Crafting Effective Prompts

You ask ChatGPT to write something, and what comes back is technically fine and completely unusable. It is too long, too formal, full of phrases you would never say, and it could have been written for anyone. So you either rewrite it yourself or decide AI is overrated.

The problem is almost never the model. It is that the prompt left most of the decisions to the model, and when a language model has to guess, it guesses the most average answer it can. This lesson gives you a simple framework for making those decisions yourself, in a sentence or two, before you hit Enter.

Why vague prompts get generic answers

In the last lesson you saw that the model predicts text based on everything in the conversation. A short, vague prompt gives it almost nothing to work with, so it draws on the broadest, most common patterns it knows. That is where the bland, one-size-fits-all tone comes from.

Every detail you add narrows the field. Who is this for? How long should it be? What should it sound like? What do you already know? Each answer removes a guess, and fewer guesses means output that fits your situation instead of everyone's.

The four pillars of a good prompt

You do not need a template for every task. You need four habits. Most strong prompts use all four; even adding one or two makes a visible difference.

1. Be specific

Say exactly what you want: the topic, the angle, the length, the tone.

Vague:

Write about marketing.

Specific:

Write a 200-word LinkedIn post about why small businesses should
invest in email marketing. Use a conversational tone.

The second version names the platform, the length, the argument, the audience, and the voice. The model no longer has to invent any of them.

2. Provide context

The model knows nothing about your situation unless you tell it. Context is the background only you have: what you are working on, what you have tried, what the constraints are.

Vague:

Fix this code.

With context:

I'm building a [TYPE OF APP] and getting this error when the page
first loads: [PASTE ERROR]. It works after I refresh. Here's the
code: [PASTE CODE]. What's causing this, and how do I fix it?

The same rule applies outside of code. "Help me reply to this email" is weaker than "Help me reply to this email from a longtime customer who is upset about a late delivery. I want to apologize without promising a refund."

3. Set the format

If you want a table, ask for a table. If you want three bullet points, say three. If you want it short enough to read on a phone, say so. Format is the cheapest pillar to add and one of the most noticeable.

Vague:

Give me ideas for my business.

Formatted:

Give me 5 marketing ideas for [TYPE OF BUSINESS] in [AREA].
For each one, include: a short name, who it's aimed at, and the
first step I'd take this week. Format as a numbered list.

4. Define the role

Telling the model to "act as" a specific role is one of the fastest ways to improve an answer. It does not turn the model into an expert, and it does not give it any new knowledge. What it does is filter which patterns the model draws from, which produces more targeted responses. "Act as a bookkeeper for small retail shops" pulls toward the vocabulary, priorities, and concerns of people who actually do that job, instead of an average of everyone.

Vague:

How should I organize my finances?

With a role:

Act as a bookkeeper who works with [TYPE OF BUSINESS] owners.
I'm a [ONE-LINE DESCRIPTION OF YOU AND YOUR SITUATION].
What's a simple weekly routine for keeping my records organized?

Keep the role specific and ordinary. "A bookkeeper for sole traders" works better than "a world-class financial genius." And remember that a role changes the framing, not the facts. For money, health, or legal questions, the answer is a starting point for a conversation with a qualified professional, not the conversation itself.

Worked example: a follow-up email

Here is all four pillars working together.

Before:

Write an email to a client.

You will get a generic, slightly stiff email with placeholder names and nothing specific to say.

After:

Act as an account manager at a small [TYPE OF COMPANY].

Write a follow-up email to [CLIENT NAME], who attended our product
demo yesterday. Thank her for coming. Mention the three things she
seemed most interested in: [BENEFIT 1], [BENEFIT 2], and [BENEFIT 3].
Suggest a 15-minute call on [DAY].

Keep it under 150 words. Friendly but professional. No exclamation
points.

Look at what each part is doing. The first line is the role. The second paragraph is specificity and context: who, what happened, what to mention, what to ask for. The last paragraph is format and tone. None of it took long to write, and almost none of it could have been guessed by the model.

The first answer is a draft

You should never assume the first response is the final one. The best results come from a short back-and-forth, the same way you would work with a new assistant.

Useful follow-ups are short and direct:

Make it shorter and cut the opening sentence.

That's too formal. Rewrite it the way I'd say it out loud.

Good. Now give me two alternative subject lines.

What information would you need from me to make this better?

That last one is worth remembering. When you are not sure what context to add, ask the model what it is missing. It will often list exactly the details you left out.

Common mistakes

Asking for everything at once. A single prompt that asks for a business plan, a logo idea, and a marketing calendar will do all three badly. Break big jobs into steps and handle them one at a time.

Leaving out the details that matter. The model cannot see your notes, your customer, or your deadline. If it matters, put it in the prompt.

Stopping after one try. A mediocre first answer is information, not failure. Tell it what is wrong and ask again.

Skipping the proofread. A well-written answer can still contain wrong facts, invented details, or promises you cannot keep. Read everything before it leaves your hands.

Try this

Set a timer for 15 minutes.

Step 1. Pick one real task from your week: an email you need to write, a post you need to publish, a question you have been putting off.

Step 2. Write the lazy version of the prompt, the one you would normally type. Run it and save the answer.

Step 3. Rewrite it using all four pillars. Check your prompt against this list:

• Specific: does it say what, how long, and in what tone?
• Context: does it include the background only you know?
• Format: does it say what shape the answer should take?
• Role: does it name a specific, realistic role?

Step 4. Run the new prompt, then send at least two follow-ups to refine it.

Step 5. Compare the final result with the lazy version. Write down the one pillar that made the biggest difference for this task. That is the habit to build first.

If you want to see how much the model matters versus the prompt, try your improved prompt in the Study AI Mastery Playground's compare mode on two or three models side by side.

What to remember

• Vague prompts get generic answers because the model fills every gap with the most average guess.
• The four pillars of a good prompt: Be Specific, Provide Context, Set Format, Define Role.
• "Act as" a role filters which patterns the model draws from, producing more targeted responses; it does not add knowledge or make facts true.
• Never accept the first response by default. Iterate with short, direct follow-ups.
• When you are unsure what to add, ask the model what information it needs from you.