Chapter 9 Quiz: Instructional Prompting and Role Assignment
Test your understanding of instructional design, verb choice, role assignment mechanics, and the limits of role-based prompting. Attempt each answer before expanding the solution.
Question 1
What is the primary difference between asking and instructing in the context of AI prompting?
A) Asking produces longer responses; instructing produces shorter ones B) Asking invites the AI to interpret the request; instructing specifies the precise cognitive operation to perform C) Asking is used for questions; instructing is used for creative tasks D) Asking is appropriate for complex tasks; instructing is for simple ones
Answer
**B) Asking invites the AI to interpret the request; instructing specifies the precise cognitive operation to perform.** The distinction is about precision of direction, not response length (A) or task type (C, D). "Can you give me feedback on this?" asks for whatever feedback the AI thinks is appropriate. "Identify the three weakest assumptions in this argument and explain what question each one would prompt from a skeptical decision-maker" instructs a specific cognitive operation with specified scope, quantity, and output format. The instructing version does not need to be longer — it needs to be more precisely targeted.Question 2
In the verb taxonomy from Section 9.2, which category does "Extract" belong to, and what does it signal to the AI?
A) Transformative — it signals a request to change the form of existing content B) Generative — it signals a request to produce new content C) Interrogative — it signals a request to identify and surface specific information from content D) Reasoning — it signals a request for logical analysis
Answer
**C) Interrogative — it signals a request to identify and surface specific information from content.** Interrogative verbs (identify, find, extract, surface, highlight, flag) signal to the AI that you want targeted retrieval of specific information from existing content, not generation of new content (B) or transformation of form (A). This is distinct from analytical verbs like "analyze" or "evaluate," which produce assessments; interrogative verbs produce located, specific outputs.Question 3
True or False: Assigning an expert role to the AI (e.g., "You are a leading expert in X") reliably improves the factual accuracy of its output about that topic.
Answer
**False.** This is one of the most important myths about role assignment. Expert role assignment reliably changes the register, vocabulary, and focus of output — but it does not reliably improve factual accuracy. Some research suggests it can increase the confidence with which output is stated, including the confidence with which potentially incorrect information is delivered. The accuracy of AI output is determined by training data quality, not by the role assigned. For accuracy-sensitive tasks, pair role assignment with explicit instructions to acknowledge uncertainty, and always verify consequential facts independently.Question 4
What is the "say-mean gap" in instructional prompting?
A) The gap between what you want the AI to do and what the AI is capable of B) The gap between what your instruction literally says and what you actually mean C) The gap between an AI's response and the correct answer D) The gap between a vague prompt and a specific one on the specificity ladder
Answer
**B) The gap between what your instruction literally says and what you actually mean.** The say-mean gap exists when your instruction is technically interpretable in multiple ways, and the AI's reasonable interpretation differs from your actual intention. "Make this better" literally means "improve this" — but what you mean might be "make this 30% shorter" or "make this more direct" or "strengthen the argument in paragraph two." Closing the gap means translating your intention into precise, specific language rather than relying on the AI to infer what "better" means in your context.Question 5
Which of the following role assignments is most likely to produce genuinely useful feedback on a written proposal?
A) "You are an expert. Review this proposal." B) "You are a skeptical CFO at a mid-sized technology company. You have approved two proposals this year and rejected eight. You care primarily about implementation risk and ROI timeline. Review this proposal and identify the three questions you would ask before approving it." C) "Review this proposal as a professional." D) "You are a helpful reviewer. Give me constructive feedback on this proposal."
Answer
**B) "You are a skeptical CFO at a mid-sized technology company. You have approved two proposals this year and rejected eight. You care primarily about implementation risk and ROI timeline. Review this proposal and identify the three questions you would ask before approving it."** Option B is the only one that specifies: the role's industry context, the role's decision history (skeptical because they reject most proposals), the role's primary concerns, and the exact output format (three questions). Option A is a role without a role's perspective or focus. Option C has no role at all. Option D names a default "helpful reviewer" mode, which is what the AI defaults to anyway, and adds nothing.Question 6
What is the primary purpose of using the "target audience member" role archetype?
A) To generate content that appeals to a target audience B) To have the AI read your content from the perspective of a specific reader, revealing how that reader would actually react C) To calibrate vocabulary and reading level to match the audience's expertise D) To identify demographic gaps in your content strategy
Answer
**B) To have the AI read your content from the perspective of a specific reader, revealing how that reader would actually react.** The audience role technique is primarily for testing — getting feedback on how a specific type of person would respond to content that already exists or is being considered, not for generating content (A) or calibrating vocabulary (C). It works by having the AI inhabit a detailed persona description and report that persona's honest reaction: what engages them, what loses them, what questions they have, what would make them take action. Demographic gap analysis (D) is a separate research task.Question 7
When should you use system-level role assignment versus message-level role assignment?
A) System-level for technical tasks; message-level for creative tasks B) System-level for long, specialized sessions where a consistent perspective is needed; message-level for perspective shifts within broader sessions C) System-level is always more effective; message-level should be avoided D) System-level for expert roles; message-level for audience roles
Answer
**B) System-level for long, specialized sessions where a consistent perspective is needed; message-level for perspective shifts within broader sessions.** The choice is about consistency vs. flexibility. System-level roles persist throughout the session, making them ideal for specialized tools or sessions that require one sustained perspective. Message-level roles are scoped to specific exchanges, making them ideal when you need different perspectives on different aspects of a project within the same session. There is no quality difference between the two approaches for tasks within their appropriate scope.Question 8
You assign the role "You are a world-leading expert in data privacy law" to an AI and ask it a complex question about a recent regulatory change. What is the most important risk to be aware of?
A) The AI will refuse to answer because the topic is too specialized B) The AI will produce output in an overly academic register C) The AI may produce a confident, expert-sounding response that contains factual inaccuracies, because role assignment does not grant expertise the AI does not have D) The AI will provide too many qualifications and disclaimers
Answer
**C) The AI may produce a confident, expert-sounding response that contains factual inaccuracies, because role assignment does not grant expertise the AI does not have.** This is the most important limitation of role assignment and one of the most dangerous misconceptions in AI use. Expert role assignment increases confidence of output — including confident-sounding incorrect claims — without necessarily increasing accuracy. For any topic where factual accuracy matters, expert role assignment should be paired with explicit uncertainty acknowledgment instructions and followed by independent verification.Question 9
What is "role stacking" and when does it work best?
A) Assigning the same role at both system and message level for consistency B) Assigning multiple roles to the AI simultaneously or sequentially to get multiple perspectives C) Using different roles in different sessions on the same project D) Building progressively more specific role descriptions over multiple prompts
Answer
**B) Assigning multiple roles to the AI simultaneously or sequentially to get multiple perspectives.** Role stacking uses two or more distinct roles in a single prompt — either simultaneously (the AI considers both perspectives at once) or sequentially (the AI produces separate assessments from each perspective, then synthesizes them). It works best when the perspectives are complementary and the task genuinely benefits from more than one viewpoint. The practical limit: stacking more than two or three roles tends to produce superficial treatment of each perspective.Question 10
Which of the following negative instructions is most appropriate in a role assignment context?
A) "Do not make any factual claims" B) "Do not use bullet points" C) "Do not shift to a balanced or supportive perspective — maintain the devil's advocate position throughout" D) "Do not write more than 500 words"
Answer
**C) "Do not shift to a balanced or supportive perspective — maintain the devil's advocate position throughout."** This negative instruction addresses a specific, documented failure mode in role assignment: the AI defaulting back to its comfortable "balanced and helpful assistant" mode when assigned a challenging role. The instruction prevents this regression, ensuring the assigned role is maintained throughout the response. Options A, B, and D are format and content constraints, not role-maintenance instructions — they are not specifically applicable to the role assignment context in the same way.Question 11
You want to use role assignment to evaluate a client proposal but are concerned about the ethical note in Section 9.17. Which approach is most appropriate?
A) Avoid role assignment for any task involving sensitive content B) Assign the expert role and override any uncertainty disclaimers to get clean, confident output C) Assign the role to shape perspective and focus, include explicit instructions to flag uncertainty, and verify consequential facts independently D) Use role assignment only for creative tasks, not analytical ones
Answer
**C) Assign the role to shape perspective and focus, include explicit instructions to flag uncertainty, and verify consequential facts independently.** The ethical note in Section 9.17 does not argue against role assignment — it argues against using role assignment specifically to suppress appropriate epistemic humility. The right approach is to use role assignment for what it reliably delivers (perspective, register, focus) while preserving the AI's ability to express uncertainty, and pairing it with your own verification for consequential claims. Overriding disclaimers (B) removes a useful quality signal. Avoiding role assignment for sensitive content (A) is unnecessarily restrictive.Question 12
What does the "naive expert" role archetype specifically test for?
A) Whether the AI can explain complex concepts simply B) Whether the AI can perform tasks outside its training data C) Places in your content where understanding requires background knowledge your intended audience may not have D) Whether an expert in your field would find your content credible
Answer
**C) Places in your content where understanding requires background knowledge your intended audience may not have.** The naive expert is an expert in an adjacent field who has no background in the field of the content — they have sophisticated analytical capability but lack domain-specific knowledge. This role is specifically designed to reveal accessibility gaps: places where jargon goes undefined, assumptions are made without explanation, or logic is skipped over. It is not a test of simplicity (A), AI capability limits (B), or expert credibility (D).Question 13
What is the recommended structure for a sequential instruction with dependencies?
A) All instructions in a single sentence, separated by commas B) Numbered steps where each step specifies what it requires from the previous step and what output it produces C) Each instruction in a separate message, sent one at a time D) A bulleted list of tasks with priorities assigned to each
Answer
**B) Numbered steps where each step specifies what it requires from the previous step and what output it produces.** Explicit numbered sequential instructions are more reliably followed than prose descriptions of the same steps (A), and more efficient than separate messages (C) when the steps can be described in advance and the AI can work through them in one response. Bulleted lists with priorities (D) do not communicate sequential dependency — they communicate parallel tasks with different importance levels, which is a different structure.Question 14
According to Section 9.18, what does research show about where role assignment should be placed for maximum consistency over long sessions?
A) At the end of the first message, so the AI processes all other context first B) System-level or session-opening placement maintains more consistent role adherence over long sessions than message-level placement C) In every message, to prevent the AI from defaulting to its standard mode D) After providing all relevant context, as a final instruction
Answer
**B) System-level or session-opening placement maintains more consistent role adherence over long sessions than message-level placement.** Research comparing placement confirms that system-level or very early session placement of role instructions produces more consistent adherence throughout a long session, while message-level role assignment can be diluted as the conversation progresses. This is consistent with the "lost in the middle" effect — instructions that appear early in the context window maintain more consistent influence than instructions that appear mid-session.Question 15
You use the target audience role technique and receive feedback suggesting your report would lose your intended reader on page 2. What is the most productive next step?
A) Accept that the content is too complex for your audience and simplify the entire report B) Discard the AI feedback — it cannot accurately predict how a real person would read your report C) Use the specific feedback to identify exactly what causes the reader to disengage (unclear argument, missing context, vocabulary barrier) and address those specific issues D) Ask the AI to rewrite the report for a less sophisticated audience