COMP 3270 / Module 2 / Unit 4 / Lesson 4

Lesson 4.4 · Week 3 · Reading: Chapter 4

Verification Requirements and PCAFE-V Integration

Prompt Design Fundamentals — the V in PCAFE-V and complete integration. Learn to specify verification and recognize the limits of prompt improvement.

Suggested time: 3.5–4.5 hours Concludes Unit 4 Builds on Lessons 4.1–4.3

Table of Contents

1. Lesson Overview

In Lessons 4.1, 4.2, and 4.3, you learned to define purpose, context, audience, format, and evaluation criteria. These elements produce a clear, usable, and reviewable output. But even a well-structured output may contain unverified claims, missing information, or unsupported assumptions.

This lesson introduces the final element of the PCAFE-V framework: Verification Requirements. You will learn to specify which claims require verification, how to handle missing information and conflicting evidence, how to require provisional rather than falsely authoritative output, and why asking an AI system to "verify itself" is not independent verification.

You will then integrate all six PCAFE-V elements into a complete prompt, compare an initial and revised prompt, evaluate whether refinement improved the output, and identify limitations that remain despite stronger prompting. By the end of this lesson, you will have a complete PCAFE-V prompt and a clear understanding of the limits of prompt improvement.

This lesson concludes Unit 4 and prepares you for Unit 5: Advanced Prompting and Context Engineering.

Responsible‑use reminder: Asking an AI system to "verify itself" is not independent verification. Human review and accountability are essential.

2. Learning Outcomes

By the end of this lesson, you should be able to:

  1. Incorporate source identification and verification requirements into a prompt.
  2. Require appropriate treatment of missing information, conflicting evidence, unsupported claims, and uncertainty.
  3. Apply all six elements of PCAFE-V in a complete prompt.
  4. Diagnose weaknesses in an initial prompt and revise it purposefully.
  5. Compare initial and revised outputs using common evaluation criteria.
  6. Identify improvements, regressions, and remaining limitations in revised outputs.
  7. Explain why "verify your answer" is inadequate and why human review remains essential.
  8. Construct a complete PCAFE-V prompt with verification requirements and a human review plan.
  9. Document prompt versions, outputs, evaluations, verification needs, and human decisions.

3. Before You Begin

4. Retrieval Practice: Concepts from Lessons 4.1–4.3

Before we add verification requirements, let's refresh the key ideas from the previous lessons.

Question 1: What is the role of evaluation criteria in the PCAFE-V framework?

Show answer

Answer: Evaluation criteria are observable qualities used to assess whether the output is fit for its intended purpose. They provide a consistent, checkable way to judge quality and compare prompt versions.

Question 2: Why is "make it professional" insufficient as a format or criteria instruction?

Show answer

Answer: "Professional" is subjective and open to interpretation. Without observable specifics, the AI cannot reliably produce it, and you cannot consistently evaluate it.

Question 3: What is the distinction between polish and substantive quality?

Show answer

Answer: Polish refers to surface-level professional appearance (grammar, formatting, tone). Substantive quality refers to content-related qualities (accuracy, completeness, relevance, evidence). Polish can conceal weak substance.

Question 4: What is the purpose of minimizing context in a prompt?

Show answer

Answer: Minimizing context removes unnecessary, irrelevant, or sensitive information, reduces the risk of exposing protected data, and helps the AI focus on the essential task.

5. Opening Scenario: "The Polished but Invented Output"

📌 Scenario

Context: You are a graduate student researching the effectiveness of remote work policies. You ask an AI system to "summarize recent research on remote work productivity." The AI produces a detailed, well-written, and professionally formatted summary with citations to recent studies. The summary includes:

  • A citation to "Smith, J. (2025). Remote Work and Productivity. Journal of Management, 45(3), 112-128."
  • A claim that "remote work increases productivity by 25% on average."
  • A statement that "all major studies since 2022 have reached the same conclusion."
  • No mention of conflicting evidence or limitations.
  • No dates or verification notes.

Upon closer inspection, you realize:

  • The citation is invented (the journal, volume, and page number do not exist).
  • The "25%" figure is unsupported and does not appear in any actual study.
  • The claim that "all major studies agree" is false; there is significant debate.
  • The summary presents no uncertainty or conflicting evidence.

Questions for reflection:

  1. What verification requirements were absent from the prompt?
  2. What should the prompt have required to prevent invented citations?
  3. How should the prompt handle conflicting evidence and uncertainty?
  4. What human responsibilities remain after the AI generates the output?
Show guided analysis

1. Absent verification requirements: The prompt did not require source identification, verification of claims, handling of conflicting evidence, no-fabrication rules, or uncertainty preservation.

2. Preventing invented citations: The prompt should require: (a) citations to be traceable to supplied or authoritative sources, (b) a clear distinction between supplied facts and generated interpretations, (c) a no-fabrication rule, and (d) marking of unverified claims.

3. Handling conflicting evidence: The prompt should instruct the AI to note when sources conflict, present both sides, and indicate uncertainty. It should not present a single, definitive conclusion when evidence is mixed.

4. Human responsibilities: The human must verify all claims, check sources, assess the quality of evidence, determine whether the output is fit for purpose, and take final responsibility for any decisions made using the output.

Key insight: Asking the AI to "verify its answer" is not enough. The prompt must include specific verification requirements, and the human must independently verify the output.

6. What Are Verification Requirements?

In the PCAFE-V framework, Verification Requirements are instructions specifying which claims, sources, calculations, dates, assumptions, conflicts, and uncertainties must be checked before the output is used.

Verification requirements answer: How will we know this output is trustworthy? and What must the human check before using it?

Common verification requirements include:

🔍 Verification in practice

Weak verification instruction: "Make sure your answer is accurate."

Strong verification instruction: "All factual claims must be traceable to the supplied sources. For each claim, include a reference (page number or URL). If a claim is not in the supplied sources, mark it as unverified. Do not invent citations. If sources conflict, present both perspectives and indicate the disagreement. Identify any information that is missing or uncertain."

Why "verify your answer" is inadequate

Asking an AI system to "verify its own answer" is not independent verification. The AI system:

Verification is a human responsibility. The prompt can instruct the AI to make verification easier (by citing sources, marking uncertainty, etc.), but the actual verification must be done by a human.

⚠️ Critical reminder

Asking the AI system to "verify itself" is not independent verification. The AI does not have access to reality, cannot check its own sources, and cannot be held accountable. Human users must independently verify all critical claims.

7. Handling Missing Information, Uncertainty, and Conflicting Evidence

Generative AI systems often produce outputs that are overly confident or overly definitive. They may present speculation as fact, ignore contradictory evidence, or invent information to fill gaps.

A responsible prompt must explicitly address these issues:

Missing information

Uncertainty

Conflicting evidence

📋 Example: handling uncertainty and conflict

Instruction: "When sources present conflicting findings, present both perspectives. Indicate the degree of consensus or disagreement. Do not present a single, definitive conclusion if the evidence is mixed. Use tentative language such as 'some studies suggest' or 'evidence is inconclusive' where appropriate."

8. Provisional Output and the No-Fabrication Rule

AI-generated output should be treated as provisional until it has been verified by a human. The prompt should explicitly state that the output is a draft or a starting point, not a final product.

The no-fabrication rule is essential: the AI must not invent citations, data, facts, or sources. This rule should be explicit in the prompt.

🔍 No-fabrication rule example

"Do not invent citations, data, or facts. All claims must be traceable to the supplied sources or explicitly marked as unverified. If a source does not exist or cannot be verified, do not include it."

9. The Limits of Prompt Improvement

It is important to recognize that even the best-designed prompt cannot guarantee:

A well-designed prompt can reduce some errors while making other errors more polished or difficult to notice. This is the AI Paradox in action: improved fluency and polish can make inaccuracies harder to detect.

Common misconceptions about prompt improvement include:

⚠️ Critical understanding

Prompt quality cannot eliminate the need for human verification and accountability. A well-designed prompt may reduce some errors while making other errors more polished or difficult to notice. Human users remain responsible for evaluating and using the generated output.

10. PCAFE-V Walkthrough: Verification Requirements (V)

🔍 Verification Requirements

Definition: Instructions specifying which claims, sources, calculations, dates, assumptions, conflicts, and uncertainties must be checked before the output is used.

Role in prompt design: Verification requirements make the output reviewable and accountable. They ensure that the human user knows what to check and what to do when information is missing or uncertain.

Guiding questions for verification

Weak vs. improved verification

Weak verification example

"Make sure your information is correct."

Why it is weak: Vague, unobservable, and does not specify how to handle missing information, uncertainty, or conflicts.

Improved verification example 
Verification Requirements:
- All factual claims must be traceable to the supplied sources or explicitly marked as unverified.
- For each claim, include a reference (page number, URL, or DOI).
- If information is missing, indicate what is needed and do not invent.
- If sources conflict, present both perspectives and note the disagreement.
- State the degree of confidence for each major claim (e.g., "well-established," "preliminary," "uncertain").
- Do not invent citations, data, or facts.
- Mark the output as a draft requiring human review before use.
- Identify any claims that require qualified human verification (e.g., legal, medical, or financial advice).

Common mistakes

Responsible-use consideration

Verification requirements make it explicit that the AI output is not the final word. They preserve human accountability by identifying what needs to be checked and what remains uncertain. They also help prevent over-reliance on AI by making the limits of the output visible.

Brief practice: verification requirement builder

Write verification requirements for a prompt that asks the AI to "provide a briefing on the current state of renewable energy policy in Canada."

Show sample verification requirements
  • All claims about current policy must be traceable to official government sources or reputable energy agencies.
  • Include the date of each source; indicate when information may be outdated.
  • If policies vary by province, note the provincial differences.
  • Identify any conflicting policy interpretations.
  • Mark the briefing as a draft requiring review by a policy specialist.
  • Do not invent data or citations.

11. PCAFE-V Integration and Prompt Refinement

11.1 The complete PCAFE-V framework

The PCAFE-V framework consists of six elements that work together to produce a responsible, reviewable prompt:

Table 1. The six PCAFE-V elements
Element Definition Guiding question
P — Purpose The legitimate human objective the AI-supported task is intended to advance. Why are we doing this? What human need is being addressed?
C — Context The information, constraints, and background the AI needs. What information does the AI need? What is excluded?
A — Audience The intended reader, user, learner, or stakeholder. Who is the output for? What do they need to know?
F — Format The required organization, length, structure, style, and presentation. How should the output be structured and presented?
E — Evaluation Criteria Observable qualities used to assess whether the output is fit for purpose. How will we know if the output is good?
V — Verification Instructions specifying what must be checked before the output is used. What must be checked before we use the output?

These elements are interdependent. The purpose informs the audience; the audience shapes the format; the format enables evaluation; the evaluation identifies verification needs. Together, they create a prompt that is purposeful, bounded, clear, reviewable, and responsible.

11.2 The prompt refinement cycle

Prompt refinement is an iterative process. The introductory cycle is:

  1. Define the purpose.
  2. Draft the prompt.
  3. Generate or inspect an output.
  4. Evaluate the output using common criteria.
  5. Diagnose the prompt and output weaknesses.
  6. Revise relevant elements of the prompt.
  7. Compare outputs using the same criteria.
  8. Identify remaining limitations.
  9. Record the human decision.
⚠️ Important

Do not treat refinement as repeatedly requesting "better" output without criteria. Each revision should be purposeful, addressing a specific diagnosed weakness. Use the same evaluation criteria to compare initial and revised outputs.

11.3 Comparing initial and revised outputs

When comparing outputs, use the same evaluation criteria for both. This allows you to identify:

This comparison informs the next iteration of refinement and helps you decide whether the revised output is fit for purpose.

11.4 Prompt documentation

For responsible use, you should document:

This documentation supports transparency, accountability, and learning.

12. Worked Example: Complete PCAFE-V Prompt and Refinement

Let's bring together all six elements in a complete prompt, compare initial and revised outputs, and identify remaining limitations.

Context: You are a project coordinator at a municipal government. You need to prepare a one-page summary of a new public transit policy for city council members. You have a draft policy document and some background notes.

Initial prompt (incomplete):

Write a summary of the new public transit policy for city council.

Initial output (excerpt):

The new public transit policy aims to improve transit efficiency and sustainability. It includes several key measures such as expanding bus routes, reducing fares for low-income riders, and implementing new technology. This policy will benefit all residents and is expected to be implemented in 2026.

Diagnosis:

Step 1: Define the purpose

"Help me prepare a one-page summary of the new public transit policy for city council members, so they can understand the key changes and make an informed decision."

Step 2: Add context

"Use the supplied policy document and background notes. Do not include internal staff names or financial details."

Step 3: Define the audience

"City council members with varying levels of technical expertise. They need clear, concise, decision-ready information. Use plain language and define technical terms."

Step 4: Specify format

"One-page briefing with sections: (1) Policy Overview, (2) Key Changes, (3) Impacts on Residents, (4) Next Steps. Use bullet points for key changes. Under 500 words."

Step 5: Define evaluation criteria

Step 6: Add verification requirements

Revised prompt (complete PCAFE-V):

Purpose: Help me prepare a one-page summary of the new public transit policy for city council members, so they can understand the key changes and make an informed decision.

Context: Use the supplied policy document and background notes. Do not include internal staff names or financial details. Definitions: "Transit policy" refers to the new municipal public transit framework.

Audience: City council members with varying levels of technical expertise. They need clear, concise, decision-ready information. Use plain language and define technical terms (e.g., "dedicated bus lane," "signal priority").

Format: One-page briefing with the following sections:
1. Policy Overview (2-3 sentences)
2. Key Changes (bullet list, max 5 items)
3. Impacts on Residents (brief descriptions)
4. Next Steps and Timeline

Length: Under 500 words. Use headings and bullet points.

Evaluation Criteria:
- Relevance: Directly addresses the new transit policy and its key changes.
- Clarity: Plain language; technical terms are defined.
- Completeness: Covers all four sections.
- Source support: All claims must be traceable to the supplied policy document.
- Concision: Under 500 words.
- Uncertainty: Notes any missing information or conflicting interpretations.
- Reviewability: Easy to verify against the policy document.

Verification Requirements:
- All factual claims must be traceable to the supplied policy document (include page references).
- If a claim cannot be verified from the supplied documents, mark it as unverified.
- Do not invent details, figures, or policy positions.
- If information is missing, indicate what is needed.
- Mark the output as a draft requiring council review and verification.
- Identify any claims that require legal or financial review.

Revised output (excerpt):

Policy Overview:
The new transit policy (Document 2025-42) introduces changes to bus routing, fare structure, and technology integration to improve service efficiency and accessibility.

Key Changes:
- Expansion of bus routes to three underserved neighborhoods (see Appendix A, p. 4)
- Reduced fares for low-income residents (50% discount, Document 2025-42, p. 12)
- Implementation of real-time tracking on all buses (p. 18)
- New dedicated bus lanes on two major corridors (p. 22)

Impacts on Residents:
- Improved access to transit for underserved communities
- Reduced transportation costs for low-income households
- More reliable service with real-time information

Next Steps:
Implementation is scheduled for Q1 2026. City council approval is required by December 2025.

Note: This is a draft summary. Please verify all details against the policy document. Some implementation details require further review.

Comparison:

Table 2. Initial vs. revised output comparison
Criterion Initial Revised
Relevance Vague; no specific policy details Clear; addresses specific policy changes
Clarity Generic language Plain language with defined terms
Completeness Missing key sections All four sections present
Source support No sources Page references included
Concision Too short, lacking detail Appropriate length
Uncertainty None; presented as fact Notes draft status and missing details

Improvements: The revised output is more relevant, clearer, more complete, better sourced, and acknowledges uncertainty.

Remaining limitations: The summary still requires verification against the full policy document. Some implementation details are still vague. Legal and financial implications are not addressed.

Human decision: The revised output is suitable as a draft for council review, but it must be verified against the original policy document before distribution.

13. Guided Learning Activities

🧩 Activity 1: Verification Requirement Builder
  • Purpose: Practice creating verification requirements for a prompt.
  • Learning Outcome: LO 1, 2.
  • Estimated Time: 15 minutes.

Instructions: Write verification requirements for a prompt that asks the AI to "prepare a briefing on the safety of a new medical device."

Show sample verification requirements
  • All safety claims must be traceable to the supplied clinical trial data or peer-reviewed research.
  • Include references to specific data tables or page numbers.
  • Clearly distinguish between established safety data and preliminary findings.
  • If adverse events are identified, present the data honestly and without minimization.
  • Do not invent data or citations.
  • Mark the briefing as a draft requiring review by a qualified medical professional.
  • Note that this briefing does not constitute medical advice.
🧩 Activity 2: Complete PCAFE-V Diagnosis
  • Purpose: Practice diagnosing an incomplete prompt using PCAFE-V.
  • Learning Outcome: LO 4.
  • Estimated Time: 15 minutes.

Instructions: Analyse the following prompt and identify what is missing for each PCAFE-V element: "Write a report on the benefits of remote work."

Show diagnosis
  • Purpose: Vague — what is the human need? Who will use the report?
  • Context: Missing — what sources should be used? What definition of "remote work"?
  • Audience: Undefined — who is the report for?
  • Format: Vague — what is the structure, length, and style?
  • Evaluation Criteria: None — how will the output be judged?
  • Verification: None — what sources should be used? How should uncertainty be handled?
🧩 Activity 3: Prompt Refinement and Comparison
  • Purpose: Practice refining a prompt and comparing outputs.
  • Learning Outcome: LO 5, 6.
  • Estimated Time: 20 minutes.

Instructions: Using the scenario from Activity 2, write a revised prompt that includes all six PCAFE-V elements. Then compare the initial and revised outputs (you can use supplied outputs or generate them). Identify improvements, regressions, and remaining limitations.

Show sample comparison framework

Evaluation criteria: Relevance, clarity, completeness, source support, concision, uncertainty.

Expected improvements: The revised output should be more focused, better organized, better sourced, and more transparent about uncertainty.

Possible regressions: The revised output may be longer or more complex.

Remaining limitations: Some claims may still require independent verification. The report may not cover all aspects of remote work.

🧩 Activity 4: Limits of Prompt Improvement
  • Purpose: Practice identifying limitations that remain despite strong prompting.
  • Learning Outcome: LO 7.
  • Estimated Time: 15 minutes.

Instructions: Evaluate the following claims about prompt improvement. Identify which are misconceptions and explain why.

  1. "A detailed prompt guarantees accuracy."
  2. "A request for citations guarantees real sources."
  3. "Specifying an expert role gives the AI professional authority."
  4. "A structured output is automatically more reliable."
  5. "A privacy warning inside a prompt replaces platform controls."
Show analysis
  1. Misconception. Detail improves clarity but does not ensure correctness. The AI may still generate inaccurate information.
  2. Misconception. Citations may be invented unless verification requirements are explicit and independently checked.
  3. Misconception. Role prompting does not create actual expertise or authority. The AI is still a language model, not a professional.
  4. Misconception. Structure improves reviewability but does not ensure content quality. A structured output can still contain errors.
  5. Misconception. Prompt instructions are not security controls. Privacy requires platform-level safeguards and human policies.
🧩 Activity 5: Prompt Documentation
  • Purpose: Practice documenting a prompt, output, and human decision.
  • Learning Outcome: LO 9.
  • Estimated Time: 15 minutes.

Instructions: Use the worked example prompt from this lesson. Create a prompt documentation record that includes: prompt identifier, date, AI system (if applicable), purpose, prompt version, context category, output, evaluation, verification needs, revision, human decision, and material AI-use disclosure.

Show sample documentation record
  • Prompt ID: PCAFE-2026-001
  • Date: 2026-08-29
  • AI System: [Institutionally approved system]
  • Purpose: Prepare a one-page summary of the new public transit policy for city council members.
  • Prompt Version: v2.0 (complete PCAFE-V)
  • Context Category: Public policy, municipal government
  • Output: One-page briefing with four sections, page references, and draft status.
  • Evaluation: Meets all criteria; source support strong; uncertainty noted.
  • Verification Needs: Verify all page references against the policy document. Review legal and financial implications.
  • Revision: Added source references and uncertainty notes.
  • Human Decision: Output is suitable as a draft for council review pending verification.
  • AI-Use Disclosure: AI used as a drafting assistant; final content reviewed and verified by the project coordinator.

14. Exercises

📝 Exercise 1: Write Source and Evidence Requirements
  • Learning Outcome: LO 1.
  • Estimated Time: 10 minutes.

Instructions: Write source and evidence requirements for a prompt that asks the AI to "prepare a summary of a court ruling."

Show sample solution
  • All legal claims must be traceable to the specific court ruling or official case summary.
  • Include the case name, docket number, and date.
  • Use direct quotations where possible, with page or paragraph references.
  • Do not interpret the ruling beyond what is stated.
  • Mark the summary as a draft requiring review by a legal professional.
📝 Exercise 2: Write Missing-Information and Uncertainty Rules
  • Learning Outcome: LO 2.
  • Estimated Time: 10 minutes.

Instructions: Write missing-information and uncertainty rules for a prompt that asks the AI to "project future sales for a new product."

Show sample solution
  • If sales data is incomplete, indicate what is missing and do not estimate.
  • Use ranges rather than single-point estimates where data is uncertain.
  • Clearly distinguish between historical data and projections.
  • State the assumptions underlying each projection.
  • Mark the projections as preliminary and requiring review.
📝 Exercise 3: Complete PCAFE-V Diagnosis
  • Learning Outcome: LO 4.
  • Estimated Time: 15 minutes.

Instructions: Diagnose the following prompt using PCAFE-V: "Explain the theory of evolution." Identify what is missing for each element.

Show diagnosis
  • Purpose: Vague — why is this being asked? For a class, a presentation, personal interest?
  • Context: Missing — what level of detail? What sources should be used?
  • Audience: Undefined — who is the explanation for?
  • Format: Vague — how long? What structure?
  • Evaluation Criteria: None — what makes a good explanation?
  • Verification: None — what sources should support the explanation?
📝 Exercise 4: Revise a Prompt Using All Six Elements
  • Learning Outcome: LO 3, 4.
  • Estimated Time: 15 minutes.

Instructions: Revise the prompt from Exercise 3 ("Explain the theory of evolution") into a complete PCAFE-V prompt. Include all six elements.

Show sample solution 
Purpose: Help me prepare a two-page plain-language explanation of the theory of evolution for high school students who are studying biology.

Context: Use the supplied textbook chapter and peer-reviewed sources. Define "evolution" as "the process by which species change over time through natural selection and genetic variation."

Audience: High school students (grades 10-12) with basic biology knowledge but no prior study of evolution. They need clear, accessible explanations with concrete examples.

Format: Two-page explanatory note with sections: (1) What is Evolution? (2) How Does Natural Selection Work? (3) Evidence for Evolution (4) Common Misconceptions. Use examples and simple diagrams described in text. Under 1000 words.

Evaluation Criteria:
- Clarity: Plain language; technical terms defined.
- Completeness: Covers all four sections.
- Evidence: Uses examples from the supplied textbook.
- Accessibility: Suitable for high school readers.
- Accuracy: Correct and up to date.
- Uncertainty: Note areas of ongoing scientific debate.

Verification Requirements:
- All claims must be traceable to the supplied textbook or peer-reviewed sources.
- Include page references for key claims.
- Mark the output as a draft requiring teacher review.
- Do not invent examples or data.
📝 Exercise 5: Compare Prompt Versions and Identify Remaining Limitations
  • Learning Outcome: LO 5, 6.
  • Estimated Time: 15 minutes.

Instructions: Using the initial and revised prompts from the worked example (public transit policy), use the evaluation criteria to compare the outputs. Identify improvements, regressions, and remaining limitations.

Show sample comparison

Improvements: Relevance (directly addresses policy), clarity (plain language), completeness (all sections), source support (page references), uncertainty (draft status).

Regressions: The revised output is slightly longer than the initial output, but this is acceptable given the added detail.

Remaining limitations: Some policy details are still vague; legal and financial implications are not addressed; the draft still requires independent verification against the policy document.

15. Homework Problems

Formative · Ungraded · Recommended for portfolio preparation and Assignment 1 evidence.

🧠 Problem 1: Source-Requirement

Learning Outcome: LO 1. Estimated Time: 5 minutes.

Instructions: Write a source-requirement instruction for a prompt that asks the AI to "provide a briefing on a proposed new law."

Show sample solution

"All claims about the proposed law must be traceable to the official bill text or legislative summary. Include section numbers and page references. If a claim is not in the bill text, mark it as commentary."

🧠 Problem 2: No-Fabrication

Learning Outcome: LO 1. Estimated Time: 5 minutes.

Instructions: Write a no-fabrication rule for a prompt that asks the AI to "write a research abstract."

Show sample solution

"Do not invent data, citations, or findings. All results and references must be drawn from the supplied research materials. If information is missing, indicate what is needed."

🧠 Problem 3: Current-Information

Learning Outcome: LO 1. Estimated Time: 5 minutes.

Instructions: Write a current-information requirement for a prompt that asks the AI to "summarize recent developments in a field."

Show sample solution

"All claims about recent developments must be supported by sources published within the last three years. If the most recent source is older, indicate this limitation. Include the date of each source."

🧠 Problem 4: Uncertainty

Learning Outcome: LO 2. Estimated Time: 5 minutes.

Instructions: Write an uncertainty rule for a prompt that asks the AI to "forecast economic trends."

Show sample solution

"Use ranges to express uncertainty (e.g., 'growth is projected between 2% and 4%'). State the assumptions underlying each projection. Clearly distinguish between 'forecast,' 'projection,' and 'speculation.' Mark the output as preliminary."

🧠 Problem 5: Conflict-Handling

Learning Outcome: LO 2. Estimated Time: 5 minutes.

Instructions: Write a conflict-handling rule for a prompt that asks the AI to "summarize research on a controversial topic."

Show sample solution

"When sources present conflicting findings, present both perspectives and note the degree of consensus. Do not cherry-pick evidence to support a preferred conclusion. Indicate the quality of evidence on each side."

🧠 Problem 6: PCAFE-V Diagnosis

Learning Outcome: LO 4. Estimated Time: 10 minutes.

Instructions: Diagnose the following prompt using PCAFE-V: "Help me write a business plan." Identify what is missing for each element.

Show sample solution
  • Purpose: Vague — what is the business? What is the plan for?
  • Context: Missing — what industry? What stage?
  • Audience: Undefined — who is the plan for?
  • Format: Vague — what sections? What length?
  • Evaluation Criteria: None — what makes a good business plan?
  • Verification: None — what sources? What assumptions?
🧠 Problem 7: PCAFE-V Diagnosis

Learning Outcome: LO 4. Estimated Time: 10 minutes.

Instructions: Diagnose the following prompt using PCAFE-V: "Make a list of healthy recipes." Identify what is missing for each element.

Show sample solution
  • Purpose: Vague — for whom? For what dietary needs?
  • Context: Missing — what is "healthy" in this context?
  • Audience: Undefined — who is the list for?
  • Format: Vague — how many recipes? What structure?
  • Evaluation Criteria: None — what makes a recipe "healthy"?
  • Verification: None — what sources for nutritional information?
🧠 Problem 8: Complete Prompt Redesign

Learning Outcome: LO 3, 4. Estimated Time: 15 minutes.

Instructions: Redesign the prompt from Problem 7 ("Make a list of healthy recipes") into a complete PCAFE-V prompt. Include all six elements.

Show sample solution 
Purpose: Help me prepare a list of 10 healthy dinner recipes for a family of four, with a focus on meals that are quick (under 30 minutes) and use common ingredients.

Context: "Healthy" means balanced nutrition (protein, vegetables, whole grains) and under 500 calories per serving. Recipes should be suitable for busy weeknights.

Audience: My family (two adults, two children aged 8 and 10). They need simple, appealing meals that are easy to prepare.

Format: A table with columns: Recipe Name, Key Ingredients, Prep Time, Calories per Serving, and Dietary Notes (e.g., gluten-free, vegetarian).

Evaluation Criteria:
- Relevance: Recipes must be quick, family-friendly, and healthy.
- Clarity: Instructions must be easy to follow.
- Completeness: All 10 recipes must have all columns filled.
- Accuracy: Nutritional information must be reasonable.
- Variety: Include different proteins, cuisines, and vegetables.

Verification Requirements:
- Nutritional information must be approximate and based on typical recipes.
- If a recipe requires special equipment, note it.
- Mark the list as a draft for family review.
- Do not invent recipes or nutritional data.
🧠 Problem 9: Prompt Comparison

Learning Outcome: LO 5. Estimated Time: 10 minutes.

Instructions: Using the initial prompt from Problem 7 and the revised prompt from Problem 8, compare the outputs using common criteria. Identify improvements, regressions, and remaining limitations.

Show sample comparison

Criteria: Relevance, clarity, completeness, accuracy, variety.

Improvements: The revised output is more relevant (quick, family-friendly), clearer (table format), more complete (10 recipes with all columns), and more accurate (nutritional notes).

Regressions: The revised output may be less creative or more constrained.

Remaining limitations: Nutritional accuracy still requires independent verification. Family preferences may vary.

🧠 Problem 10: Limitation-Identification

Learning Outcome: LO 6. Estimated Time: 5 minutes.

Instructions: Even with a complete PCAFE-V prompt, what limitations remain? Identify at least three limitations that no prompt can eliminate.

Show sample solution
  1. The AI may still generate inaccurate information if the supplied sources are unreliable.
  2. The AI cannot independently verify facts; it depends on the human to check.
  3. The AI cannot exercise professional judgment or ethical decision-making.
  4. The AI cannot understand context in the way a human does.
  5. The AI is not accountable for its outputs; the human remains responsible.
🧠 Problem 11: Documentation

Learning Outcome: LO 9. Estimated Time: 10 minutes.

Instructions: Create a prompt documentation record for the revised prompt from Problem 8. Include: prompt identifier, date, AI system, purpose, prompt version, context category, output, evaluation, verification needs, revision, human decision, and AI-use disclosure.

Show sample solution
  • Prompt ID: PCAFE-REC-001
  • Date: 2026-08-29
  • AI System: [Institutionally approved system]
  • Purpose: Generate 10 healthy family dinner recipes.
  • Prompt Version: v1.0
  • Context Category: Personal, family meal planning
  • Output: Table of 10 recipes with nutritional notes.
  • Evaluation: Meets criteria; some nutritional values approximate.
  • Verification Needs: Verify nutritional information with a reliable source.
  • Revision: Initial version; may revise based on family feedback.
  • Human Decision: Use as a starting point; family to review.
  • AI-Use Disclosure: AI used as a recipe assistant; final selection by family.
🧠 Problem 12: Transfer Problem

Learning Outcome: LO 8. Estimated Time: 10 minutes.

Instructions: Apply the complete PCAFE-V framework to a task from your own life or work. Write a complete prompt that includes all six elements.

Show sample solution design

Task: Planning a team meeting agenda.

  • Purpose: Help me prepare a 60-minute team meeting agenda for a weekly project update meeting.
  • Context: The team has 8 members. The meeting covers project status, blockers, and next steps.
  • Audience: Team members who need a clear, concise agenda to prepare for the meeting.
  • Format: Agenda with sections: (1) Welcome and Check-in, (2) Project Updates, (3) Blockers and Risks, (4) Next Steps. Include time allocations.
  • Evaluation Criteria: Clarity, completeness, time feasibility, actionability.
  • Verification: Mark as a draft; team members should verify time allocations and add their own topics.

16. Mini‑Project: Complete PCAFE-V Prompt and Evaluation Record

📁 Complete PCAFE-V Prompt and Evaluation Record

Estimated Effort: 60–90 minutes.

Context: Extend the mini-project from Lessons 4.1, 4.2, and 4.3. Use the same task you selected there.

Deliverable: Create a document with the following components:

  1. Context and human problem: Describe the situation.
  2. Legitimate purpose: State the human objective.
  3. AI-supported task: What the AI will do.
  4. Intended use: How the output will be used.
  5. Initial prompt: The initial draft (without PCAFE-V).
  6. Initial output: The generated output (or supplied).
  7. PCAFE-V diagnosis: Identify what is missing for each element.
  8. Minimized context: Necessary context only.
  9. Audience profile: Detailed audience description.
  10. Output format: Specific format specification.
  11. Evaluation criteria: 5-8 observable criteria.
  12. Verification requirements: Source, uncertainty, conflict, no-fabrication.
  13. Revised prompt: Complete PCAFE-V prompt.
  14. Revised output: The output (or supplied).
  15. Side-by-side comparison: Compare initial and revised outputs using criteria.
  16. Claims requiring verification: List of claims to check.
  17. Remaining limitations: What still needs improvement.
  18. Human decisions: What you decide to do with the output.
  19. Final assessment: Is the output fit for purpose?
  20. Prompt documentation record: Documentation as described.
  21. Material AI-use disclosure: Transparency statement.

Self-Assessment Criteria:

  • Is the purpose legitimate, clear, and human-centred?
  • Is the context minimized and appropriate?
  • Is the audience defined with sufficient detail?
  • Is the format specified clearly?
  • Are the evaluation criteria observable and relevant?
  • Are the verification requirements comprehensive?
  • Is the revised prompt a clear improvement over the initial?
  • Are remaining limitations identified honestly?
  • Is the human review plan realistic?
  • Is the documentation complete?
Show staged hints

Hint 1: Start with the work you have already done in Lessons 4.1, 4.2, and 4.3. You already have purpose, context, audience, format, and criteria. Now add verification.

Hint 2: For verification, think about what you would need to check before trusting the output. What sources should be used? What uncertainties exist?

Hint 3: The comparison should be honest. If the revised output is better in some ways but worse in others, note that.

Hint 4: The human decision should be concrete: Will you use the output? With what caveats? What will you do next?

Hint 5: The documentation record helps you track your work and supports transparency if you later need to disclose AI use.

Show solution architecture
  1. Context and human problem: [Description]
  2. Legitimate purpose: [Statement]
  3. AI-supported task: [Description]
  4. Intended use: [Description]
  5. Initial prompt: [Text]
  6. Initial output: [Text]
  7. PCAFE-V diagnosis: [List missing elements]
  8. Minimized context: [Text]
  9. Audience profile: [Description]
  10. Output format: [Specification]
  11. Evaluation criteria: [List]
  12. Verification requirements: [List]
  13. Revised prompt: [Complete PCAFE-V prompt]
  14. Revised output: [Text]
  15. Side-by-side comparison: [Table or list]
  16. Claims requiring verification: [List]
  17. Remaining limitations: [List]
  18. Human decisions: [Description]
  19. Final assessment: [Statement]
  20. Prompt documentation record: [Record]
  21. AI-use disclosure: [Statement]
Show illustrative partial example (different context)

Context: Professional development workshop (from Lesson 4.3 partial example).

Verification requirements:

  • All workshop design claims must be traceable to the supplied best-practice resources.
  • If an activity is suggested without a source, mark it as "proposed activity requiring review."
  • If time estimates are uncertain, use ranges.
  • Do not invent activities or time allocations.
  • Mark the agenda as a draft requiring team review.

Revised prompt (complete PCAFE-V): "Help me design a half-day workshop on communication skills for a team of 15. Use the supplied best-practice resources. The agenda should include modules on active listening, feedback, and communication styles. Format: a 4-hour agenda with 5-6 modules, each with a topic, activity description, and time allocation. Evaluation criteria: relevance, clarity, completeness, time feasibility, audience suitability. Verification: all activities must be traceable to resources; mark uncertain time estimates as tentative. Mark as a draft requiring team review."

17. Self‑Check Quiz

Test your understanding of the key concepts from this lesson.

Question 1: What does the V in PCAFE-V stand for?

Select one:
Show answer

Correct answer: Verification.

Explanation: PCAFE-V stands for Purpose, Context, Audience, Format, Evaluation Criteria, and Verification Requirements.

Question 2: Why is "verify your answer" inadequate as a verification instruction?

Select one:
Show answer

Correct answer: It is not independent verification; the AI cannot check its own sources.

Explanation: The AI does not have access to external reality, cannot distinguish between accurate and fabricated information, and cannot be held accountable. Verification is a human responsibility.

Question 3: What should a prompt require when sources conflict?

Select one:
Show answer

Correct answer: Present both perspectives and note the disagreement.

Explanation: Responsible handling of conflicting evidence requires transparency. The AI should identify the conflict and present both sides, not cherry-pick evidence.

Question 4: Which of the following is a valid verification requirement?

Select one:
Show answer

Correct answer: All factual claims must be traceable to supplied sources.

Explanation: This is an observable, checkable requirement. The others are vague value judgments.

Question 5: What is the purpose of marking output as "provisional" or "draft"?

Select one:
Show answer

Correct answer: To indicate it requires human review and verification.

Explanation: Marking output as provisional reminds users that it is not final and requires checking before use. It preserves human accountability.

Question 6: Which of the following is a common misconception about prompt improvement?

Select one:
Show answer

Correct answer: A detailed prompt guarantees accuracy.

Explanation: This is a misconception. Detail improves clarity and reviewability but does not ensure correctness. The AI may still generate inaccurate information.

Question 7: What does the no-fabrication rule require?

Select one:
Show answer

Correct answer: The AI must not invent citations, data, or facts.

Explanation: The no-fabrication rule is essential to prevent hallucinations and ensure that the output is based on real sources or explicitly marked as unverified.

Question 8: What is the final step in the prompt refinement cycle?

Select one:
Show answer

Correct answer: Record the human decision.

Explanation: The cycle ends with the human recording their decision about the output's fitness for purpose, what actions to take, and what remains to be verified.

18. Assignment 2 Portfolio Connection

Assignment 2: Prompt Engineering Portfolio requires you to submit three complete prompt-engineering entries. For each entry, you will need to document your verification requirements and show the complete PCAFE-V prompt.

For Lesson 4.4: Save the following for each of your three portfolio contexts (educational, professional, and personal):

  • Verification requirements (source identification, uncertainty, conflict, no-fabrication);
  • Complete PCAFE-V prompt (all six elements);
  • Initial and revised outputs with comparison using common criteria;
  • Remaining limitations and human decisions;
  • Prompt documentation record and AI-use disclosure.

Use the cumulative mini-project from this lesson as the foundation for your portfolio entry. The work you do here will be submitted as part of Assignment 2 at the end of Module 2.

⚠️ Important

Do not use real personal, confidential, or proprietary information. Your portfolio should demonstrate your prompt-design skills, not share private data.

19. Assignment 1 Journal Connection

📝 Unit 4 Journal Prompts

Assignment 1: AI Learning and Reflection Journal is an ongoing reflective journal throughout the course. For Unit 4, consider the following prompts related to this lesson.

  1. How did adding verification requirements change the prompt? Did it make the output more trustworthy?
  2. What challenges did you face in specifying verification requirements? How did you address them?
  3. How did you handle missing information or uncertainty in your prompts? What did you learn?
  4. What misconceptions about prompt improvement did you identify in your own thinking or in others?
  5. How did the side-by-side comparison of initial and revised outputs help you understand the value of PCAFE-V?
  6. What limitations remained after you refined your prompt? What does this tell you about the role of human judgment?
  7. How has your understanding of "responsible AI use" changed after completing Unit 4?
Show reflection-planning checklist
  • Choose 2–3 prompts that are most relevant to your experience.
  • Write 150–250 words per selected prompt.
  • Be specific about what you did, learned, and would do differently.
  • Connect your reflection to specific concepts (verification, uncertainty, no-fabrication, human responsibility).
  • Include concrete examples from your work (e.g., your cumulative project, exercises, or homework).
  • Identify any remaining questions or areas of uncertainty.

20. Key Takeaways

21. Key Terms

Verification requirements
Instructions specifying which claims, sources, calculations, dates, assumptions, conflicts, and uncertainties must be checked before the output is used.
Source identification
Linking claims to specific, traceable sources (e.g., page numbers, URLs, DOIs).
Source traceability
The ability to trace a claim back to its original source for verification.
No-fabrication rule
A requirement that the AI must not invent citations, data, facts, or sources.
Provisional output
An output marked as a draft or preliminary version that requires human review and verification before use.
Uncertainty
The degree of confidence in a claim, which should be stated explicitly in the output.
Conflicting evidence
Information from different sources that contradicts or disagrees with other information.
Prompt refinement cycle
The iterative process of defining purpose, drafting, evaluating, diagnosing, revising, comparing, and recording human decisions.
PCAFE-V framework
The six-element framework for responsible prompt design: Purpose, Context, Audience, Format, Evaluation Criteria, and Verification Requirements.
Prompt documentation
The record of prompt versions, outputs, evaluations, verification needs, and human decisions.

22. Looking Ahead: Unit 5

Congratulations! You have completed all four lessons in Unit 4. You have learned to define purpose, context, audience, format, evaluation criteria, and verification requirements. You can now construct a complete PCAFE-V prompt, diagnose weaknesses, refine prompts purposefully, and identify remaining limitations.

In Unit 5: Advanced Prompting and Context Engineering, you will extend these skills to more complex workflows:

Unit 5 will build directly on the foundations you have established in Unit 4. You will also begin work on Assignment 2: Prompt Engineering Portfolio, which will be submitted at the end of Module 2.

🎉 End of Unit 4

You are now well prepared for the advanced prompting techniques in Unit 5.

Continue to Unit 4 Consolidation Return to Unit 4 Home

23. Sources & Further Reading

For additional reading on verification and accountability, consult your institution's AI-use policy, professional standards in your field, and guidance from organizations such as the Responsible AI Institute or the Partnership on AI.

⚠️ Note

Some textbook content, definitions, examples, and figures are marked with placeholders. These will be supplied by the course coordinator or subject-matter expert.

👤