Lesson 1

Why Simple Prompts Fail — and What to Do About It

You've probably had this experience: you type a perfectly reasonable request into ChatGPT, Claude, or Gemini — something like "Write a marketing email" — and the result is a generic, fluffy paragraph that could apply to any business on Earth. You ask for a follow-up, and it gets worse. You feel like you're arguing with a very confident intern who never quite understands the assignment.

This isn't a flaw in the AI. It's a flaw in the prompt. Large language models are not mind-readers. They are next-token predictors trained on billions of documents. When you give them a vague instruction, they have no reason to produce a specific output — so they produce the most statistically average output, which is exactly what you're seeing: bland, generic, and useless.

Prompt engineering is the discipline of writing instructions that eliminate ambiguity. It's not about "tricking" the AI or using magic phrases. It's about giving the model the same level of detail you'd give a competent human contractor — before they start working. This chapter will show you exactly why simple prompts fail, what a good prompt contains, and how to write one step by step.

The Anatomy of a Failed Prompt

Let's look at a real example. Suppose you're a product manager at a SaaS company called FlowDesk, which sells customer-support software. You need a short email to announce a new feature: an AI-powered ticket triage system.

Here's a simple prompt:

Write an email announcing our new feature.

What does the model produce? Something like this:

"We are excited to announce a new feature that will help your team work more efficiently. This update brings powerful capabilities to streamline your workflow..."

This is useless. It doesn't mention FlowDesk, the feature name, the benefit, or the call to action. The model had to guess everything — the product, the audience, the tone, the length, the goal — and it guessed wrong.

Now consider this prompt:

Write a 150-word email to FlowDesk's existing customers announcing our new AI-powered ticket triage feature. The feature automatically categorizes incoming support tickets by urgency and routes them to the right team. Tone: confident but not salesy. Goal: get customers to try the feature via the in-app banner. Include a subject line and a single call-to-action button labeled "Try It Now".

The difference is night and day. The second prompt gives the model context (who the audience is), specificity (what the feature does), constraints (word count, tone, goal), and format (subject line, CTA button). The output will be usable, often with minor edits.

This is the core lesson: a simple prompt fails because it delegates all decisions to the model's statistical average. A good prompt makes those decisions for it.

The Three Pillars of a Good Prompt

Every effective prompt — whether it's for writing, coding, analysis, or translation — rests on three pillars. If you're missing one, your output quality drops.

1. Clarity

Clarity means the model knows exactly what you want it to do. Use imperative verbs: "Write," "List," "Compare," "Debug," "Translate." Avoid ambiguous phrasing like "Can you help me with..." or "I need something about..." — these invite the model to guess your intent.

Clarity also means defining your terms. If you say "a short story," the model doesn't know if "short" means 100 words or 1,000. Say "a 300-word story." If you say "a professional tone," the model doesn't know if that means corporate jargon or plain language. Say "formal but accessible, no jargon."

2. Context

Context is the background information the model needs to produce relevant output. This includes:

• Who you are (your role, your industry)
• Who the audience is (their knowledge level, their needs)
• Why you need this (the goal, the use case)
• What the model should assume (facts, constraints, prior decisions)

For example, "Write a product description" is weak. "Write a 50-word product description for a noise-cancelling headphone aimed at frequent flyers, emphasizing battery life and comfort over technical specs" is strong. The model now knows the audience, the selling points, and the length.

3. Specificity

Specificity is about constraints. The more constraints you give, the less room the model has to drift. Specify:

• Length (word count, character count, number of paragraphs)
• Format (bullet points, table, JSON, markdown, email, script)
• Structure (introduction, body, conclusion; or problem, solution, benefit)
• Style (tone, voice, reading level)
• Exclusions (what NOT to include)

Here's a practical example. Suppose you want a Python function to calculate compound interest. A simple prompt:

Write a Python function for compound interest.

You'll get a basic function, but it might not handle edge cases, might use floating-point errors, or might not have a docstring. A specific prompt:

Write a Python function called calculate_compound_interest that takes principal (float), annual_rate (float, as a percentage), years (int), and compounds_per_year (int, default 12). Return the final amount rounded to 2 decimal places. Include a docstring and handle the case where principal is negative by raising a ValueError. Use the formula A = P(1 + r/n)^(nt).

Now the model knows the function name, the parameters, the default values, the error handling, the formula, and the output format. The code will be correct and production-ready.

Step-by-Step: Building a Prompt from Scratch

Let's walk through a real workflow. You're a freelance writer who needs a blog post outline about remote work productivity. Here's how you'd build the prompt, layer by layer.

Step 1: Start with the task.

Create a blog post outline.

Step 2: Add the audience and purpose.

Create a blog post outline for mid-level managers at tech companies who are struggling with remote team productivity.

Step 3: Add structure and length.

Create a blog post outline for mid-level managers at tech companies who are struggling with remote team productivity. The outline should have 5 main sections, each with 3-4 sub-bullets. Target length: 1,500 words.

Step 4: Add tone and constraints.

Create a blog post outline for mid-level managers at tech companies who are struggling with remote team productivity. The outline should have 5 main sections, each with 3-4 sub-bullets. Target length: 1,500 words. Tone: practical and empathetic, not preachy. Avoid generic advice like "communicate more." Focus on specific tactics like async standups and time-blocking.

Step 5: Add a format requirement.

Create a blog post outline for mid-level managers at tech companies who are struggling with remote team productivity. The outline should have 5 main sections, each with 3-4 sub-bullets. Target length: 1,500 words. Tone: practical and empathetic, not preachy. Avoid generic advice like "communicate more." Focus on specific tactics like async standups and time-blocking. Format the output as a markdown list with H2 headings for each section.

Now you have a prompt that will produce a genuinely useful outline. Notice how each step added a constraint that reduced ambiguity.

Common Mistakes

Expert Tip

Your Practice Task

Here's a 15-minute exercise to cement what you've learned.

Task: Write a prompt that generates a short story. Then improve it.

Part 1 (2 minutes): Write a simple prompt: "Write a short story." Run it in any AI tool (ChatGPT, Claude, Gemini). Note the output's weaknesses: Is it generic? Does it lack a clear protagonist? Is the ending weak?

Part 2 (10 minutes): Rewrite the prompt using the three pillars. Include:

• Clarity: Specify the genre (e.g., science fiction), the length (e.g., 400 words), and the point of view (e.g., first person).
• Context: Give the setting (e.g., a Mars colony in 2147), the protagonist (e.g., a botanist who discovers a plant that grows in Martian soil), and the conflict (e.g., the colony's leadership wants to destroy it).
• Specificity: Require a twist ending, a specific tone (e.g., melancholic but hopeful), and a title.

Part 3 (3 minutes): Run your improved prompt. Compare the two outputs. You should see a dramatic difference in quality, specificity, and emotional impact.

Self-verification: Your improved story should have a named protagonist, a clear setting, a conflict that escalates, and a twist you didn't explicitly request in the prompt. If it doesn't, add more constraints and try again.

This exercise takes less than 15 minutes and will teach you more about prompting than reading ten articles. The key takeaway: every detail you add is a decision the model doesn't have to guess. And fewer guesses mean better output.