Before composing a prompt, what should the writer decide first?
A The model's hidden weights
B A decorative emoji
C The intended task outcome
D The browser window size
Answer: C. The intended task outcome. A prompt should begin with a clear outcome so its instructions and context can be chosen for that goal.
Which prompt element explains why the task matters?
A A random example
B Relevant context
C The output font
D A secret key
Answer: B. Relevant context. Context explains the situation or purpose that helps the model tailor its response.
How can an audience description improve a prompt?
A Change the model version
B Increase file storage
C Guarantee a true answer
D Set suitable detail and tone
Answer: D. Set suitable detail and tone. Audience information guides vocabulary, depth, and tone, but does not guarantee factual accuracy.
Why specify the desired output format?
A Make responses easier to use
B Increase training data
C Remove every limitation
D Create more context
Answer: A. Make responses easier to use. Naming a format such as a table or checklist reduces uncertainty about how the response should be organized.
Which instruction is more actionable for a summary task?
A Use three bullets, keep caveats
B Write a polished summary
C Write something useful
D Use an impressive style
Answer: A. Use three bullets, keep caveats. Specific requirements are easier to follow and assess than vague requests for quality.
What helps prevent a prompt from asking for conflicting results?
A Add more unrelated examples
B Hide the final goal
C Review its constraints together
D Use every format
Answer: C. Review its constraints together. Checking constraints together can reveal contradictions, such as requesting both a one-sentence answer and extensive detail.
How should a prompt describe a task that has several ordered stages?
A Mix all goals together
B List the stages clearly
C Remove the final output
D Use unrelated context
Answer: B. List the stages clearly. Clear ordering helps distinguish steps and their expected outputs.
When a prompt contains both instructions and source text, what improves clarity?
A Repeat everything twice
B Remove the task
C Put everything in one sentence
D Separate them with labels
Answer: D. Separate them with labels. Labels or delimiters can help distinguish the task instructions from the material to process.
Do delimiters alone make untrusted source text safe from prompt injection?
A No, applications need controls
B Yes, always
C Only for PDFs
D Only for short inputs
Answer: A. No, applications need controls. Delimiters may clarify structure but are not a complete security boundary; applications need additional controls.
What should a prompt say when the source lacks required information?
A Invent a likely value
B Hide the gap
C State what is missing
D Use unrelated facts
Answer: C. State what is missing. A prompt can ask the model to identify missing information rather than fill gaps with unsupported guesses.
Which prompting approach uses no examples?
A One-shot prompting
B Few-shot prompting
C Fine-tuning
D Zero-shot prompting
Answer: D. Zero-shot prompting. Zero-shot prompting gives an instruction without adding worked examples.
What does one-shot prompting provide?
A No task instruction
B One illustrative example
C Many training epochs
D A permanent model update
Answer: B. One illustrative example. One-shot prompting includes one example to demonstrate the desired pattern or output.
How does few-shot prompting guide a model?
A Changes model weights
B Expands the context automatically
C Shows several example pairs
D Removes the user's input
Answer: C. Shows several example pairs. Few-shot prompts include examples that demonstrate the task or response pattern.
Why should prompt examples resemble the real use case?
A They guide relevant behavior
B They change model licensing
C They prove every answer
D They replace evaluation
Answer: A. They guide relevant behavior. Relevant examples better show the desired scope, vocabulary, and format for the actual task.
What can happen when every example follows the same narrow pattern?
A The model gains more facts
B The model may overgeneralize
C The prompt becomes private
D The output is guaranteed
Answer: B. The model may overgeneralize. Narrow examples can encourage an unintended pattern, so examples should cover meaningful variations.
Why include edge cases in a set of prompt examples?
A Increase response length
B Hide the main task
C Remove all constraints
D Show how exceptions should work
Answer: D. Show how exceptions should work. Edge cases demonstrate how the task should handle uncommon or boundary inputs.
Which positive instruction is clearer than saying only what to avoid?
A Do not be bad
B Avoid all problems
C Never make mistakes
D Return a neutral, factual summary
Answer: D. Return a neutral, factual summary. Positive directions state the behavior wanted, while broad prohibitions may leave the target unclear.
How can a writer set a useful response-length constraint?
A Ask for appropriate length
B Use more examples
C Give a measurable limit
D Mention several audiences
Answer: C. Give a measurable limit. A measurable limit, such as a word range, is easier to follow than an undefined request for brevity.
What should a prompt specify when requesting JSON output?
A Required keys and value types
B Only the file name
C A preferred background color
D A public password
Answer: A. Required keys and value types. Naming required keys and expected value types makes the requested structure clearer and easier to validate.
Can an instruction requesting valid JSON guarantee parsable output?
A Yes, always
B No, validate the result
C Only for short prompts
D Only with examples
Answer: B. No, validate the result. Models may still produce malformed or extra text, so software should validate structured output.
Why include an example of the expected answer format?
A Increase model accuracy automatically
B Change tokenization
C Make the structure concrete
D Replace user context
Answer: C. Make the structure concrete. A sample output makes formatting expectations concrete, especially for repeated or structured tasks.
Which prompt instruction helps an assistant distinguish facts from assumptions?
A Label assumptions separately
B Hide uncertain details
C Use a confident tone
D Remove source notes
Answer: A. Label assumptions separately. Separating assumptions from sourced facts helps readers see what is known and what is inferred.
When asking for an answer grounded in supplied documents, what should be requested?
A Add unsupported background
B Ignore the source text
C Use memory only
D Cite the relevant passages
Answer: D. Cite the relevant passages. Asking for relevant passages makes it easier to check whether the response is grounded in the supplied material.
What should a prompt require when source documents disagree?
A Choose one without explanation
B Describe the disagreement
C Average their statements
D Ignore both sources
Answer: B. Describe the disagreement. Reporting conflicting source statements is more transparent than silently selecting one.
How can a prompt reduce ambiguity in a classification task?
A Use shorter labels only
B Add unrelated instructions
C Ask for a confident guess
D Define each category
Answer: D. Define each category. Category definitions help distinguish labels that might otherwise be interpreted differently.
What does a role instruction mainly help establish?
A Perspective or response style
B Verified expertise
C Legal authority
D Access to private data
Answer: A. Perspective or response style. A role can guide tone or perspective, but it does not give the model verified credentials or authority.
Why should a prompt avoid vague pronouns such as it or they?
A They increase token limits
B Their references may be unclear
C They improve source quality
D They change model weights
Answer: B. Their references may be unclear. Clear nouns reduce ambiguity about which document, person, or object an instruction refers to.
When should abbreviations be expanded in a prompt?
A Only in titles
B Never in technical tasks
C When meaning is unclear
D After every sentence
Answer: C. When meaning is unclear. Expanding unfamiliar abbreviations prevents different readers or models from interpreting them differently.
How can long input documents be handled more clearly?
A Label each document and source
B Blend them without headings
C Remove source names
D Repeat the prompt
Answer: A. Label each document and source. Labels help distinguish documents and preserve source context when multiple inputs are included.
What is a useful prompt for an unfamiliar term in source material?
A Replace it with a guess
B Ignore its surrounding text
C Invent a new acronym
D Ask for contextual definition
Answer: D. Ask for contextual definition. Contextual definitions reduce the risk of choosing a meaning unrelated to the passage.
Why should a complex request be broken into smaller prompts when needed?
A Make answers automatically correct
B Reduce task ambiguity
C Avoid all review
D Change training data
Answer: B. Reduce task ambiguity. Smaller stages can make complex tasks easier to direct and review, especially when outputs depend on earlier steps.
Which revision strategy helps identify why a prompt's output changed?
A Rewrite every constraint
B Change prompts and model
C Change one thing per revision
D Remove all examples
Answer: C. Change one thing per revision. Changing one element at a time makes it easier to attribute differences in output to that change.
What is prompt iteration?
A Retraining the model weights
B Deleting old context
C Changing the model's code
D Testing and refining instructions
Answer: D. Testing and refining instructions. Prompt iteration involves trying instructions, reviewing results, and refining the prompt.
Why test a prompt on several representative inputs?
A Check behavior across cases
B Guarantee universal success
C Reduce source diversity
D Hide failure modes
Answer: A. Check behavior across cases. Multiple representative inputs reveal whether a prompt works beyond a single example.
How should prompt versions be compared fairly?
A Change every setting at once
B Select only the best result
C Use the same evaluation inputs
D Ignore failures
Answer: C. Use the same evaluation inputs. Using the same test inputs helps compare prompt changes under more consistent conditions.
What is a prompt evaluation rubric used for?
A Change the model's architecture
B Score outputs against criteria
C Store private passwords
D Create training labels automatically
Answer: B. Score outputs against criteria. A rubric defines criteria such as completeness, factual support, tone, and formatting for consistent review.
Which result suggests a prompt has a repeatability problem?
A Outputs vary on key requirements
B The prompt has a title
C The answer uses paragraphs
D The input contains examples
Answer: A. Outputs vary on key requirements. Variation in important requirements across similar inputs may indicate unclear directions or model variability.
Why should a prompt library be versioned?
A Increase model memory
B Guarantee no drift
C Hide earlier wording
D Track changes and results
Answer: D. Track changes and results. Versioning helps teams identify which prompt changed and compare its effect over time.
When a model changes, what should happen to prompts used in production?
A Assume behavior stays identical
B Reevaluate them on examples
C Delete evaluation records
D Skip user feedback
Answer: B. Reevaluate them on examples. Different model versions can respond differently, so prompts should be checked again against representative tasks.
What privacy practice applies when creating reusable prompt examples?
A Include real customer records
B Share private credentials
C Remove unnecessary personal data
D Keep identifiers for realism
Answer: C. Remove unnecessary personal data. Examples should avoid unnecessary personal information and follow the data rules of the service and organization.
Why avoid placing passwords or API keys in a prompt?
A Secrets improve output quality
B Prompts may expose secrets
C Models need passwords to reason
D Keys reduce token counts
Answer: B. Prompts may expose secrets. Prompts may be processed or retained by a service, so secrets should not be included.
Which instruction is appropriate for uncertain answers?
A Say when evidence is insufficient
B Always choose an answer
C Hide uncertainty
D Invent a citation
Answer: A. Say when evidence is insufficient. Explicitly permitting uncertainty can help avoid unsupported answers, though output still needs review.
How can a prompt request useful clarification before a high-stakes answer?
A Assume all missing facts
B Use the first search result
C Provide a final decision
D Ask when details are missing
Answer: D. Ask when details are missing. Clarifying questions can surface missing details that materially affect an answer.
Why should system-level and user-level instructions be checked for conflicts?
A They change the model license
B They reveal the dataset
C Contradictions can confuse behavior
D They guarantee output structure
Answer: C. Contradictions can confuse behavior. Conflicting instructions can lead to inconsistent responses, so their requirements should align.
What should be done if a prompt has become too long and repetitive?
A Remove redundant instructions
B Add more examples blindly
C Repeat the goal again
D Hide the constraints
Answer: A. Remove redundant instructions. Concise prompts retain useful context while reducing repetition that can obscure the main task.
Which consideration matters when adapting prompts across different AI models?
A Assume identical responses
B Test model-specific behavior
C Reuse all settings unchanged
D Ignore output differences
Answer: B. Test model-specific behavior. Models can interpret instructions differently, so evaluate adapted prompts on the target model.
What should a prompt ask the model to do with conflicting requirements?
A Silently choose one
B Ignore the stated goal
C Identify the conflict before proceeding
D Invent a third requirement
Answer: C. Identify the conflict before proceeding. Surfacing conflicting requirements lets the user resolve them instead of relying on an unannounced assumption.
When is a prompt ready for broader use?
A It worked once
B It sounds professional
C It contains many examples
D It passes defined evaluation criteria
Answer: D. It passes defined evaluation criteria. A prompt should meet defined quality criteria across representative cases before wider use.
How can a prompt writer make output easier to review?
A Request brief rationale or evidence
B Ask for hidden internal reasoning
C Request maximal verbosity
D Remove citations
Answer: A. Request brief rationale or evidence. A concise explanation or cited evidence can support review without requiring inaccessible internal reasoning.
What is a good next step when repeated prompt edits do not improve results?
A Add random instructions
B Assume users are wrong
C Revisit the task design
D Publish without review
Answer: C. Revisit the task design. If iterations do not help, reconsider whether the task, source data, or evaluation criteria are well defined.