A practical guide to law firm AI use, including safe workflows, confidentiality, verification, supervision, policy, and marketing applications.
Useful starting points
- Brainstorming outlines and questions
- Summarizing nonconfidential background material
- Reformatting approved content
- Drafting internal checklists
- Repurposing reviewed marketing copy
- Creating first-pass email variations
Start with low-risk, reversible tasks and require human review before anything reaches a client, court, prospect, or public channel.
The five controls every firm needs
- Approved toolsDefine which products and account types may be used.
- Data rulesState what information may never be entered and when consent or contractual protection is required.
- VerificationRequire source checking, citation review, and substantive attorney approval.
- SupervisionTrain lawyers and staff and assign responsibility for the output.
- DocumentationKeep the policy, workflow, and review standard current.
Ethics don't disappear because a tool is new
ABA Formal Opinion 512 applies existing duties such as competence, confidentiality, communication, candor, supervision, and reasonable fees to generative AI. State rules and opinions may add or clarify requirements.
Marketing uses still need guardrails
AI can help create newsletter concepts, subject-line variants, interview questions, content briefs, and repurposed drafts. It shouldn't invent results, testimonials, awards, statistics, citations, or jurisdiction-specific claims. Keep an attorney in the approval loop.
Choose tools based on data handling
Review training practices, retention, access, sharing, integrations, deletion, and contract terms. Product tier matters: business and consumer offerings may handle data differently.
This article is general marketing and technology information, not legal advice. Rules, products, and platform practices change; confirm current requirements for your firm and jurisdiction.
Where ChatGPT fits
How lawyers can use ChatGPT for low-risk workflows while protecting confidentiality, checking accuracy, and maintaining professional judgment.
What ChatGPT can help with
- Generate questions for an attorney interview
- Turn approved notes into a first-pass outline
- Rewrite legal marketing copy in plain English
- Create multiple subject-line approaches
- Summarize public, nonconfidential material
- Build checklists for human review
What it shouldn't replace
Don't treat model output as verified law, a source, a conflict check, professional judgment, or final work product. It can produce plausible but false information, including invented authorities and quotes.
Decide what data is permitted before using any AI tool. Confidentiality analysis depends on the information, product, settings, terms, and applicable rules.
A safe five-step workflow
- Classify the taskIs it low risk, internal, and reversible?
- Remove sensitive dataUse approved anonymization and don't assume removing a name is enough.
- Prompt for structureAsk for questions, options, or a draft, not a declaration of truth.
- Verify independentlyCheck every legal, factual, statistical, and quoted claim against reliable sources.
- Approve as a humanThe responsible lawyer or reviewer owns the final result.
Understand the account you're using
OpenAI states that business data from ChatGPT Business, Enterprise, and its API isn't used for model training by default. Consumer settings and other products may differ, so review current terms and controls rather than relying on a general assumption.
Source: OpenAI business data privacy.
This article is general marketing and technology information, not legal advice. Rules, products, and platform practices change; confirm current requirements for your firm and jurisdiction.
Claude vs. ChatGPT is the wrong first question
Compare products, not brand names
Consumer, team, enterprise, and API versions may have different privacy controls, retention terms, administrative features, and integrations. Evaluate the exact plan your firm would buy.
Seven comparison criteria
- Data use and retention terms
- Administrative and access controls
- Quality on the firm’s real tasks
- Source handling and verification support
- Integrations and document workflow
- Cost predictability
- Training, supervision, and auditability
Run a controlled bake-off
Create ten low-risk representative tasks with a scoring rubric. Remove client facts, use the same instructions, and compare accuracy, omissions, tone, revision effort, and reviewer confidence.
Don't skip governance
No model comparison replaces duties involving competence, confidentiality, supervision, communication, and verification. Establish approved uses and prohibited inputs before adoption.