Question

A legal-tech team asks Claude to (1) extract all obligations from a 40-page contract, (2) classify each obligation by risk level, and (3) draft a remediation memo for high-risk items. When they try to do all three in a single prompt, the output frequently skips obligations, mislabels risk, and produces an inconsistent memo. Which approach should the team adopt to improve reliability?

Accepted Answer

Break the work into a prompt chain: one prompt extracts obligations, its output feeds a second prompt for risk classification, and that output feeds a third prompt to draft the memo.

Answer

Raise the temperature so Claude explores more of the contract and catches the obligations it was missing.

Answer

Keep the single prompt but increase max_tokens so the model has enough room to complete all three tasks accurately.

Answer

Switch to a smaller, faster model so each of the three tasks completes before hitting the context window limit.