Connecting an LLM to Make is the easy part; getting reliable output is the craft. These twelve scenarios start with the Anthropic Claude modules (a raw API call and Create a Message with a system prompt), show a practical spreadsheet workflow where Claude fills in a column, and then walk through the prompt patterns that make answers trustworthy and machine-readable.

The patterns covered: let the model admit uncertainty, ground answers in direct quotes, require citations, define the output format, preload the assistant response to force structure, show an example format, and answer only from a knowledge base. Two further scenarios call the API directly to demonstrate extended thinking and streaming.

The prompts behind each scenario are included as plain text files so you can reuse them in any tool, along with fictional sample data. A start-here guide explains the connection setup and, importantly, that the model names inside the blueprints are retired and must be switched to a current model.