Prompting With Role Guardrails (Templates You Can Reuse)
Good prompting is less about clever wording and more about guardrails: role limits, source grounding, and explicit “do not guess” instructions.
Guardrail prompt template (staff / paralegal)
Role: You are assisting a legal team. You are not providing legal advice.
Task: Summarize the provided document(s) accurately.
Constraints: Do not speculate. If the document does not contain the answer, say “Not found in the provided text.”
Output: Use headings and bullet points. Provide page/section references when possible.
Guardrail prompt template (attorney)
Role: Act as a legal drafting assistant. Output is a draft for attorney review.
Sources: Use only the materials provided. Do not invent citations.
Verification: Flag any uncertainties or missing sources.
Output: Provide a draft + a checklist of items that must be verified before filing.
Prompt pattern: “Source anchored” drafting
When possible, paste the relevant section or attach the source document and use a prompt like:
- “Use only the text below…”
- “Quote directly with citations to page/paragraph…”
- “If you are unsure, say so and list what would resolve uncertainty…”
Prompt pattern: “Failure-mode checklist”
Before you accept the output, ask the model to critique itself:
List potential errors in your answer. Identify any assumptions you made. Provide a verification checklist for a human reviewer.
What to avoid
- “Just draft the brief” prompts with no sources
- Prompts that include sensitive data unnecessarily
- Prompts that ask the model to decide privilege or to provide definitive legal advice for a client
Practice tip: Your prompt is part of your audit trail. Write prompts the way you would write an instruction to a human assistant.