How to Use ChatGPT Effectively: Better Prompts, Better Results
The secret is not a magic phrase. A useful ChatGPT request sets a goal, supplies relevant details and gives you a way to check the answer. Here is how to put that into practice.

Ask ChatGPT to 'write an email' and you will usually get an email. Whether it is the one you needed is another matter.
The difference between a generic response and a useful one often comes down to four details: who it is for, what you want to achieve, what constraints apply and what a good answer should look like. You do not need to learn a specialist language to provide them.
Start with the job, not a fashionable prompt formula
Consider this request: 'Write a follow-up email.' It leaves ChatGPT guessing about the recipient, subject and tone.
Try something more specific:
I sent a proposal to a small business client last Tuesday and have not heard back. Draft a friendly follow-up email under 120 words. Do not sound annoyed or imply a deadline. End with one clear question about their availability for a short call.
Now the model has a real task to solve. The goal is follow-up, the reader is a client, the tone is friendly and the format has a limit. It can still make a poor choice; you can correct it without starting over.
The five-part prompt that works for ordinary tasks
When a request matters, include the information below in natural language. You can write a paragraph rather than filling in a form.
- Task: What should ChatGPT do — compare, explain, revise, plan or calculate?
- Context: Who is this for, and what facts or documents matter?
- Constraints: Give a budget, length, deadline, location, reading level or exclusions.
- Output: Ask for a checklist, table, draft, worked example or short answer.
- Quality check: Say what would make the result useful, and ask it to flag missing information.
For example: 'I am choosing a laptop for university. My budget is £800, I carry it daily and need good battery life. Explain the trade-offs, show a compact comparison and tell me which specifications I should check before buying. Do not invent prices.'
Ask follow-up questions instead of restarting
A good first answer is often a working draft. Tell ChatGPT exactly what missed the mark: 'Keep the second paragraph, shorten the introduction and replace the jargon with plain English.' For an analysis, try 'Which assumption would change your conclusion?' or 'What important information is still missing?'
Breaking a complicated task into stages can also help. For a report, ask for an outline first, then the evidence needed for each section, then a draft. Review each stage before moving on. This makes errors easier to spot.
Give examples, especially for writing
If you want a distinctive voice, show a short sample that represents it and explain what you like. 'Use short sentences and concrete examples, like this paragraph' is more useful than 'make it sound human'. Be clear about who owns any example you upload and avoid sharing private customer material without permission.
For editing, specify what must stay untouched: figures, quotations, legal wording, names or links. ChatGPT may rewrite something it should have preserved unless you say otherwise.
Use documents and search deliberately
For a contract, spreadsheet or long report, attaching the document — when the feature is available on your account — gives ChatGPT something specific to work from. Ask it to identify the relevant section and distinguish a quotation from its interpretation. For current information, use web search where available and inspect the original links.
Do not assume a source is genuine because a fluent answer lists one. University library guidance and OpenAI's own documentation both recommend checking important facts and verifying citations. This matters especially for health, legal, financial and breaking-news questions.
Set recurring preferences once
If you repeatedly ask for British English, concise introductions or explanations aimed at beginners, use Custom Instructions in ChatGPT's Personalisation settings when available. For an ongoing piece of work, a Project can keep relevant chats, files and instructions together, depending on your plan and workspace setup. You can still override preferences for an individual task.
These are conveniences, not guarantees that ChatGPT will be correct. Keep checking what it produces.
Protect your information before you paste it
Remove passwords, bank details, private identification numbers and other sensitive data unless you have a clear, authorised reason to share them. Review Settings → Data controls to understand your available model-improvement choices. Temporary Chat offers different history and memory behaviour; check the current description before using it for sensitive work.
Three prompts you can reuse
Learn a topic: 'Explain compound interest to a beginner. Start with a £1,000 example, show the arithmetic and then give me three questions to check my understanding.'
Improve a draft: 'Edit this introduction for clarity. Keep all factual claims and names unchanged. Show the revised paragraph, then list any claims that need verification.'
Make a decision: 'Compare these two options against cost, time and maintenance. Tell me what information you need before making a recommendation. Do not assume missing figures.'
Common questions
Is a longer prompt always better? No. Relevant details improve the task; unrelated instructions can bury it. Start clear, then refine.
Can ChatGPT research facts for me? It can help research when search tools are available, but you should inspect reliable source material for consequential claims.
What is the single best habit? Ask for an answer you can verify: a clear method, stated assumptions and the sources or figures behind important conclusions.
Further reading: OpenAI's prompting guide and Penn State University Libraries on checking AI output.