A developer at Coho Winery is configuring a 'Create generative answers' node in a topic. The node should generate a wine pairing recommendation, but responses must only use content from the winery's official pairing guide and must not speculate beyond it. Which prompt engineering technique should be applied in the node's prompt configuration?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Grounding a model to a specific source is like telling an expert witness 'only speak to what's in the evidence you were given.' Explicit grounding instructions in the prompt prevent the model from inventing answers beyond the knowledge source. The correct answer is B.
Full explanation below image
Full Explanation
## Why B is Correct In the Generative answers node's custom prompt, you can include grounding instructions — explicit directives that constrain the model's response to the provided context. Phrases like 'Only use information from the knowledge source below' and 'If the information is not available, say I do not know' are standard RAG grounding prompts. They reduce hallucination by explicitly forbidding the model from using its parametric knowledge beyond what the retrieved context contains.
## Why the Distractors Are Wrong A (Temperature = 0): Temperature controls the randomness/creativity of generation. Setting it to 0 makes responses more deterministic but does not prevent the model from using information outside the knowledge source — it just makes the response more consistent.
C (Content moderation toggle): Content moderation filters harmful or inappropriate content — it does not enforce factual grounding to a specific document corpus.
D (List pairings as trigger phrases): Trigger phrases are for topic invocation intent-matching. Listing content in trigger phrases does not affect the model's knowledge access or response generation.
## Exam Tip For exam questions about preventing hallucination or limiting AI to a specific source: the answer is prompt-level grounding instructions (explicit constraints in the prompt text), not model parameters or UI toggles.