Chapter 8 Exercises: Context Is Everything

These exercises develop your ability to identify missing context, build context packets, and structure context-loading effectively. Work through them in order — the skills from early exercises are required for the later ones.

Part A: Context Identification

Exercise 1: Context Type Mapping

For each AI output failure below, identify which of the six context types was missing (Background, Task, Audience, Style, Constraint, Reference). Explain what specific information would have prevented the problem.

Failure A: The AI produced a blog post about workplace wellness that was written at a general-public level. The actual audience was registered nurses with 10+ years of clinical experience.

Failure B: The AI generated an executive summary that included detailed technical specifications. The actual purpose was for a board presentation — the board members are all non-technical.

Failure C: The AI drafted a client email that mentioned a competitor by name and referenced an ongoing lawsuit. The writer's legal team prohibits both.

Failure D: The AI wrote marketing copy using an upbeat, exclamation-heavy tone. The brand's established voice is calm, minimalist, and completely free of exclamation marks.

Failure E: The AI produced a competitive analysis that included accurate general information about the industry but missed all of the company's specific differentiators, because the company's actual value proposition was never provided.

Exercise 2: The Context Audit

Apply the context audit framework from Section 8.14 to a real AI interaction you have experienced (or create a realistic scenario from your own work).

For your chosen scenario: 1. Describe the task you gave the AI and the disappointing output you received 2. Work through all six context types: what was present and what was missing? 3. Write the revised prompt with the missing context added 4. Explain what specific differences you would expect in the output

Exercise 3: Context Strip and Rebuild

Read the following complete, well-contextualized prompt. Then answer the analysis questions below.

Full prompt: "I am a senior communications director for a publicly traded retail company (approximately 4,000 employees). I am working on the internal communication about our upcoming store closures — we are closing 12 stores over 90 days, affecting approximately 340 employees, 280 of whom will be offered roles at other locations.

I need to draft the initial announcement email from our CEO to all company employees. This email goes out the same morning that local media will receive the press release.

Key constraints: - HR and legal have approved the core facts I will provide; do not add facts or figures I haven't given you - Do not use the word 'unfortunately' or begin with an apology - The 280 employees with transfer opportunities should feel genuinely prioritized — this is real and we want them to feel it - The 60 employees without transfer options will be given severance packages; do not mention the specific terms in this email - Maintain confidence in the company's overall direction — this is not a company in crisis, it is a company making a strategic decision - 400 words maximum

Company voice: Direct, human, no corporate euphemisms. We use 'employees' not 'associates,' 'team members,' or 'colleagues.'

Please draft the email."

Questions: a) Which of the six context types are present? Cite specific elements. b) Are there any context types missing that would improve the output further? c) Which constraint is doing the most important work, and why? d) What would happen if the "Company voice" section were removed? e) What would happen if the "Key constraints" section were removed?

Part B: Building Context Packets

Exercise 4: Personal Context Packet

This is the primary practical exercise of the chapter. Build a complete context packet for your own most frequent recurring AI task.

Step 1 — Select your task category: Choose a type of work you do regularly and expect to continue using AI to assist with. Examples: writing team updates, reviewing vendor proposals, drafting client-facing reports, creating training materials, preparing board presentations, writing recruitment communications.

Step 2 — Analyze your context needs: For your chosen task, answer: - What background about you and your organization should the AI know? - Who is the audience for this type of work? What do they know? What do they value? - What style/voice is required? Can you find one sample sentence that captures it? - What are the standing constraints? (What would a reasonable AI output include that you absolutely cannot use?) - What reference materials, if any, apply to every instance of this task?

Step 3 — Draft your context packet: Using the template from Section 8.8, build a complete context packet for your selected task category.

Step 4 — Test it (if you have access to an AI tool): Open a new AI session, paste your context packet, then issue a representative task from your category. Compare the output to your typical output without the packet.

Step 5 — Reflect: In 3-5 sentences, describe what changed in the output and what you would revise in your context packet based on the test.

Exercise 5: The Multi-Context Challenge

You need to create AI-assisted content for three different clients in the same week. Each has a different voice, audience, and constraint set.

Client A: A regional hospital network — conservative, compliance-sensitive, writing for patients and families Client B: A venture-backed consumer tech startup — casual, bold, writing for early adopters aged 18–35 Client C: A global law firm — formal, precise, writing for corporate clients

For each client, draft the core sections of a context packet: - Standing constraints (at least 3 specific, non-obvious constraints) - Style guidance (at least one descriptive sentence + one example sentence) - Audience description (specific enough to influence vocabulary and depth)

Then: compare your three packets. What are the most significant differences? What does this exercise reveal about how much context actually varies between clients?

Exercise 6: Context Packet for a Team

You are a team lead for a content marketing department of 8 people. You want all team members to use AI tools for drafting, but the output quality varies dramatically across the team.

Design a shared context packet that would: - Give every team member consistent brand voice grounding - Include standing constraints that apply to all content (not project-specific) - Be specific enough to matter but general enough to apply to all content types - Be maintainable — someone on the team can update it as the brand evolves

Write the complete shared context packet, then write a brief (2-paragraph) instruction memo explaining to the team how and when to use it.

Part C: Style Context Practice

Exercise 7: Voice Extraction Exercise

Choose a piece of writing you admire — a newsletter, a piece of journalism, a book passage, a corporate blog post, or any other nonfiction writing that has a distinctive and useful voice.

Extract a style reference from it that you could use in an AI context packet: 1. Write a 2–3 sentence description of the voice (what makes it distinctive) 2. Identify 3 specific techniques the writer uses (sentence length, verb choice, use of examples, etc.) 3. Write a Do/Don't pair for at least 3 elements 4. Pull one example sentence that perfectly represents the voice

Then: write a style context block you could use in a prompt to get AI output that matches this voice.

Exercise 8: The Adjective Limitation Test

This exercise demonstrates why style description alone is insufficient.

Step 1: Write a style context using only adjective descriptions: "Write in a style that is [5-7 adjectives]." Do not include examples.

Step 2: Write a style context using only an example: "Write in the style of the following example: [your example]. No further description."

Step 3 (requires AI tool): Submit both versions with the same writing task and compare the outputs.

Step 4: Write a paragraph analyzing: which produced output closer to your intent? What does this suggest about how AI processes style instructions?

If you do not have access to an AI tool: predict which would produce better results and explain your reasoning in detail.

Exercise 9: Do/Don't Pair Construction

For each of the following brand voice descriptions, write at least three specific Do/Don't pairs that operationalize the description:

Brand A: "We are bold and irreverent — we challenge conventional wisdom in our industry without being cynical."

Brand B: "We are empathetic and expert — we know more than our clients do about this subject, but we never make them feel that way."

Brand C: "We are precise and efficient — our clients are time-poor executives who want information without narrative."

A Do/Don't pair should be specific enough that a writer (or AI) could apply it without guessing. "Do be bold" is not a pair — it is still a vague description. "Do challenge the conventional industry assumption in your first paragraph, rather than building to your point of difference" is a pair.

Part D: Reference Material Handling

Exercise 10: Framing Reference Documents

You have a 3,000-word market research report and need AI help with two different tasks:

Task A: Drafting an executive summary of the entire report (400 words, for your CEO) Task B: Extracting and presenting only the competitive landscape findings (for a sales deck)

Write two different prompts — one for each task — that: - Frame the reference material appropriately for each task - Tell the AI exactly how to use the document - Include any relevant audience or format context - Specify what to include and what to ignore from the document

Then: write a paragraph explaining how the framing differs between the two prompts and why.

Exercise 11: Multi-Document Context Prompting

You need to produce a strategic recommendation document. You have access to: - Document A: Internal sales data from the last 12 months - Document B: A competitor analysis from last year - Document C: A rough draft that your colleague started but did not finish

Write a prompt that: - Establishes the relationship between all three documents - Tells the AI how to use each one differently - Specifies what the final output should look like - Includes appropriate constraints about data accuracy (the AI should not invent figures)

Part E: Platform-Specific Context Features

Exercise 12: Custom Instructions Design

Design a set of Custom Instructions for your own ChatGPT account (or a hypothetical account representing your professional role).

Field 1 (What would you like ChatGPT to know about you): Write 150–200 words covering: your role, your organization type, the types of tasks you use AI for most, your expertise level, and any consistent audience you write for.

Field 2 (How would you like ChatGPT to respond): Write 150–200 words covering: preferred output formats for your common tasks, standing style constraints, length preferences, and any specific behaviors you want consistently (e.g., "always end recommendations with a specific next action").

Then: identify at least one thing your Custom Instructions would not cover that still requires per-session context packets. What does this reveal about the limits of persistent context?

Exercise 13: System Prompt Design

You are implementing an AI writing assistant for your HR department. The assistant will be used for: job description writing, offer letter drafting, policy communication, and manager coaching resources.

Design a system prompt that: - Establishes the assistant's role and scope - Sets standing tone and voice guidelines for HR communications at your organization - Lists the constraint categories that apply to HR communications (legal, sensitivity, etc.) - Provides guidance on when to flag uncertainty rather than generate an answer - Is concise enough to remain useful (under 400 words)

Part F: Advanced Integration

Exercise 14: Context for Sensitive Communications

You are a people manager who needs to communicate a team restructuring to your 6-person team. Two people will be reassigned to new teams; the other four are unaffected. This is sensitive, nuanced communication.

Build a complete context packet for this communication that: - Provides necessary background for the AI without providing confidential personnel details - Specifies the emotional register required (honest, empathetic, clear about uncertainty) - Includes constraints around what cannot be said in advance of formal HR conversations - Specifies the audience's likely emotional state and what they need to hear

Then write the first email using this context. Reflect: what made this context packet different from a typical professional communication packet?

Exercise 15: The Context Refresh Protocol

You are in a long AI working session — you have been working for 2 hours on a complex client proposal. The session has now accumulated 35 exchanges. You notice the AI's last three outputs have drifted: the tone has become more formal than your style guidance, the AI included a technical term you asked it to avoid, and the length of outputs has grown well beyond what you specified.

Write: 1. A lightweight context refresh message (under 100 words) that addresses only the most critical drift 2. A full context reset message (under 300 words) that re-establishes all key parameters for the second half of the session 3. A brief explanation of when you would use each type, and what the tradeoff is between them

Reflection Questions

After completing the exercises, consider:

Which of the six context types do you most consistently provide in your current AI use? Which do you most consistently omit?
How long would it take you to build context packets for your three most frequent AI-assisted work categories? What is stopping you from doing it?
When you think about AI output that has disappointed you in the past, which context failure pattern was most often responsible?
If you were to explain the blank slate problem to a colleague who has never read this chapter, what analogy would you use? (Not the "new employee" analogy from the chapter — find your own.)
What is the hardest type of context to articulate for your specific work? Why is it hard, and what would help you get better at expressing it?