Housing staff paste the relevant lease FAQ excerpt into the prompt so answers stay grounded. What prompt-crafting basic is demonstrated?

Housing staff pasting the relevant lease FAQ excerpt into the prompt demonstrates grounding: providing source context so answers stay tied to approved civic text. Generative models otherwise fill gaps from general training data, which may conflict with local eligibility, rent rules, or program dates. Including the authoritative excerpt makes the model a drafting assistant over agency-approved language rather than an inventor of housing policy.

Providing source context in the response so civic answers stay grounded in approved text fails because grounding must be supplied in the prompt before generation; text that appears only in the output does not constrain drafting and does not prevent the model from inventing rent rules and eligibility details. Providing source context in the prompt so civic answers stay grounded in model memory fails because model memory reflects general training data, not source grounding; it does not stop the model from inferring eligibility rules that the FAQ never actually states. Providing source context in the prompt so civic answers stay grounded in sample replies fails because sample replies control tone and structure, not content provenance; they do not ground answers in the approved FAQ excerpt or prevent the model from using unauthorized sources like housing blogs or forums.

Exam caveat: grounding reduces but does not eliminate errors; staff must still verify that the excerpt is current and that the answer quotes or paraphrases it faithfully. Operational check: build prompts that insert the latest FAQ section by topic, instruct "answer only from the provided text; say if the excerpt is insufficient," and route edge cases to a housing specialist rather than escalating model speculation.