CCDV-F Prompt and Context Engineering Practice Question
You are building a customer support bot using Claude 3.5 Sonnet. You notice the model sometimes hallucinates policies that do not exist when the user asks about obscure edge cases. Which technique most effectively grounds the model's responses to your internal documentation?
⚠ Common exam trap
Candidates often suggest fine-tuning as the primary solution for grounding, ignoring that fine-tuning is for style and behavior, while RAG is the standard for factual grounding.
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
✓
Implement RAG to dynamically inject relevant documentation snippets into the system prompt at runtime.
Retrieval-Augmented Generation (RAG) is the industry standard for grounding LLMs. By providing context from your source documentation within the prompt, you constrain the model's output to factual, verifiable data. This reduces reliance on training-set knowledge, which might be outdated or insufficient for specific company policies. Effective prompt engineering ensures that the model is instructed to strictly rely on provided context or state ignorance if the answer is unavailable.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Increase the system prompt temperature to 1.0 to ensure more creative and comprehensive answers.
Why it's wrong here
Increasing temperature makes the model more stochastic and prone to hallucination. High temperature is counterproductive when grounding is required, as it reduces the model's focus on the provided context, leading to more erratic generation rather than strict adherence to the source material provided in the prompt.
- ✗
Fine-tune the model on your entire history of support tickets to teach it the specific tone.
Why it's wrong here
Fine-tuning is excellent for style, but it is poor for fact-based grounding. Facts in support tickets can be outdated or incorrect. Relying on fine-tuning for knowledge storage leads to stale information and is much harder to maintain compared to dynamic RAG systems that update in real-time.
- ✓
Implement RAG to dynamically inject relevant documentation snippets into the system prompt at runtime.
Why this is correct
RAG is the most reliable way to ground Claude to specific knowledge. By injecting retrieved documentation into the context window, you provide the model with the exact source of truth, minimizing hallucinations and ensuring responses are current, accurate, and tied directly to the relevant company policy documents.
- ✗
Request that the model adopts a strict 'no hallucination' persona by repeating the instruction five times.
Why it's wrong here
Repetitive prompting is an unreliable mitigation strategy. While system instructions help, they cannot overcome the inherent probabilistic nature of LLMs without actual data grounding. If the model does not have the necessary context, persona constraints are insufficient to prevent it from filling in gaps with plausible-sounding fabrications.
About these practice questions
This CCDV-F question is part of Courseiva's 257-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 CCDV-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 CCDV-F exam.