D685 Practical Applications of Prompt Exam Questions &
Answers
2026/2027 • 105 Original Practice Questions • AI, Machine Learning, Generative AI, LLMs, Prompt Engineering & AI Ethics
Important: This study guide contains newly authored practice questions aligned to the requested D685 subject areas. It does not
reproduce or claim to contain leaked, proprietary, or actual WGU examination questions.
1. Which prompting technique improves performance on a complex task by showing worked examples
before a new task?
■ A. Zero-shot prompting
■ B. Few-shot prompting
■ C. Random sampling
■ D. Token truncation
Correct Answer: B. Few-shot prompting
Rationale: Few-shot prompting gives examples that demonstrate the desired task pattern.
2. A prompt specifies labels but provides no examples. Which approach is this?
■ A. Few-shot prompting
■ B. Chain-of-thought prompting
■ C. Zero-shot prompting
■ D. Retrieval-augmented generation
Correct Answer: C. Zero-shot prompting
Rationale: Zero-shot prompting provides instructions without task-specific examples.
3. Which practice best reduces ambiguity in a technical-summary prompt?
■ A. Use vague wording
■ B. Specify audience, length, scope, and output format
■ C. Remove context
■ D. Ask the model to invent evidence
Correct Answer: B. Specify audience, length, scope, and output format
Rationale: Explicit requirements make the requested output more predictable and useful.
4. What is the primary purpose of retrieval-augmented generation (RAG)?
■ A. Eliminate prompts
■ B. Ground generation using retrieved external information
■ C. Increase parameter count
■ D. Guarantee truth
Correct Answer: B. Ground generation using retrieved external information
Rationale: RAG supplies retrieved information as context; it improves grounding but does not guarantee truth.
D685 Practical Applications • Page 1
,5. An LLM produces a plausible but unsupported statement. What is this commonly called?
■ A. Hallucination
■ B. Overfitting
■ C. Encryption
■ D. Tokenization
Correct Answer: A. Hallucination
Rationale: A hallucination is unsupported, fabricated, or ungrounded generated content.
6. Which prompt is best when another application must parse the response?
■ A. Write something useful
■ B. Return valid JSON using specified field names
■ C. Be creative
■ D. Use any format
Correct Answer: B. Return valid JSON using specified field names
Rationale: An explicit schema makes output easier for downstream software to parse.
7. What does temperature generally control in a generative language model?
■ A. Physical temperature
■ B. Randomness or variability in token selection
■ C. Training documents
■ D. Authentication level
Correct Answer: B. Randomness or variability in token selection
Rationale: Temperature adjusts sampling randomness; higher values generally produce more variation.
8. Which statement best describes an LLM?
■ A. A database of every answer
■ B. A neural model trained on large amounts of text to generate language
■ C. A spreadsheet engine
■ D. A rules-only system
Correct Answer: B. A neural model trained on large amounts of text to generate language
Rationale: LLMs learn statistical patterns in language and generate text from context.
9. A prompt says: identify requirements, draft an answer, then check it against requirements. What
principle is used?
■ A. Decomposition and self-checking
■ B. Data deletion
■ C. Network segmentation
■ D. Password hashing
Correct Answer: A. Decomposition and self-checking
Rationale: Breaking work into stages and checking the result can improve reliability.
D685 Practical Applications • Page 2
, 10. What risk is most directly associated with placing sensitive personal information in an external AI
prompt?
■ A. Privacy and data governance
■ B. Screen resolution
■ C. CPU speed
■ D. File compression
Correct Answer: A. Privacy and data governance
Rationale: Sensitive prompt data can create confidentiality, retention, privacy, and governance risks.
11. What is prompt injection?
■ A. Physical installation of a model
■ B. Crafted input intended to manipulate an AI system into violating intended instructions
■ C. Increasing RAM
■ D. Backing up files
Correct Answer: B. Crafted input intended to manipulate an AI system into violating intended instructions
Rationale: Prompt injection attempts to influence model behavior through adversarial instructions or content.
12. How should retrieved webpage text be treated in an AI application?
■ A. As higher-priority instructions
■ B. As data separated from trusted application instructions
■ C. As a replacement for system instructions
■ D. As automatically trustworthy
Correct Answer: B. As data separated from trusted application instructions
Rationale: Retrieved content should generally be treated as untrusted data, with instruction boundaries and validation.
13. Which metric measures the proportion of actual positive cases correctly identified?
■ A. Recall
■ B. Latency
■ C. Token count
■ D. Temperature
Correct Answer: A. Recall
Rationale: Recall is true positives divided by all actual positives.
14. What does precision measure?
■ A. The proportion of predicted positives that are actually positive
■ B. Response latency
■ C. Model size
■ D. Recall exactly
Correct Answer: A. The proportion of predicted positives that are actually positive
Rationale: Precision is TP divided by TP plus FP.
D685 Practical Applications • Page 3
Answers
2026/2027 • 105 Original Practice Questions • AI, Machine Learning, Generative AI, LLMs, Prompt Engineering & AI Ethics
Important: This study guide contains newly authored practice questions aligned to the requested D685 subject areas. It does not
reproduce or claim to contain leaked, proprietary, or actual WGU examination questions.
1. Which prompting technique improves performance on a complex task by showing worked examples
before a new task?
■ A. Zero-shot prompting
■ B. Few-shot prompting
■ C. Random sampling
■ D. Token truncation
Correct Answer: B. Few-shot prompting
Rationale: Few-shot prompting gives examples that demonstrate the desired task pattern.
2. A prompt specifies labels but provides no examples. Which approach is this?
■ A. Few-shot prompting
■ B. Chain-of-thought prompting
■ C. Zero-shot prompting
■ D. Retrieval-augmented generation
Correct Answer: C. Zero-shot prompting
Rationale: Zero-shot prompting provides instructions without task-specific examples.
3. Which practice best reduces ambiguity in a technical-summary prompt?
■ A. Use vague wording
■ B. Specify audience, length, scope, and output format
■ C. Remove context
■ D. Ask the model to invent evidence
Correct Answer: B. Specify audience, length, scope, and output format
Rationale: Explicit requirements make the requested output more predictable and useful.
4. What is the primary purpose of retrieval-augmented generation (RAG)?
■ A. Eliminate prompts
■ B. Ground generation using retrieved external information
■ C. Increase parameter count
■ D. Guarantee truth
Correct Answer: B. Ground generation using retrieved external information
Rationale: RAG supplies retrieved information as context; it improves grounding but does not guarantee truth.
D685 Practical Applications • Page 1
,5. An LLM produces a plausible but unsupported statement. What is this commonly called?
■ A. Hallucination
■ B. Overfitting
■ C. Encryption
■ D. Tokenization
Correct Answer: A. Hallucination
Rationale: A hallucination is unsupported, fabricated, or ungrounded generated content.
6. Which prompt is best when another application must parse the response?
■ A. Write something useful
■ B. Return valid JSON using specified field names
■ C. Be creative
■ D. Use any format
Correct Answer: B. Return valid JSON using specified field names
Rationale: An explicit schema makes output easier for downstream software to parse.
7. What does temperature generally control in a generative language model?
■ A. Physical temperature
■ B. Randomness or variability in token selection
■ C. Training documents
■ D. Authentication level
Correct Answer: B. Randomness or variability in token selection
Rationale: Temperature adjusts sampling randomness; higher values generally produce more variation.
8. Which statement best describes an LLM?
■ A. A database of every answer
■ B. A neural model trained on large amounts of text to generate language
■ C. A spreadsheet engine
■ D. A rules-only system
Correct Answer: B. A neural model trained on large amounts of text to generate language
Rationale: LLMs learn statistical patterns in language and generate text from context.
9. A prompt says: identify requirements, draft an answer, then check it against requirements. What
principle is used?
■ A. Decomposition and self-checking
■ B. Data deletion
■ C. Network segmentation
■ D. Password hashing
Correct Answer: A. Decomposition and self-checking
Rationale: Breaking work into stages and checking the result can improve reliability.
D685 Practical Applications • Page 2
, 10. What risk is most directly associated with placing sensitive personal information in an external AI
prompt?
■ A. Privacy and data governance
■ B. Screen resolution
■ C. CPU speed
■ D. File compression
Correct Answer: A. Privacy and data governance
Rationale: Sensitive prompt data can create confidentiality, retention, privacy, and governance risks.
11. What is prompt injection?
■ A. Physical installation of a model
■ B. Crafted input intended to manipulate an AI system into violating intended instructions
■ C. Increasing RAM
■ D. Backing up files
Correct Answer: B. Crafted input intended to manipulate an AI system into violating intended instructions
Rationale: Prompt injection attempts to influence model behavior through adversarial instructions or content.
12. How should retrieved webpage text be treated in an AI application?
■ A. As higher-priority instructions
■ B. As data separated from trusted application instructions
■ C. As a replacement for system instructions
■ D. As automatically trustworthy
Correct Answer: B. As data separated from trusted application instructions
Rationale: Retrieved content should generally be treated as untrusted data, with instruction boundaries and validation.
13. Which metric measures the proportion of actual positive cases correctly identified?
■ A. Recall
■ B. Latency
■ C. Token count
■ D. Temperature
Correct Answer: A. Recall
Rationale: Recall is true positives divided by all actual positives.
14. What does precision measure?
■ A. The proportion of predicted positives that are actually positive
■ B. Response latency
■ C. Model size
■ D. Recall exactly
Correct Answer: A. The proportion of predicted positives that are actually positive
Rationale: Precision is TP divided by TP plus FP.
D685 Practical Applications • Page 3
