Transit customer service prompts require an apology, next-bus ETA fields, and an escalation path when riders are angry. What effective-prompt structure is shown?

Transit customer service prompts that require an apology, next-bus ETA fields, and an escalation path when riders are angry illustrate multi-part structured prompts for resident-facing replies. Upset riders need acknowledgment, operational facts, and a clear path to a human agent—not a shrug or an unauthorized promise. Few-shot exemplar prompts that require apology, operational fields, and escalation for resident-facing replies fail for this use case because showing examples does not satisfy the requirement for enforced multi-part structured sections with apology, ETA, and escalation. Sequential chain-of-thought prompts that require apology, operational fields, and escalation for resident-facing replies miss the point because eliciting step-by-step internal reasoning does not provide the required multi-part structured reply with apology, operational fields, and escalation. Output-constrained schema prompts that require apology, operational fields, and escalation for resident-facing replies fail for this use case because constraining output to a rigid schema alone does not ensure the required resident-facing multi-part reply with apology, live ETA fields, and escalation. Exam caveat: structured fields must be populated from real AVL/CAD or app feeds, not invented ETAs; if data are missing, the prompt should require saying so and escalating rather than guessing. Operational check: maintain a reply template with apology + ETA placeholders + escalation instructions; integrate live ETA where possible; train agents never to let the model invent refunds or schedule guarantees; audit samples weekly for unauthorized commitments.