A prompt is the request you write to an AI model. The same tool can give you a generic, useless answer or a precise one that saves hours; the difference is how you write. Here you learn the skill step by step, with a builder for strong prompts and ready-made templates for MIS students.
The model can't read your mind. It only sees the words you wrote. Anything in your head that isn't on the page doesn't exist for it.
A vague request gets a generic answer. When you don't specify, the model picks the “average”: an answer that fits everyone and no one in particular.
Picture a brilliant new hire on their first day: they don't know your project, your audience, or what you mean by “a good report”. Whatever you'd explain to a new colleague, explain to the model.
A clearer prompt means fewer rewrites. One extra minute writing the request saves five rounds of “no, that's not what I meant”.
The six parts of a strong prompt
Not every prompt needs all six, but the more a task matters, the more of them you'll want. Hover the colours in the example below:
1
Context and role
Who are you, what's the situation, and who will read the result?
2
Task
A clear verb: write, summarise, compare, fix, explain… What exactly do you want?
3
Input
The material to work on: a table, text, code, an error message. Fence it off from the instructions.
4
Constraints
Length, level, what to avoid, what to assume, and any limits to respect.
5
Output format
A table? Bullets? Numbered steps? Code only? In which language?
6
Example
One example of what you want replaces a whole paragraph of explanation.
I'm a third-year MIS student working on a database project about the university library. Suggest the entities and the relationships between them based on this description: “Students borrow books, each book has copies, staff record loans and returns, and late returns get a fine.” Keep it to 5–7 entities, and don't add anything the description doesn't mention unless you explain why. Show the result as a table: entity, key attributes, primary key, relationships. Example row: Book | title, isbn, year | book_id | 1:N with Copy.
Context and roleTaskInputConstraintsOutput formatExample
Build your own prompt
Fill in the fields (or pick a ready example) and watch the prompt come together, with a check that tells you what's missing. Copy it into ChatGPT, Claude or Gemini.
Examples:
Prompt strengthWeak
○ The task is clear and specific
○ There's context: who you are and who it's for
○ Input is included and fenced off
○ Constraints are stated
○ The output format is defined
○ There's an example (a big accuracy boost)
○ No vague words like “something nice” or “anything”
Final prompt
Start with the task at least…
Techniques that make a difference
01
Be specific, use numbers
Instead of “summarise the article”, say “summarise it in 5 bullets, one line each”. Numbers close the door on guessing.
02
Explain why
“Keep it short because the talk is only 3 minutes” beats “keep it short”. When it knows the goal, it makes smarter choices in the details.
03
Fence off your data
Wrap text or tables in markers like """ or tags like <data>…</data> so the model knows where your instructions end and the material begins.
04
Give one or two examples
Want user stories in a certain shape? Write one as a sample. This is called few-shot prompting, and it's one of the most powerful techniques.
05
Break big tasks down
Instead of “do my graduation project”: first requirements, then entities, then queries. Review each step before the next.
06
Let it reason first
For problems involving calculation or logic, ask it to “work through it step by step, then give the final answer”. Errors drop noticeably.
07
Let it ask you first
Add: “If you need more information to answer well, ask me before you start.” Very useful when the task is big and you're unsure what to include.
08
Say what to do, not only what not to do
“Don't be formal” is weaker than “write in a friendly tone, as if to a classmate”. A positive instruction is clearer.
09
Ask for honesty
Add: “If you're not sure about something, say so plainly and don't invent sources.” It reduces hallucination and tells you what to verify.
10
Iterate in conversation
The first answer is a draft. Tell it exactly what to change: “point 3 is too general, give an example from a distribution company.” The conversation is part of the prompt.
Before and after: examples from your major
A SQL query
Before
Write me SQL for sales
After
I have two MySQL tables:
orders(order_id, customer_id, order_date)
order_items(order_id, product_id, quantity, unit_price)
Write a query returning total sales per month in 2025, newest first. After the code, explain each part in one line.
What changed? Added the table and column names (input), the period and ordering (constraints), and asked for an explanation (format).
System requirements
Before
Write requirements for a restaurant app
After
I'm a business analyst drafting an initial requirements document for a small restaurant's ordering app (one branch, 20 tables, delivery within the neighbourhood only).
Write 8 functional and 4 non-functional requirements, each a single testable, numbered sentence (FR-01, NFR-01).
Leave out online payment; it's out of scope for now. If anything is unclear, list your assumptions under “Assumptions”.
What changed? A clear role, size and scope, an exact count, a “testable” standard, an out-of-scope item, and assumptions instead of silent guesses.
Understanding a concept for an exam
Before
What is normalisation?
After
I have a databases exam tomorrow and I don't get normalisation (1NF, 2NF, 3NF).
Explain it with one running example: an orders table with repetition, normalised step by step up to 3NF.
Then ask me 3 short questions to check I understood, and wait for my answers before correcting them.
What changed? We stated the goal and level, asked for one running example instead of definitions, and turned the model into a tutor that tests us.
Fixing an error
Before
My code doesn't work, why?
After
This Python code averages grades and raises this error:
"""
TypeError: unsupported operand type(s) for +: 'int' and 'str'
"""
The code:
"""
[paste the code here]
"""
First explain the cause in plain words, then give the fix. I want to understand, not just copy.
What changed? Pasted the full error and code, fenced off, and set the order: cause first, then the fix.
Practice: what's missing from the prompt?
Read each prompt and pick the most important missing part.
Exercise 1
“Analyse this data and tell me what you see.” (no data attached)
Exercise 2
“I'm an ERP project manager with a meeting with senior management on Sunday about a two-week delay caused by data migration.”
Exercise 3
“Write 10 user stories for a university room-booking app in the form: As a… I want… so that….”
Exercise 4
“Compare SAP and Odoo for a 30-person company with a limited budget, focusing on cost and ease of implementation.”
Ready-made templates for MIS students
Copy a template and replace whatever is in [brackets].
Explain a concept
I'm an MIS student in [year]. Explain [concept] as you would to a classmate who hasn't studied it, with an example from a real company or shop. Then give me a 3-bullet summary and one question to test myself.
Write a SQL query
My [MySQL/SQLite] database has these tables:
[table_name(columns)]
[table_name(columns)]
Write a query that returns [exactly what you need], ordered by [column]. Explain it line by line after the code.
Test data
Give me INSERT statements for [number] sample rows in [table(columns)]. Make the data realistic and varied, with two edge cases (such as a NULL or a very old date) so I can test my queries.
User stories
I'm a business analyst on [system description]. The users are [roles]. Write [number] user stories in the form “As a… I want… so that…”, each with two acceptance criteria in Given/When/Then. Don't add features beyond this description.
ERD entities
From this description:
"""
[system description]
"""
Extract the entities, their attributes and primary keys, and the relationships (1:1, 1:N or M:N) with the reason for each. Show it as a table. For any M:N relationship, name the junction table.
Meeting notes to requirements
Here are notes from a meeting with [stakeholder]:
"""
[text]
"""
Extract the functional requirements, the non-functional requirements, and the open questions we need to take back to the client. Don't add requirements that weren't mentioned.
Interview prep
I have an interview for [job title] at [type of company]. Ask me 5 realistic interview questions, one at a time, and after each of my answers rate it and tell me how to improve it.
A formal email
Write an email to [professor/manager] asking for [request], because [reason]. Respectful and brief (under 120 words), with a subject line. Write it in [Arabic/English].
Common mistakes
Many asks in one message: “summarise, write an email, make a table and translate it”. You get mediocre results at everything. Split them.
Believing everything: a model can invent a number or a citation with full confidence. Check figures and sources, and run the code.
Pasting sensitive data: passwords, customer records or confidential company documents. Follow your organisation's policy and swap in made-up examples.
Giving up after one answer: if you don't like it, don't repeat the same request. Say exactly what's wrong.
Submitting its work as yours: many universities have an AI policy for coursework. Know it, and use AI to learn, not to skip learning.
Spotted a mistake on this page, or have an idea to improve the site?