---
title: "A legal AI prompt library: research, drafting and review prompts"
url: https://notionforlawyers.com/post/ai-prompt-library-legal-research-drafting-review
updated: 2026-10-17
author: Raghav Handa
---

# A legal AI prompt library: research, drafting and review prompts

A legal prompt library is a shared database of tested prompts organised by stage: research, drafting and review. Each entry stores the prompt text with placeholders, the intended use, known limits, the last test date and an owner, so the team reuses what works and retires what does not.

## Key takeaways

- Organise prompts by stage of work, not by tool.
- Store placeholders, limits and a verification step with every prompt.
- Version and date prompts; models change and prompts drift.
- Never store client data inside prompts.

## Database layout

| Property | Purpose |
| --- | --- |
| Name | What the prompt does |
| Stage | Research, Drafting or Review |
| Practice area | Optional tag |
| Prompt text | The reusable text with [placeholders] |
| Limits | What it must not be used for |
| Last tested / Owner | Accountability |


## Research prompts

Issue spotting: I am a [practice area] lawyer in [jurisdiction]. From the anonymised facts below, list research questions I should investigate, ranked by importance. Do not state the law and do not cite any authority. Facts: [paste]

Search plan: Suggest keywords, synonyms and Boolean search strings I could run in a legal research database for this question: [question]. Do not describe results; only propose searches.


## Drafting and review prompts

First draft: Draft a [document type] for a [party A] and [party B]. Include only these terms: [list]. Use bracketed placeholders for anything I have not given you. Add a list of assumptions you made.

Contract review checklist: Review the clause below against this checklist: [your checklist]. For each item answer Met / Not met / Unclear and quote the words you relied on. Do not suggest legal conclusions. Clause: [paste]

Consistency check: Check the document below for inconsistent defined terms, cross-references that look wrong, and dates that do not match. List each with its location. Document: [paste]

Verify before use: Always verify AI output before relying on it. Check every fact, quotation and citation against the primary source, and follow your own bar or regulator's guidance on confidentiality and supervision.


## Maintaining the library

1. Test: Run each new prompt on three anonymised examples.
2. Approve: A second person reviews the prompt and its limits.
3. Date it: Record the model and date tested.
4. Review quarterly: Retire prompts that no longer perform.

LawyerOS does not ship a prompt library or AI model. You can build this database in the same workspace using the approach above; Pro with AI ($39) includes a Custom AI Agents page about Notion agents.

- Everyday AI prompts: https://notionforlawyers.com/post/ai-prompts-for-legal-work
- Clause library in Notion: https://notionforlawyers.com/post/clause-precedent-library-notion


## FAQ

### Why organise by stage?

Lawyers think in stages of work, so stage-based filters are quicker than tool-based ones.

### How many prompts should a library have?

Start with ten to fifteen that you actually use.

### Should prompts name a jurisdiction?

Yes, as a placeholder, and always verify the output for that jurisdiction.

### Can the library include real client facts?

No. Keep prompts generic and anonymise inputs at use time.