The Shelly Palmer Prompt Crafting Workbook

Introduction

Artificial intelligence can do extraordinary things, but only when you know how to tell it what to do. The skill that unlocks everything is prompt crafting.

This workbook teaches you how to think, write, and work with AI systems as collaborators. You’ll learn how to build prompts that produce clear, useful, and business-ready results.

You don’t need a technical background. You do need curiosity, precision, and a willingness to iterate.

• Keep an AI assistant window open (ChatGPT, Claude, Gemini, Copilot, or another) and follow along.
• Copy the examples, paste them into your assistant, and test your results.

Every section in this workbook is designed to be used, not just read.

Note: I strongly recommend you have a paid version of whatever AI assistant you wish to use. Free versions generally offer far fewer features than paid versions and they also restrict token usage which, as you shall learn here, is an unacceptable restriction for professional use.

1. Why Prompt Crafting Matters

Prompt crafting is the discipline of communicating with AI systems in a structured, repeatable way.

It is how you convert intent into predictable action.

Every interaction with an AI model is a coded instruction.

You’re not “chatting with a bot”; you’re programming language.

The better you frame your instruction, the better the model’s reasoning, accuracy, and tone.

Why It Matters

• Every prompt is a business instruction.
  It drives an outcome that will be acted on or published.
• Precision saves time and tokens.
  Clear input eliminates rework and wasted cycles.
• Clarity defines quality.
  If you can’t explain the goal clearly, the model can’t deliver it.
• Consistency builds trust and scale.
  Reusable prompt frameworks create repeatable, brand-safe results across teams.

AI will not make you smarter; it will amplify your clarity.

Prompt crafting is how you give the machine that clarity and keep control of its output.

Example

Vague: “Write something about our new product launch.”

Effective: “You are my marketing strategist. Write a 150-word summary announcing the Q4 product launch to retail partners. Use plain English and end with a one-sentence call to action.”

Now You Try

Open a blank document. List one or two recurring tasks you could make faster or clearer by writing better prompts.

Example: “Summarizing a 30-page market report into a 2-paragraph executive brief.”

⬆️ Back to Table of Contents

Section 3. Context Is Everything

Context is what tells the model who you are, what you want, and how you expect the work delivered. Without context, even powerful models will guess, often incorrectly.

Every effective prompt contains six context fields. These create a predictable frame the AI can use to reason and write.

The Six Context Fields

1. Role – Who the model should be.
   Example: marketing strategist, financial analyst, comms director.
2. Audience – Who will read or use the output.
   Example: board, customers, internal team.
3. Objective – What needs to be accomplished.
   Example: summarize, evaluate, propose.
4. Constraints – Rules, limits, and compliance boundaries.
   Example: 200 words, plain language, brand tone.
5. Format – Structure, length, or file type.
   Example: bullets, slide outline, memo, table.
6. Quality bar – How success will be judged.
   Example: factual accuracy, actionable recommendation, correct tone.

Example Prompt

TEXT
You are my marketing strategist. Write a 250-word executive summary of our Q4 social-campaign performance for the CMO. Include spend, reach, and ROI. Keep it factual and actionable. End with one recommendation.

This prompt works because it defines all six fields before requesting output.

How to Apply Context in Practice

• Start with who you are and who it’s for.
• Add why you’re doing it and how long it should be.
• Finish with how you’ll measure success.

Now You Try

Choose one recurring task you delegate (report, memo, campaign brief).

Write a contextual prompt using all six fields above.

Review it against your own standards for clarity, brevity, and purpose.

⬆️ Back to Table of Contents

5. Pre-Prompts: Setting the Frame

A pre-prompt is the short instruction you give the AI before every exchange. It sets tone, depth, and behavioral expectations.

If your JSON Context Profile defines the “who”, your pre-prompt defines the “how”.

Professionals start with a frame. The frame sets expectations for how the work will be done and judged.

Why Pre-Prompts Matter

AI starts each session as a blank slate. A consistent pre-prompt aligns voice, rigor, and scope from the first token.

• Saves time by defining behavioral rules upfront.
• Enforces brand voice and communication standards.
• Reduces rework and keeps outputs consistent across models.
• Anchors tone, level of detail, and purpose for every conversation.
• Teaches the model what “good” looks like in your terms.

What a Good Pre-Prompt Includes

1. Role behavior – What kind of collaborator is the model? (trusted advisor, analyst, creative partner)
2. Tone and depth – How should it speak? formal or conversational, tactical or strategic
3. Success condition – What does a good answer look like? (concise, actionable, persuasive)

One to three sentences is ideal: short enough to remember, specific enough to set expectations.

Formula

Pattern: “You are my [role]. Respond with [tone and depth]. Focus on [objective]. Avoid [undesired behavior]. End with [success criterion].”

Detailed Examples

Strategy (Trusted Advisor)

TEXT
You are my trusted advisor. Answer with clarity, business relevance, and candor. Avoid jargon and speculation. Use complete sentences. End with two actionable recommendations.

Finance/Operations (Risk Review)

TEXT
Assume the role of a risk officer. Identify key risks with likelihood and mitigation. Use a simple RAG rating. Limit to 200 words.

Marketing (Creative Ideation)

TEXT
Act as a senior creative director. Generate three on-brand ideas with a one-sentence rationale each. Be imaginative and practical. Avoid clichés.

Sales (C-Suite Prep)

TEXT
You are a sales strategist preparing for a C-suite meeting. Summarize the prospect’s pain points, align to business priorities, and propose a concise value message. Keep it under 250 words.

Best Practices

• Clarity: One goal per pre-prompt.
• Consistency: Reuse for similar tasks.
• Conciseness: Keep under ~60 words.
• Scalability: Pair each JSON profile with 1–3 standard pre-prompts.
• Adaptability: Adjust tone to the audience (board vs. internal).

How to Test a Pre-Prompt

1. Run a task without a pre-prompt. Note tone, structure, and clarity.
2. Run the same task with your pre-prompt.
3. Compare edit effort, alignment to audience, and scope control.
4. Iterate wording until it reliably reduces edits by 25% or more.

Common Mistakes and Fixes

• Too vague: Be explicit about tone and outcome.
• Too long: Noise dilutes guidance; keep to three sentences max.
• Style only, no purpose: Add a clear business objective.
• No success criterion: End with “Deliver…” or “End with…”.

Advanced Pre-Prompt Patterns

Clarity Enforcer

TEXT
Answer as a senior advisor. Use one paragraph to summarize, then three bullet points of evidence. End with a recommended decision.

Data Interpreter

TEXT
Act as a data translator. Explain findings in plain English for executives. Highlight trends, context, and business impact.

Comms Polisher

TEXT
Act as a corporate communications editor. Make the text concise, positive, and audience-appropriate. Preserve meaning. End with a readability note.

Now You Try

Write two pre-prompts you could reuse: one for creative ideation and one for executive summaries. Test each by running the same task with and without the pre-prompt, then compare the results.

⬆️ Back to Table of Contents

6. Understanding the Difference: JSON Context Profile vs. Pre-Prompt

We’ve learned about these two tools. Both control how an AI assistant behaves. It is logical to ask if they can be combines. They can, but each serves a distinct purpose and it is better to keep the separated. Here’s some additional detail.

• JSON Context Profile defines who the AI is. This includes its identity, tone, and rules. It’s the operating system for your AI assistant.
• Pre-Prompt defines how the AI should act in this specific conversation. How it behaves, its focus, and immediate objective.

Think of It Like This

• The JSON Context Profile is your permanent job description.
• The Pre-Prompt is your meeting brief. It tells the AI what’s happening right now and how to respond today.

Example Pair

JSON Context Profile (the “Who”)

JSON
{
  "name": "Executive Communications Assistant",
  "purpose": "Draft concise, business-relevant summaries for Fortune 500 leaders.",
  "audience": "Executives and board members",
  "style": {
    "tone": "Professional, direct, confident",
    "rules": [
      "Use active voice",
      "Short paragraphs only",
      "Avoid em dashes",
      "No contrastive constructions"
    ]
  },
  "format": {
    "defaults": {
      "length": "300–450 words",
      "structure": "short intro + bullet points + conclusion"
    }
  },
  "quality": {
    "hallucination_tolerance": "low",
    "checks": [
      "Factual accuracy",
      "Clarity of recommendation",
      "Business relevance"
    ]
  }
}

Pre-Prompt (the “How”)

TEXT
You are my communications strategist. Respond with clarity, precision, and business relevance. Avoid jargon. Write in plain English. End with one actionable next step.

How They Work Together

1. The JSON Context Profile sets the foundation. It tells the AI who it is, its tone, and quality standards.
2. The Pre-Prompt activates that profile for a specific session, shaping the context, focus, and outcome.
3. Used together, they create consistency (profile) and precision (pre-prompt).

Analogy

Imagine you’re leading a team:

• The job description says what your communications director is responsible for. That’s the JSON Context Profile.
• Before each meeting, you give a quick briefing: “We’re presenting to the board, keep it concise and visual.” That’s the Pre-Prompt.

Quick Summary

Executive Takeaway

Use the JSON Context Profile to create consistency across your team’s AI workflows.

Use the Pre-Prompt to create precision in a specific conversation.
Together, they make AI output predictable, on-brand, and ready for business.

⬆️ Back to Table of Contents

8. Meta-Prompting and Reasoning

A meta-prompt is a prompt about prompts. It tells the AI how to think before it writes. It forces method selection, makes assumptions visible, and ends with a self-check so you can audit the work.

Why Meta-Prompts Matter

Executives already do this with teams: “Before you brief me, tell me what data you used and why.” Meta-prompts do the same for AI, replacing guesswork with a visible process.

• Forces a clear reasoning method (analysis, synthesis, comparison).
• Surfaces assumptions you can accept or reject.
• Requires a self-check so quality is explicit.
• Makes outputs auditable for leadership review.

Core Pattern

1. Identify the task type – strategic, analytical, creative, or operational.
2. Select a method – e.g., scenario analysis, benchmarking, synthesis.
3. List assumptions – what the model is taking for granted.
4. Perform the task – produce the requested output.
5. Self-check – verify relevance, accuracy, and tone.

Templates

General Meta-Prompt

TEXT
Before answering, identify the task type (strategic, analytical, creative, or operational). Choose an appropriate reasoning method. List your key assumptions. Produce the answer. Finish with a three-item self-check (accuracy, clarity, business relevance).

Simplified Meta-Prompt (Fast Tasks)

TEXT
Identify method → list assumptions → answer → self-check.

Advanced Meta-Prompt (With Trace)

TEXT
Step 1: Classify the task. 
Step 2: Select a reasoning framework and explain why.
Step 3: List 3–5 assumptions.
Step 4: Perform the analysis.
Step 5: Rate confidence (0–100%) and note data gaps.

Business Examples

Marketing Strategy

TEXT
Classify the task (analytical). Choose a framework (SWOT, 5Cs, or JTBD) and explain why. State three assumptions. Produce the analysis. End with a self-check on accuracy, tone, and usefulness for an executive deck.

Sales Enablement

TEXT
Classify the task (operational). Compare current pitch vs. revised version. List assumptions about buyer persona and stage. Deliver the new version. End with what changed and why.

Corporate Comms

TEXT
Decide whether this is a reputation, messaging, or clarity issue. Choose a messaging framework and justify it. List stakeholder assumptions. Draft two statements and evaluate them for clarity, empathy, and brand tone.

How to Test a Meta-Prompt

1. Run the same task twice: once normally, once with the meta-prompt.
2. Compare structure, transparency, and edit effort.
3. Keep the version that is easier to review, summarize, and defend.

Common Mistakes and Fixes

• Style requests instead of process: Ask for a method and assumptions.
• No self-check: Add an explicit closing check.
• Overcomplication: Start with one sentence and expand as needed.
• Using for trivia: Reserve for reasoning and synthesis tasks.

Now You Try

Choose a live task (strategy memo, campaign analysis, comms outline). Write one meta-prompt that forces the model to declare method and assumptions, then run an A/B test to compare outputs for logic, structure, and confidence.

Starter

TEXT
Task: Create a competitive analysis of Q4 ad spend.
Meta-Prompt: Before analyzing, identify the analytical method (benchmark or trend). List data assumptions. Produce the table and explain findings. End with a self-check: accuracy, completeness, and actionability.

⬆️ Back to Table of Contents

10. Variables and Reusable Prompts

AI gets faster, cheaper, and more useful when your prompts become reusable templates instead of one-off instructions. Variables make that possible. They turn static text into a flexible system that can generate endless versions of high-quality content without sacrificing brand consistency.

What Variables Are

A variable is a placeholder you insert inside a prompt — usually written in double curly brackets like {{topic}} or {{audience}}.
When you substitute real values for those placeholders, the AI instantly adapts the same logic to new inputs.

Example:

TEXT
Write a {{length}} executive summary of {{topic}} for {{audience}}. Focus on clarity and measurable outcomes.

Replace the placeholders:

• {{length}} = “150-word”
• {{topic}} = “Q4 marketing performance”
• {{audience}} = “the CMO”

…and the AI will generate:

“Write a 150-word executive summary of Q4 marketing performance for the CMO...”

Why Variables Matter

Variables convert prompting from art to process. They:

• Save time — one template replaces dozens of manual rewrites.
• Ensure consistency — same tone, structure, and quality across all outputs.
• Enable scale — you can feed entire datasets (CSV or JSON) into the same prompt.
• Reduce risk — less human error, fewer ad-hoc edits.
• Support governance — prompt templates can be reviewed, versioned, and approved like any enterprise asset.

This is exactly how you would standardize marketing copy in a content management system — same principle, new tool.

How Variables Work Inside Prompts

Each variable acts like a blank field the model fills in. The surrounding prompt — tone, length, and structure — stays constant.

When you run the same template multiple times, you simply provide new values.

JSON

{
  "prompt": "Create a {{tone}} paragraph describing {{product}} for {{audience}}. Highlight {{benefit}} and end with a call to action.",
  "defaults": {
    "tone": "professional",
    "audience": "B2B marketing leaders"
  },
  "example_input": {
    "product": "AI-driven analytics platform",
    "benefit": "faster decision-making"
  }
}

How to Design a Variable Template

1. Start with a solid prompt. Write one great version first. Don’t abstract too early.
2. Identify the changeable parts. Anything that might vary (topic, tone, audience, medium, region) becomes a variable.
3. Replace those with placeholders. Use double curly brackets — {{this_format}} keeps it visible and standard.
4. Set defaults for stability. If some fields rarely change, include default values so the template never breaks.
5. Test at least three inputs. If all produce high-quality results, the template is ready for reuse.

Examples from Marketing and Comms

1. Ad Copy Template

TEXT
Write a {{length}} social-media post promoting {{product}} to {{audience}}. Tone is {{tone}}. Include one benefit and one call to action.

2. Email Template

TEXT
Draft a {{length}} email for {{audience}} introducing {{offer}}. Focus on {{value_proposition}}. End with a one-line CTA.

3. Press Release Template

TEXT
You are a corporate communications writer. Create a press release announcing {{announcement}}. Include a quote from the CEO, a market context paragraph, and a closing line on next steps.

Scaling with Data

Once you have a tested variable template, you can run it against structured data.

Example Record Set

JSON

{
  "records": [
    { "product": "AI Insights Platform", "audience": "enterprise marketers", "benefit": "faster campaign optimization" },
    { "product": "Data Integrity Monitor", "audience": "CFOs and finance teams", "benefit": "error-free reporting" }
  ]
}

Each row produces a personalized, brand-consistent message without rewriting the prompt.

Best Practices for Variable Prompts

• Use descriptive names. {{audience}} is better than {{a}}.
• Keep scope tight. No more than 5–6 variables per prompt.
• Validate inputs. If a field can be empty, set a default.
• Combine with Context Profiles and Pre-Prompts. The profile enforces tone; the pre-prompt defines purpose; variables handle scale.
• Document everything. Save templates and example inputs in your team’s prompt catalog.

Testing Your Variable Prompt

1. Choose three distinct inputs (topics or audiences).
2. Run the prompt with each input.
3. Compare for clarity, tone, and brand consistency.
4. Revise your defaults until the outputs require minimal editing.

Advanced Tip — Nested Variables

For more complex workflows (like campaign generation), you can nest variables.

TEXT
Write a {{length}} {{content_type}} promoting {{product}}. 
Use tone: {{tone}}. 
Audience: {{audience}}. 
Include three hashtags from this list: {{hashtags}}.

The nested structure lets you build modular libraries:

• One template for copy structure
• One file of approved variables (e.g., tone, hashtags, CTAs)

Governance and Security Considerations

When scaling variable prompts across teams:

• Store templates in approved repositories (Google Drive, SharePoint, or secure prompt catalogs).
• Tag each template with owner, version, and last review date.
• Redact sensitive data before uploading to consumer LLMs.

Now You Try

Step 1: Choose a recurring task (e.g., social copy, email subject lines, meeting summaries).
Step 2: Write one solid prompt for that task.
Step 3: Identify what changes between instances (topic, audience, tone, offer).
Step 4: Replace those with variables and add defaults.
Step 5: Test with three inputs and record what you learn.

Executive Takeaway

Variables turn prompting into process.

They allow you to scale creativity without sacrificing control, the same way templated brand systems let you scale design.

Once you start working this way, your prompts stop being one-off requests and become operational assets that compound in value over time.

⬆️ Back to Table of Contents

12. Skills (Anthropic Framework)

Skills are reusable reasoning patterns. Think of them as mental functions you can call on demand.

Naming the skill tells the AI what kind of thinking to perform before it writes.

Core Skills and When to Use Them

• Summarize — Condense content to the essentials. Use for reports, decks, meeting notes, and research.
• Compare — Evaluate options against criteria. Use for vendor selection, media plans, tools, or creative options.
• Synthesize — Merge inputs and reconcile conflicts. Use for cross-functional updates and trend briefs.
• Evaluate — Judge against measurable standards. Use for QA, policy checks, and compliance language.
• Plan — Create steps, owners, and timing. Use for 30-60-90-day plans and campaign workbacks.
• Generate — Produce variants or options. Use for headlines, copy, angles, and creative prompts.

Skill Descriptions with Business Examples

Summarize

Purpose: Reduce to essentials for fast decisions.

Example: Weekly marketing performance into five bullets for the CMO.

TEXT
Summarize this deck for a CMO audience. Five bullets max. One risk. One next step.

Compare

Purpose: Rank options with transparency.

Example: Choose between three attribution tools.

TEXT
Create a comparison table with columns: option, capabilities, integration effort, cost, risk. Highlight the best fit and why.

Synthesize

Purpose: Combine inputs into one message.

Example: Merge research, sales feedback, and social listening into a single narrative.

TEXT
Synthesize these sources into a single POV. Reconcile conflicts. Call out assumptions. End with a one-sentence thesis.

Evaluate

Purpose: Apply standards or policy checks.

Example: Check messaging for readability, accuracy, and brand rules.

TEXT
Evaluate this draft against criteria: accuracy, readability, brand voice, compliance language. Score 1–5 and justify each score.

Plan

Purpose: Create a practical path to execution.

Example: Turn a strategy into a 30-60-90 plan with owners.

TEXT
Draft a 30-60-90 day plan. Include goals, actions, owners, and success metrics. Add dependencies and risks.

Generate

Purpose: Explore options within constraints.

Example: Create five on-brand subject lines with different angles.

TEXT
Generate 5 subject lines. Each uses a different angle: benefit, urgency, curiosity, proof, objection. Keep under 55 characters.

Skill Chains

A skill chain sequences two or more skills to move from raw input to a decision.

Pattern: Summarize → Evaluate → Recommend → Plan

TEXT
Summarize the brief in 5 bullets → Evaluate the proposed approach against our criteria → Recommend the best option with rationale → Draft a 30-60-90 plan with owners.

Domain Examples

Marketing

TEXT
Summarize Q4 media performance → Compare 3 optimization options → Recommend one → Plan next steps with timeline.

Strategy

TEXT
Synthesize market signals → Evaluate risks and dependencies → Recommend a growth path → Plan milestones and metrics.

Comms

TEXT
Summarize the issue → Generate 2 statements → Evaluate against clarity and empathy → Recommend one and plan stakeholder rollout.

Sales

TEXT
Summarize account notes → Identify key pains → Generate a tailored value message → Plan next best action with owner.

Now You Try

Design a skill chain for a complex problem you manage. Keep it to 3–4 skills and end with owners.

TEXT
Summarize → Compare → Recommend → Plan

⬆️ Back to Table of Contents

14. Your Prompt Crafting Checklist

Use this checklist before you send any AI request. If you can check every box, your prompt is production-ready.

• Role defined
• Audience defined
• Objective stated
• Constraints listed
• Format specified
• Quality bar clear
• Pre-prompt active
• Meta-prompt guiding
• Variables reusable
• Review complete

Tip: Save all prompts that pass this checklist to your team catalog with an owner, version, and last reviewed date.

⬆️ Back to Table of Contents