied, open it and confirm that it supports the claim.
When using AI for schoolwork, follow your teacher’s rules. Use it to practise, get feedback, generate questions or clarify a concept rather than submitting an output as if it were entirely your own when that is not allowed. Add your own reasoning and keep the sources you relied on.
Prompt examples by task
Study: “Teach me the difference between speed and velocity. Use one everyday example, one formula and three practice questions. Do not show the answers until after the questions.” This gives a learning objective and a structure for practice.
Creator work: “Suggest five short-video concepts for beginners interested in phone photography. For each, give a truthful hook, a three-scene outline and one production tip. Avoid exaggerated promises.” This produces options that can be evaluated against real production constraints.