Lesson 8

Why This Matters: The Difference Between Asking and Getting

You have likely already used a chatbot or an AI image generator. You typed something, got a result, and maybe felt one of two things: either "this is magic" or "this is useless." The truth is neither. The output you get from a large language model (LLM) like ChatGPT, Claude, or Gemini is not random. It is a direct, deterministic response to the exact text you provided. The model has no intent, no hidden agenda, and no ability to read your mind. It only has your prompt.

This is the single most important skill you will learn in this entire course: prompting is not typing a question. Prompting is writing a specification. When you ask a colleague to do a task, you rely on shared context, tone of voice, and their ability to ask clarifying questions. An LLM has none of that. It takes your words literally, and it will happily produce a confident, well-structured, completely wrong answer if your prompt is ambiguous.

Here is a real problem you will face within your first week of using AI: you ask for something simple, like "write an email to my landlord," and the model produces a generic, passive-aggressive draft that sounds nothing like you and misses the key fact that your rent is late by three days, not two weeks. You then spend more time editing the AI's output than you would have spent writing the email yourself. That is not a failure of AI. That is a failure of prompting. This chapter fixes that.

The Anatomy of a Good Prompt

Professional prompt engineers — a real job title that now exists at companies like Anthropic and OpenAI — break a strong prompt into four components. You do not need to use all four every time, but you should be aware of them:

• Specificity: Exact numbers, names, dates, and formats. Not "a few" but "exactly three." Not "recently" but "in the last 30 days."
• Context: The background information the model needs to understand the situation. Who are you? Who is the audience? What has already happened?
• Constraints: Limits on length, tone, format, or content. "Under 150 words," "no jargon," "bullet points only."
• Structure: The order and format of the output. "First give me a summary, then a list of risks, then a recommendation."

When you omit any of these, the model fills the gap with its own default assumptions. Those defaults are generic, because they are averaged over millions of users. That is why you get generic output.

Worked Example: The Vague Prompt vs. The Specified Prompt

Let us use a realistic scenario. You are a small business owner who sells handmade leather wallets on Etsy. You want to use AI to write a product description. Here is the vague prompt most beginners would type:

Write a product description for my wallet.

Here is what the model will likely produce: a generic paragraph about "a stylish and durable wallet, perfect for any occasion, with multiple card slots and a sleek design." It will be 80 words, use the word "perfect" twice, and could apply to any wallet sold by any of the 4 million sellers on Etsy. It is useless.

Now here is a properly specified prompt:

Write a product description for a handmade leather wallet I sell on Etsy.

Context: I am a one-person shop. The wallet is made from full-grain vegetable-tanned leather, hand-stitched with waxed thread. It holds 6 cards and has a hidden cash pocket. It measures 4.5 x 3.5 inches when folded. It comes in brown and black. My target customer is a man aged 30-50 who values craftsmanship and is tired of mass-produced accessories.

Constraints:
- Maximum 120 words.
- Tone: warm, confident, not salesy. No exclamation marks.
- Mention the vegetable-tanned leather and the hand-stitching specifically.
- End with a single sentence about the 2-year warranty I offer.

Format: Three short paragraphs. No bullet points.

Run both prompts in ChatGPT or Claude and compare. The second one will give you a description you could paste into your Etsy listing with only minor edits. The first one gives you a template you must rewrite entirely. This is not a subtle difference. It is the difference between a tool and a toy.

Step-by-Step: Rewriting a Weak Prompt

Let us walk through the exact process of upgrading a prompt, using a different example. Suppose you want AI to help you plan a weekly meal schedule. Your first instinct might be:

Give me a meal plan.

That will produce a generic list of chicken, rice, and salad. Here is the step-by-step upgrade process you should follow every time:

Step 1: Add the "Who" and "Why"

Tell the model who this is for and what problem it solves. Write: "I am a single person living alone, cooking for one. I work 9-to-6 and have 30 minutes max to cook on weeknights. I want to reduce food waste and spend under $60 per week on groceries."

Step 2: Add the "What" — Specific Constraints

Now add hard limits. Write: "I need 5 dinners, Monday to Friday. Each dinner must take under 30 minutes of active cooking time. I do not eat pork. I have a rice cooker, an air fryer, and a standard oven. I prefer recipes with fewer than 8 ingredients."

Step 3: Add the "How" — Output Format

Tell the model exactly how to structure the answer. Write: "For each day, give me: (1) the dish name, (2) a one-line description, (3) the full ingredient list with estimated quantities, (4) step-by-step cooking instructions in numbered form, (5) an estimated total cost. Put all of this in a table."

Step 4: Combine and Test

Here is the full prompt you would paste into ChatGPT, Claude, or Gemini:

I am a single person cooking for one. I work 9-to-6 and have 30 minutes max to cook on weeknights. I want to reduce food waste and spend under $60 per week on groceries.

I need 5 dinners, Monday to Friday. Each dinner must take under 30 minutes of active cooking time. I do not eat pork. I have a rice cooker, an air fryer, and a standard oven. I prefer recipes with fewer than 8 ingredients.

For each day, give me:
1. The dish name.
2. A one-line description.
3. The full ingredient list with estimated quantities.
4. Step-by-step cooking instructions in numbered form.
5. Estimated total cost.

Put all of this in a table.

This prompt is 110 words. It takes 90 seconds to write. It will produce a meal plan that is actually usable, with a shopping list you can take to the store. The vague version takes 5 seconds to write and produces nothing usable. The time you "save" by writing a short prompt is lost tenfold when you edit the output.

Asking for Step-by-Step Reasoning

One of the most powerful techniques for beginners is to ask the model to show its work. LLMs are trained to produce final answers, but they can also produce intermediate reasoning. This is not the same as the model "thinking" — it is the model generating text that mimics reasoning, but for your purposes, the result is the same: you get a traceable, verifiable answer.

Add this phrase to any prompt where the answer involves logic, math, or decision-making:

Before giving your final answer, list the steps you took to reach it. Then give the final answer.

For example, if you ask "Should I buy a used car with 80,000 miles or a new car that costs twice as much?" the model will give you a generic pros-and-cons list. If you add the step-by-step instruction, it will first list the assumptions it is making about your budget, your driving habits, and your maintenance tolerance — and then you can correct those assumptions. This turns a one-shot answer into a conversation.

Common Mistakes Beginners Make

Your Practice Task

This task takes under 15 minutes and you can verify the result yourself.

Task: Open any LLM of your choice (ChatGPT, Claude, Gemini, or any other). Do the following:

1. Type this exact prompt: Write a short email to my boss asking for a day off next week. Copy the output.
2. Now rewrite the prompt using the four components from this chapter. Use these specifics: You need Thursday off. You are requesting it 5 days in advance. You have already completed your major deadline. You want to suggest a specific colleague who can cover your afternoon meeting. The email must be under 80 words and use a professional but warm tone.
3. Paste your new prompt into the same model. Copy the output.
4. Compare the two outputs. Count the number of sentences in the first output that you would have to rewrite completely. Then count for the second output.

Self-verification: If your second prompt produced an email that you could send with zero or one minor edit, you have succeeded. If you still had to rewrite more than one sentence, go back and add more constraints — specify the exact date, the exact reason, and the exact tone. Then run it again.

This is the loop you will use for the rest of your life with AI: write, test, identify the gap, add a constraint, test again. It is not magic. It is engineering.