Two master’s students ask a generative AI assistant about logistic regression. The first types “Explain logistic regression” and gets a five-paragraph lecture that is correct but reads like any textbook. The second writes: “I’m a public health master’s student who has studied linear regression. I’m reading a paper with an odds ratio of 2.3 (95% CI 1.4–3.8) for smoking. Explain what that number means in everyday language, then ask me two questions to check I’ve understood.” She gets exactly what she needs, plus a small exercise.
The difference is not a trick or a magic phrase. It is giving the assistant enough context that it does not have to guess. This article focuses on legitimate academic uses: understanding concepts, stress-testing arguments, self-testing and organising your own ideas.
Five components of a good prompt
| Component | Answers the question | Example |
|---|---|---|
| Context | Who am I, what am I doing, what do I already know? | “Second-year education PhD student, comfortable with interviews, new to statistics” |
| Task | What exactly do I need? | “Identify the three weakest steps in this argument,” not “comment on this paragraph” |
| Input | What should the assistant work on? | Your paragraph, outline or an anonymised table |
| Constraints | What must it not do? | “Do not rewrite the paragraph, only ask questions”; “do not invent references” |
| Format | What should the output look like? | “A three-column table”, “under 150 words”, “bullet points” |
The most neglected component is constraints, and it is the one that keeps you on the right side of integrity rules. A prompt that says “do not write this for me, question me instead” turns the assistant from a ghostwriter into a tutor.
Six templates for common situations
1. Understanding a difficult concept
“I am [context]. Explain [concept] using an example from [your field]. Then point out one common misunderstanding about it. Finally, ask me one question to test my understanding and wait for my answer before commenting.”
2. Stress-testing an argument
“Here is the main claim of the chapter I’m writing: [paste]. Act as a sceptical reviewer and give the three strongest objections. Do not edit or rewrite my text.”
3. Self-testing after reading
“I have just read a chapter on [topic]. Ask me five questions of increasing difficulty, one at a time. After each answer, tell me where I went wrong and why.”
4. Organising scattered notes
“These are my notes from a pilot interview: [paste]. Group them into clusters and give each a provisional label. Do not add any idea that is not in my notes.”
5. Preparing for a supervisor meeting
“I’m meeting my supervisor about [issue]. Here is where things stand: [summary]. Help me draft three specific questions, each leading to a decision.”
6. Checking analysis code you wrote
“Here is my R code to [purpose]. Explain what each line does and point out anything that could be wrong. Do not rewrite it all.”
Iterate: the first answer is rarely the best
Treat each session as a conversation, not a lookup. Useful follow-ups:
- “Make it half as long and drop the introduction.”
- “That example doesn’t fit my field; use one from [context].”
- “How confident are you about the second point? What could make it wrong?”
- “Summarise what I misunderstood during this exchange.”
The confidence question matters most. Generative AI assistants sound equally sure whether they are right or wrong; asking what would make a claim false tells you where to check.
Requests to avoid
- “Write my literature review on…”: over the integrity line at most institutions, and prone to fabricated sources.
- “Give me ten references on…”: chat assistants can produce sources that do not exist. Use a real literature search tool.
- Pasting personal data or a manuscript someone sent you for review.
- Leading questions: “Explain why my method is the best” invites flattery. Ask “What is the biggest weakness of this method?”
Which language should you prompt in?
Either works, with a few practical caveats. For technical concepts, ask the assistant to give the English term alongside any translation, since terminology in other languages is often inconsistent and you will need the original term to search the literature. If English is not your first language and you are working on an English paragraph, you can ask for feedback in your own language to understand it properly, then make the edits yourself in English. And do not trust local specifics, such as national regulations, agency names or statistics, without checking an official source; that is where assistants go wrong most often.
Keep a prompt log
For work feeding into a thesis or paper, save important prompts and responses (many tools let you export chat history). The log helps you write an accurate AI disclosure, demonstrate your process if questions arise, and reuse prompts that worked. A plain text file noting the date, purpose, prompt and how you used the output is enough.
A 20-minute exercise
- Pick a concept you are hazy on in a course or your project.
- Write a prompt with all five components, using template 1.
- Answer the assistant’s check question without looking anything up.
- Follow up with “What else have I misunderstood?”
- Write three sentences summarising the concept in your own words, then compare them with your textbook.
The last step matters most: knowledge becomes yours only when you can restate it without looking at the screen. Save the prompt you just wrote in your log as a template for next time.
Câu hỏi thường gặp
How do I write a good prompt for studying with AI?
Include five components: your context, a specific task, the input, constraints on what it must not do, and the output format. Then iterate with follow-up questions rather than accepting the first answer.
Should I ask an AI assistant for references?
No. Chat assistants can produce plausible references that do not exist. Use a proper literature search tool and open the original before citing anything.
How can I use AI without breaking academic integrity rules?
Use it to understand, question, self-test and get critique, and state in your prompt that it must not write for you. Always check your course or institution’s policy and disclose use where required.
Should I save my AI chat history?
For anything feeding into a thesis or paper, yes. It helps you write an accurate AI disclosure and show your process if questions arise.
Why are AI answers often so generic?
Usually because the prompt lacks context and a specific task. Saying who you are, what you know, what you need and what format you want improves answers considerably.
