A prompt is the instruction you give an LLM. A prompt template is a reusable instruction with placeholders for the variables that change between uses. The difference between a property that uses AI well and one that uses it badly is whether they have a prompt template library — versioned, shared, and improved over time — or whether every employee re-invents their prompt from scratch every time.
Anatomy of a good prompt
A good prompt has five parts, in this order:
A prompt missing the role produces inconsistent voice. A prompt missing the voice produces generic AI-sounding output. A prompt missing the context produces responses that sound like they could be from any hotel. A prompt missing the input produces nothing useful at all.
Template versus one-off
A template is the prompt with placeholders: "Respond to the following review for {{PROPERTY_NAME}}, tone {{TONE_GUIDE}}, length {{LENGTH_TARGET}}, context {{RECENT_CONTEXT}}, input {{REVIEW_TEXT}}." When a reservations agent needs to respond to a review, they fill in the placeholders and run the prompt. Time per response: 60 seconds. Quality: consistent across the team.
A one-off prompt is what every employee writes from scratch when they have not been given a template: "hey can you write a polite response to this review, here it is..." Time per response: 4-7 minutes (because the agent has to remember to add tone and context themselves). Quality: variable.
Building your library
Start with the four top use cases — review response, multilingual guest email, F&B copy, pre-arrival template. Author one template per use case. Store them in a shared Google Doc or Notion page. Use a {{VARIABLE}} convention so a team member can spot the placeholders. Review and improve the templates monthly based on what produced good vs. bad output. After 90 days you will have 8-15 templates that handle 80% of the property's AI workload.