CCAO-F Prompting and Context Engineering Practice Question
A company wants to minimize 'hallucinations' when Claude answers questions based on a large internal wiki. Which TWO prompting strategies are recommended to keep the model grounded in the provided text?
⚠ Common exam trap
Candidates assume the model will naturally prioritize accuracy over helpfulness, failing to explicitly instruct the model to admit ignorance or provide evidence, which leads to creative hallucination.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Instruct the model to answer 'I don't know' if the information is not in the context.
Reducing hallucinations requires a combination of structural guidance and behavioral constraints. By giving the model a 'way out' (admitting it doesn't know) and forcing it to cite its sources, you significantly increase the probability that the generated answer is based on the provided context rather than the model's internal weights.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Instruct the model to answer 'I don't know' if the information is not in the context.
Why this is correct
Explicitly giving the model permission to fail is one of the best ways to prevent hallucinations. Without this instruction, LLMs often feel 'pressured' to provide an answer, leading them to fabricate plausible-sounding but incorrect information based on their training data instead of the provided wiki content.
- ✓
Ask the model to provide direct quotes or citations from the text to support its answer.
Why this is correct
Requiring citations forces the model to locate and process the specific relevant text before formulating a response. This grounding mechanism makes it much harder for the model to hallucinate, as it must match its output to the actual strings present in the provided source documents.
- ✗
Use the 'top_p' parameter to limit the model to only the most likely next tokens.
Why it's wrong here
Top-p sampling (nucleus sampling) controls the diversity of the output but does not ensure factual grounding. While it can make the model more 'conservative,' it doesn't stop the model from being confidently wrong if the most likely tokens are derived from its internal knowledge rather than the context.
- ✗
Provide the entire wiki in the system prompt rather than the user message.
Why it's wrong here
Whether the information is in the system prompt or user message does not inherently change the model's tendency to hallucinate. While system prompts are good for general rules, the model's grounding depends more on the specific instructions and constraints provided rather than the message role used for the data.
- ✗
Repeat the core data three times within the prompt to increase its 'weight'.
Why it's wrong here
Repeating data is an inefficient use of tokens and can actually confuse the model, leading to degraded performance. Claude is capable of attending to information effectively without repetition; instead of repeating data, focus on better structure using XML tags and clearer instructions for extraction.
About these practice questions
This CCAO-F question is part of Courseiva's 259-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official Anthropic exam blueprint
This CCAO-F practice question is part of Courseiva's free Anthropic certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the CCAO-F exam.