PROMPT ENGINEERING · GENERATIVE AI
WGU D685
OBJECTIVE ASSESSMENT AND PA
Practical Applications of Prompt Engineering
2026/2027 Official Exam A+
Upper-Level Undergraduate / Professional Practice Examination
A+ 5 100%
QUESTIONS VERIFIED EXAM DOMAINS COVERED RATIONALES INCLUDED
CATEGORIES
1. Foundations of AI & Why Prompt Engineering Matters
2. Anatomy of Effective Prompts
3. Advanced Prompting Techniques
4. Evaluation, Bias, Hallucinations & Iteration
5. Ethics, Applications & Research Optimization
EXAM FOCUS
Prompt components (persona, instructions, context, format) · Zero/few-shot · Chain-of-thought
Hallucinations · Bias types · Iterative evaluation · FATE ethics · Research & business applications
Application, analysis & synthesis level items aligned to WGU D685 competencies
Original practice content for Objective Assessment and Performance Assessment readiness
STUVIAACTUALEXAM
, SECTION 1: Foundations of AI & Why Prompt Engineering Matters
Q1. A project manager asks why the team should invest time in prompt engineering rather than simply typing natural
questions into a large language model. The most accurate professional justification is that well-designed prompts:
A. Guarantee that the model will never produce incorrect information.
B. Improve relevance, consistency, and controllability of outputs for specific professional tasks.
C. Eliminate the need to verify any AI-generated content.
D. Convert a narrow AI system into artificial general intelligence.
Correct Answer: B
Rationale: Prompt engineering shapes model behavior toward task requirements. It does not remove the need for verification or
transform model capabilities into AGI, but it substantially raises output quality and reliability for defined use cases.
Q2. A data analyst is comparing supervised and unsupervised learning. The analyst correctly notes that supervised
learning differs because it:
A. Requires no labeled examples and discovers clusters autonomously.
B. Learns a mapping from inputs to known target labels using training examples.
C. Relies exclusively on reward signals from an environment without any data.
D. Can only be applied to image-generation models.
Correct Answer: B
Rationale: Supervised learning uses labeled input–output pairs to learn a predictive mapping. Unsupervised learning finds structure
without labels; reinforcement learning optimizes via rewards.
Q3. A student describes an AI system that plays chess at a superhuman level but cannot transfer that skill to any other
domain. This system is best classified as:
A. Artificial general intelligence capable of any human intellectual task.
B. A fully conscious agent with human-like understanding.
C. Narrow (weak) AI specialized for a single or limited set of tasks.
D. An unsupervised clustering algorithm.
Correct Answer: C
Rationale: Narrow AI excels at specific tasks such as game playing or image classification. General AI would match or exceed human
flexibility across domains and remains hypothetical.
Q4. A research team notices that a generative model sometimes invents plausible-sounding citations that do not exist.
This phenomenon is known as:
A. Tokenization of the input vocabulary.
B. Successful few-shot learning.
C. Hallucination, in which the model generates content not grounded in its training data or provided context.
D. Deterministic retrieval from a knowledge base.
Correct Answer: C
Rationale: Hallucinations are fluent but factually unsupported outputs. Recognizing them is essential for evaluation and for designing
prompts that reduce their occurrence.
STUVIAACTUALEXAM · Page 2
, Q5. An organization wants to understand the role of context when interacting with large language models. Context is
critical because models:
A. Condition their next-token predictions on the tokens supplied in the current prompt window.
B. Possess permanent long-term memory of every prior conversation across all users.
C. Ignore all surrounding text and respond only to single keywords.
D. Automatically expand every query into a multi-document research report.
Correct Answer: A
Rationale: LLMs generate outputs conditioned on the prompt (and conversation history within the context window). Supplying relevant
context is therefore a primary lever for controlling results.
Q6. A developer claims that because a model was trained on vast internet text it already “knows” everything needed for
any business task. The prompt-engineering specialist should respond that:
A. Training data alone guarantees perfect performance on every specialized task without further guidance.
B. Domain-specific instructions, constraints, and examples are still required to elicit reliable, on-task outputs.
C. Prompt engineering is only useful for image models, not text models.
D. The model will refuse to answer any question outside its training cutoff date.
Correct Answer: B
Rationale: Pre-training provides broad capabilities, yet task success still depends on clear instructions, scope limits, and often
examples or constraints supplied at inference time.
Q7. A team is evaluating whether to use a generative model for customer-support email drafting. The primary limitation
they must design around is that the model:
A. Can never produce grammatical English sentences.
B. May generate plausible but inaccurate or outdated information and lacks real-time access to private company data unless
supplied.
C. Is incapable of following any formatting instructions.
D. Always discloses confidential training data in every response.
Correct Answer: B
Rationale: Generative models can be fluent yet ungrounded. Effective prompts and retrieval or tool use are needed to supply current,
private, or verified information.
Q8. An analyst distinguishes generative AI from traditional discriminative models. Generative AI is characterized by its
ability to:
A. Only classify existing data points into predefined categories.
B. Produce new content such as text, images, or code that resembles the distribution of its training data.
C. Operate exclusively on structured tabular data without language understanding.
D. Require labeled data for every possible output token.
Correct Answer: B
Rationale: Generative models learn to sample new instances from an approximated data distribution, enabling creation of novel text,
images, and other media.
STUVIAACTUALEXAM · Page 3
WGU D685
OBJECTIVE ASSESSMENT AND PA
Practical Applications of Prompt Engineering
2026/2027 Official Exam A+
Upper-Level Undergraduate / Professional Practice Examination
A+ 5 100%
QUESTIONS VERIFIED EXAM DOMAINS COVERED RATIONALES INCLUDED
CATEGORIES
1. Foundations of AI & Why Prompt Engineering Matters
2. Anatomy of Effective Prompts
3. Advanced Prompting Techniques
4. Evaluation, Bias, Hallucinations & Iteration
5. Ethics, Applications & Research Optimization
EXAM FOCUS
Prompt components (persona, instructions, context, format) · Zero/few-shot · Chain-of-thought
Hallucinations · Bias types · Iterative evaluation · FATE ethics · Research & business applications
Application, analysis & synthesis level items aligned to WGU D685 competencies
Original practice content for Objective Assessment and Performance Assessment readiness
STUVIAACTUALEXAM
, SECTION 1: Foundations of AI & Why Prompt Engineering Matters
Q1. A project manager asks why the team should invest time in prompt engineering rather than simply typing natural
questions into a large language model. The most accurate professional justification is that well-designed prompts:
A. Guarantee that the model will never produce incorrect information.
B. Improve relevance, consistency, and controllability of outputs for specific professional tasks.
C. Eliminate the need to verify any AI-generated content.
D. Convert a narrow AI system into artificial general intelligence.
Correct Answer: B
Rationale: Prompt engineering shapes model behavior toward task requirements. It does not remove the need for verification or
transform model capabilities into AGI, but it substantially raises output quality and reliability for defined use cases.
Q2. A data analyst is comparing supervised and unsupervised learning. The analyst correctly notes that supervised
learning differs because it:
A. Requires no labeled examples and discovers clusters autonomously.
B. Learns a mapping from inputs to known target labels using training examples.
C. Relies exclusively on reward signals from an environment without any data.
D. Can only be applied to image-generation models.
Correct Answer: B
Rationale: Supervised learning uses labeled input–output pairs to learn a predictive mapping. Unsupervised learning finds structure
without labels; reinforcement learning optimizes via rewards.
Q3. A student describes an AI system that plays chess at a superhuman level but cannot transfer that skill to any other
domain. This system is best classified as:
A. Artificial general intelligence capable of any human intellectual task.
B. A fully conscious agent with human-like understanding.
C. Narrow (weak) AI specialized for a single or limited set of tasks.
D. An unsupervised clustering algorithm.
Correct Answer: C
Rationale: Narrow AI excels at specific tasks such as game playing or image classification. General AI would match or exceed human
flexibility across domains and remains hypothetical.
Q4. A research team notices that a generative model sometimes invents plausible-sounding citations that do not exist.
This phenomenon is known as:
A. Tokenization of the input vocabulary.
B. Successful few-shot learning.
C. Hallucination, in which the model generates content not grounded in its training data or provided context.
D. Deterministic retrieval from a knowledge base.
Correct Answer: C
Rationale: Hallucinations are fluent but factually unsupported outputs. Recognizing them is essential for evaluation and for designing
prompts that reduce their occurrence.
STUVIAACTUALEXAM · Page 2
, Q5. An organization wants to understand the role of context when interacting with large language models. Context is
critical because models:
A. Condition their next-token predictions on the tokens supplied in the current prompt window.
B. Possess permanent long-term memory of every prior conversation across all users.
C. Ignore all surrounding text and respond only to single keywords.
D. Automatically expand every query into a multi-document research report.
Correct Answer: A
Rationale: LLMs generate outputs conditioned on the prompt (and conversation history within the context window). Supplying relevant
context is therefore a primary lever for controlling results.
Q6. A developer claims that because a model was trained on vast internet text it already “knows” everything needed for
any business task. The prompt-engineering specialist should respond that:
A. Training data alone guarantees perfect performance on every specialized task without further guidance.
B. Domain-specific instructions, constraints, and examples are still required to elicit reliable, on-task outputs.
C. Prompt engineering is only useful for image models, not text models.
D. The model will refuse to answer any question outside its training cutoff date.
Correct Answer: B
Rationale: Pre-training provides broad capabilities, yet task success still depends on clear instructions, scope limits, and often
examples or constraints supplied at inference time.
Q7. A team is evaluating whether to use a generative model for customer-support email drafting. The primary limitation
they must design around is that the model:
A. Can never produce grammatical English sentences.
B. May generate plausible but inaccurate or outdated information and lacks real-time access to private company data unless
supplied.
C. Is incapable of following any formatting instructions.
D. Always discloses confidential training data in every response.
Correct Answer: B
Rationale: Generative models can be fluent yet ungrounded. Effective prompts and retrieval or tool use are needed to supply current,
private, or verified information.
Q8. An analyst distinguishes generative AI from traditional discriminative models. Generative AI is characterized by its
ability to:
A. Only classify existing data points into predefined categories.
B. Produce new content such as text, images, or code that resembles the distribution of its training data.
C. Operate exclusively on structured tabular data without language understanding.
D. Require labeled data for every possible output token.
Correct Answer: B
Rationale: Generative models learn to sample new instances from an approximated data distribution, enabling creation of novel text,
images, and other media.
STUVIAACTUALEXAM · Page 3
