Question

A team built a single large prompt that asks Claude to (1) read a legal contract, (2) extract all obligations, (3) flag risky clauses, and (4) draft a client-facing summary. Output quality is inconsistent, and when a summary is wrong the team cannot tell which step introduced the error. They want to improve reliability AND make it easier to pinpoint where failures occur. Which change best addresses both goals?

Accepted Answer

Split the work into a prompt chain where each step (extract, flag, summarize) is a separate prompt whose output feeds the next, so each subtask can be inspected and validated independently.

Answer

Keep the single prompt but raise the temperature so Claude explores more varied phrasings for the summary.

Answer

Keep the single prompt but add more multishot examples showing the full four-step output at once.

Answer

Switch to a smaller, faster model to reduce latency across the combined four-step task.