AI Prompt Book

Overview

If your AI outputs read like a messy chat log - half ideas, missing details, wrong format - that’s not “bad AI.” It’s usually a prompt structure problem. This chapter gives you 10 core prompt patterns you can reuse for writing, images, and practical tasks, with a consistent mini-format so you can spot what’s missing fast.

You’ll learn how to pair role + task framing, add context + constraints, force output formatting, control assumptions, and then tighten results through iterative refinement. Each item includes a quick “what breaks” and a “do this instead” you can copy.

Takeaway check: Ask yourself: when you paste a prompt, do you tell the AI what you want, who it’s acting as, and how you’ll use the result? If any of those are fuzzy, these patterns will fix it.

The Breakdown

#1: Role + Job-to-be-Done (Not Just “You Are…”)

Problem: Prompts that only say “You are helpful” or “You are an expert” often produce generic answers. You get the right topic but not the right deliverable - like advice when you needed a checklist, or a summary when you needed exact steps.

Solution: Start your prompt with a clear role and a single job. Example structure: “Act as a [role] who must [deliverable].” Then name the target: “for a [specific user]” and “to achieve [specific outcome].”

Result: The AI locks onto a useful format instead of wandering into broad explanations.

#2: Task Framing with a Deliverable Spec

Problem: “Write a blog post about lead generation” is too open, so the output may miss length, tone, or sections. You end up editing for 30 minutes because the model guessed wrong.

Solution: Specify the deliverable like you would for a contractor. Include: type (email, landing page, caption), audience, goal, and what must be included (e.g., 3 benefits, 1 CTA, 5 FAQs). Keep it short but concrete.

Result: You get an output you can publish or reuse with minimal rework.

#3: Context That Matters (Include the Inputs, Exclude the Noise)

Problem: Without context, the AI invents details - then your facts don’t match reality. With too much context, it misses the key constraint buried under paragraphs.

Solution: Add only the inputs the model needs to be accurate: product/service name, audience, examples you like, and any “must-not” facts. Use a small block like:

• What I’m selling: …
• Who it’s for: …
• What I already have: …
• What can’t be said: …

Result: Outputs align with your actual situation instead of drifting into guesses.

#4: Constraints as Guardrails (Length, Tone, Must/No Must)

Problem: If you don’t set constraints, you’ll get outputs that don’t fit your workflow - too long for a caption, too formal for a text thread, or missing the one thing your client needs.

Solution: Add measurable constraints. Example: “Write 120-160 words, friendly tone, no jargon, include one question, end with a CTA.” For images: “Use a 1:1 square crop, high-contrast lighting, avoid readable text.”

Result: You receive results that match the exact format you need.

#5: Output Formatting Rules (Force Structure the First Time)

Problem: Free-form answers are hard to skim and hard to reuse. Even when the content is good, you lose time reorganizing it into your template.

Solution: Tell the AI to use a strict structure with headings, bullet limits, or a table. Example: “Return as:

1) Summary (2 lines)

2) Steps (numbered, 1-7)

3) Risks (3 bullets)

4) Checklist (10 items).”

Result: The output drops straight into your doc, spreadsheet, or content calendar.

#6: Assumptions You Control (Ask for Clarifications or State Defaults)

Problem: When the prompt doesn’t say what to assume, the AI fills gaps inconsistently - like choosing the wrong platform for an ad or the wrong audience for a tone shift.

Solution: Either (A) force questions or (B) lock defaults. Use one of these lines:

• “If anything is missing, ask me up to 3 questions before writing.”
• “If missing, assume [default], and tell me what you assumed.”

Pick one approach every time.

Result: You avoid silent guesswork and can trust the output to be aligned with your intent.

#7: Include Examples (Style, Tone, and “What Good Looks Like”)

Problem: “Write in a modern tone” doesn’t help much. The AI may mimic a vibe you didn’t mean, and you won’t notice until you compare it to real examples.

Solution: Provide 1-2 short examples: a paragraph you like, a product description, or a previous caption. Then specify what to copy: “Match the rhythm, keep sentences under 12 words, use one line breaks after each sentence.”

Result: The AI reproduces your style more reliably than instructions alone.

...