Prompting Fundamentals

Introduction

Most people meet AI by typing whatever comes to mind and hoping. Sometimes it works. Often the answer is vague, generic, or subtly wrong — and it is easy to conclude the tool is not very good.

Usually the tool is fine. It simply was not told enough to do the job well. This module is about the difference between asking and briefing, and about the short loop that turns a mediocre first answer into a useful one.

You will not need an account, a subscription, or a single line of code. You will need one small task you actually care about.

By the end of this module you will be able to take a request that produced a weak answer and say precisely why — then fix it.

What a Prompt Actually Is

A prompt is not a question. It is a brief: you hand over a task, plus what is needed to do that task well.

The difference matters because a search engine and an assistant do different jobs. A search query is a set of keywords used to find documents that already exist — you do the reading, the judging, and the writing. A prompt asks something to be produced for you: a summary, a draft, an explanation, a plan. Producing something requires knowing what it is for.

Think about how you would hand the same task to a capable new colleague. You would not walk up and say "write an email." You would say who it is going to, what needs to happen, and roughly how it should sound. None of that is complicated. It is simply the information the work requires — and a prompt is where you supply it.

• •A search query finds things that exist. A prompt produces something new.
• •A brief carries the task and what is needed to do it.
• •If a person could not do the task from your words alone, an assistant cannot either.

Why Vague Requests Produce Vague Results

When information is missing, an assistant does not stop and ask. It fills the gap with the most ordinary assumption available — and ordinary assumptions produce ordinary output.

Compare these two requests for the same job:

1. "Write a summary of this report." — Missing: who reads it, how long, what they need to decide. The likely result is a competent, shapeless paragraph you will rewrite yourself.
2. "Summarize this report in five bullets for our operations lead, who needs to decide whether to extend the supplier contract. Lead with anything that affects cost or delivery risk." — Same report, and now the answer has a job to do.

The second request is not longer for the sake of it. Every extra phrase removes a guess the assistant would otherwise have made for you.

Start With Intent — Know What You Want

Intent is the outcome you are actually after, and it is the step most people skip. It is not the same as the task. The task might be "write an email"; the intent might be "they agree to a new time and do not feel messed around."

Naming the intent first does two things. It tells you what to put in the prompt — because you can now see which details serve that outcome. And it gives you something to judge the answer against, which is what makes the next loop possible. Without a stated intent, "is this good?" has no answer beyond taste.

This takes about fifteen seconds and it is the highest-leverage habit in this course. Before you type: what would a good result let me do?

• •Task: what should be produced.
• •Intent: what should be true once it exists.
• •Write the intent down, even roughly. You will use it twice — once to write, once to judge.

The Parts of a Useful Prompt

Useful prompts tend to draw on the same seven parts. Treat this as a menu you choose from, never a form you fill in.

• •Intent — what am I actually trying to achieve?
• •Task — what exactly should the assistant do?
• •Context — what does it need to know that it cannot infer?
• •Audience — who is this for?
• •Output — what form should the result take?
• •Constraints — what are the limits: length, tone, what to avoid?
• •Examples — what does "good" look like?

The right structure depends on the task. A quick rewrite needs intent and task. A client email needs audience and tone. A summary needs output and constraints. Forcing all seven into every prompt just replaces vague requests with bloated ones.

Your First Result Is Information, Not a Verdict

Almost nobody gets the prompt right first time, and that is not a failure — it is how the process works. The first response shows you what was understood and what was not. That is evidence, and it is the input to your next prompt.

The loop is short: ASK → READ → JUDGE → REFINE.

1. ASK — send your brief.
2. READ — actually read it, rather than skimming for something that looks plausible.
3. JUDGE — hold it against the intent you wrote down. Did it do the job? What is missing or wrong?
4. REFINE — change something specific in the prompt to address what you found, and ask again.

Reading and judging are separate steps on purpose. Skimming and accepting is the most common way people end up with mediocre output and blame the tool. And pressing send again without changing anything is not refinement — it is a coin toss.

You Are Still the One Deciding

A clearer prompt improves the probability of a useful result. It does not guarantee correctness. Assistants can state wrong things fluently and confidently, and a well-written brief does not stop that — it only makes it less likely and easier to spot.

So the last step never moves: you decide. You check anything that matters — names, numbers, dates, quotes, claims about people, anything you would be embarrassed or liable to get wrong. The assistant drafts; the responsibility stays with you.

One safety habit belongs here too, because it comes up immediately in practice.

• •A clearer prompt raises the chance of a useful result. It is not a guarantee.
• •There is no perfect prompt, and no phrasing that removes the need to check.
• •The assistant does not know what you meant but did not say.
• •Important decisions and verification stay with you.

Do not paste passwords, API keys, card or account numbers, confidential work information, other people's private details, or regulated data into an AI tool. A genuine task rarely needs them — and if you think it does, that is a signal to change the task, not to paste.