A library helpdesk AI answers overdue-fine questions more consistently after staff paste one example reply into the prompt. Which prompting technique are they using?

Pasting one example overdue-fine reply into the library helpdesk prompt is one-shot prompting: a single worked example guides format, tone, and content without updating model weights. Consistency improves because the model conditions on that demonstration when answering similar fine questions. Zero-shot prompting with no examples included in the prompt fails by concept because the staff explicitly supplied one example—the defining difference from zero-shot. Few-shot prompting that supplies several worked examples fails because the staff pasted only one example, not several demonstrations, so it does not meet the few-shot definition. Chain-of-thought prompting that asks for step-by-step reasoning fails because the staff provided a sample reply, not a request for the model to reason step-by-step before answering. Exam caveat: one-shot quality depends on example correctness; a bad sample can systematically bias every fine answer, and one example may not cover edge cases such as waivers or multi-branch policies. Operational check: curate the official one-shot example under librarian review, version the prompt template, evaluate a suite of overdue scenarios, and escalate novel policy questions to staff rather than forcing a matched pattern.