AI for Lawyers Playbook

How to Use This Template

Replace every [[Token]] with your actual practice details. The verification protocol and ethics sections are non-negotiable starting points. Adapt prompts to your jurisdiction and matter types. Always keep the lawyer in control of final work product.

What to Provide

Section 1: Practice-Area Use Cases

Drafting

• First drafts of routine contracts, clauses, and correspondence from a solid precedent or spec.

• Redlines and markups with explanations of changes.

Research and memos

• Case law and statute summaries with citations to verify.

• Compare treatment of an issue across jurisdictions.

Review and due diligence

• Extract obligations, deadlines, and anomalies from long agreements.

• Organize discovery or data room materials into issue summaries.

Client intake and communications

• Turn client facts into structured intake notes and initial advice outlines.

• Draft client updates that stay within scope.

Your priority use cases (start with 2):

1. [[Task that consumes the most hours]]

1. [[Task where consistency and speed add the most value]]

Section 2: Tool Categories and Examples

Specialized legal AI (when available and budget permits):

• Harvey, CoCounsel (Thomson Reuters), Lexis+ AI, Casetext (now part of CoCounsel), Spellbook for Word.

These are built with legal data, citation handling, and often better privilege/security posture.

General frontier models (with strict protocols):

• Claude (excellent for long documents), GPT-5 class, Gemini 3.5 Flash for very large contexts.

Use these for drafting, summarization, and brainstorming when specialized tools are not yet justified.

Supporting tools:

• Document comparison (Word, Litera, or PDF tools)

• Secure file sharing and matter management

• Citation verification back to primary sources (Westlaw, Lexis, official reporters, .gov)

Your starting stack: [[Primary legal AI or general model]] + [[document tool]] + [[research database]]

Section 3: Hallucination and Citation Verification Protocol

This is the most important section. Follow it on every piece of work that will be relied upon.

Mandatory verification steps (no exceptions for client-facing or filed work):

1. AI generates draft or research summary with citations.

1. Lawyer personally opens every cited case, statute, regulation, or secondary source.

1. Confirm the quote or proposition is accurate and current (check for negative treatment).

1. Note the "as of" date of the verification in the work product or research log.

1. If a citation cannot be verified in 2 minutes, remove it or mark it clearly as unverified and do not rely on it.

Red flags that require extra scrutiny:

• Very recent "cases" or statutes the AI claims exist.

• Overly specific holdings that sound right but have no pin cite.

• AI answers that contradict your own knowledge without strong sourcing.

• Any numerical or date detail presented without a primary source.

Simple rule: If the AI cannot point you to a real, checkable source in under a minute, treat the claim as unverified.

Section 4: Confidentiality, Privilege, and Data Handling

Core professional duties remain yours.

• Do not input confidential client information into consumer AI tools that lack appropriate data processing agreements and zero-retention options.

• Prefer enterprise or legal-specific versions that contractually protect confidentiality and do not train on your inputs.

• When using general tools for non-confidential brainstorming or public law research, strip identifying details.

Practical habits:

• Use matter names or numbers instead of party names when possible.

• Redact or placeholder sensitive facts before pasting.

• For highly sensitive matters, do the first pass yourself or with approved secure tools only.

• Log any use of AI on a matter (tool, date, what was asked, what was verified).

Privilege considerations: Using AI does not automatically waive privilege, but sloppy handling of outputs or inputs can create risk. Treat AI-assisted drafts the same way you treat any other draft.

Section 5: Professional Conduct and Competence

Model Rule 1.1 (competence) and equivalents: You must understand the tools you use well enough to supervise them. Blind reliance is not competent practice.

Best practice checklist before relying on AI output:

• I understand how the tool works at a basic level.

• I have verified all material factual and legal claims against primary sources.

• The final work product reflects my independent judgment.

• I can explain to a court, client, or partner how I used technology and why I reached the conclusion.

Never claim or imply that AI performed legal analysis or made decisions on the matter. The lawyer remains responsible.

Section 6: Billing and Workflow Integration

Billing transparency (recommended):

• Track time spent prompting, reviewing, and verifying separately at first.

• Many clients expect efficiency gains to be passed on in some form (lower blended rate, fixed fees for routine work, or capped matters).

• Document AI use in time entries or matter notes when it materially affected the work.

Workflow pattern that protects quality and margin:

1. Human defines the precise task and success criteria.

1. AI produces first draft or research extract (fast).

1. Human reviews, verifies citations, edits for strategy and judgment (the high-value part).

1. Final polish and sign-off by responsible lawyer.

1. Store the verified output and any useful prompt in the matter knowledge base for reuse.

When to charge premium vs efficiency rates: Premium for novel strategy and high-stakes judgment. Efficiency pricing for high-volume routine drafting and review once the process is proven.

Section 7: 60-Day Safe Adoption Plan

Days 1-14: Pick one low-risk task (internal memo, first draft of a standard clause, public statute research). Run 10 examples with full verification protocol. Measure time and error rate.

Days 15-30: Add a second task. Create 3-5 reusable prompt templates with verification checklists built in. Train any juniors or support on the protocol.

Days 31-60: Integrate into one live matter type with client communication if appropriate. Add matter logging. Review all AI-assisted work in team meeting for lessons.

Gate to expand: Zero material citation errors on verified work and time savings documented on at least 15 matters.

Worked Prompts (adapt and verify everything)

Clause redline explanation:

"Review this clause against the attached precedent. Identify every material change, explain in one sentence why it matters to our client, and flag any new risk created. Cite the specific language that creates the risk."

Research memo outline:

"From the attached cases and statutes, produce a 2-page outline on [[legal issue]] in [[jurisdiction]]. Group by sub-issue. For each point list the key holding + pin cite. Flag any split or recent change."

Pro Tips

• The value is in the verification and the strategic edit, not the first generation.

• Build a private library of your best precedents and prompts: this compounds faster than any model improvement.

• For court filings and opinions, the verification bar is highest. For internal brainstorming, you can move faster with lighter review.

• Talk to clients early about AI use. Most appreciate transparency and documented efficiency.

Common Pitfalls

• Treating the AI's first draft as almost done.

• Relying on citations the AI invented or misstated.

• Inputting client confidences into tools without proper contracts.

• Using AI on novel legal questions without extra human research.

• Failing to log use so you cannot explain your process later.

AI is a powerful research and drafting accelerator for lawyers who stay in control. The lawyers and firms that benefit most are the ones who treat it like a very fast, sometimes unreliable junior associate: give clear instructions, check the work against the record and the law, and take full responsibility for the final product. Verify everything that matters. The rules of professional responsibility have not changed.