What You Can Put Into ChatGPT as a Lawyer, and What You Can't

What client information can a lawyer safely put into ChatGPT? A flat office rule sheet with green, amber and red lists, based on state bar ethics guidance.

Written by Miriad with AI assistance and checked against the sources at the end. An office rule sheet for everyday practice, not ethics advice: your state bar's opinions decide.

TL;DR: In a public tool like the free ChatGPT, put in your own knowledge, public material and generic text; keep out anything that identifies a client or their matter. Client information belongs only in a tool the firm has vetted, and in many cases only with the client's specific informed consent. Whatever comes out, a lawyer checks before it is used. The rule sheet below is built to be printed and pinned up.

The fear we hear most from small firms is simple: "Client data can't go into AI, so we can't use AI." The first half is mostly right for public tools. The second half does not follow. Most of what a small office writes in a week, marketing text, form letters, plain-language explanations, does not need client data at all.

What a small office needs is a flat rule that anyone at the desk can apply in five seconds. This article is part of our small law firm marketing guide, and it gives that rule as three lists.

What do the ethics opinions actually say about client data in AI?

The ethics opinions agree on the core: do not put client-specific or confidential information into a public AI tool, vet any tool before client data goes into it, get specific informed consent where the opinions require it, and verify every output. They do not ban AI. The details differ by state, so your own state bar's opinion decides.

The main texts, in plain terms:

  • ABA Formal Opinion 512 (July 2024): boilerplate engagement letter provisions are not sufficient informed consent to put client data into self-learning GenAI tools (National Law Review, 2024).
  • North Carolina 2024 FEO 1: lawyers should avoid inputting client-specific information into publicly available AI resources (North Carolina State Bar, 2024).
  • California 2026 guidance: no confidential client information in a tool with material confidentiality or security risks without informed client consent; relying on a vendor's "generalized marketing assurances" is not enough (State Bar of California, 2026).
  • Texas Opinion 705 (February 2025): lawyers are not required to use GenAI, but those who do need a "reasonable and current understanding" of it (Texas Opinion 705, 2025).
  • CCBE guide for European lawyers: no personal or confidential client data without appropriate safeguards, and verify outputs before use (CCBE, 2025).

What can go into a public AI tool?

Into a public AI tool you can put material that belongs to you or to the public and identifies no client: your own notes on a practice area, published statutes and court rules, your firm's existing marketing text, generic templates and questions you write yourself. If the text would be fine on your website, it is fine in the prompt.

Green list (public tools are fine):

  • Your own explanation of how a type of matter works, written from memory.
  • The text of a statute, court rule, agency notice or published decision.
  • Your existing website pages, newsletters and brochures, to rewrite or translate.
  • Generic letter structures with no client details ("a letter confirming an appointment").
  • Questions in the abstract: "What are the usual steps in an uncontested probate in general terms?"
  • Marketing tasks: practice area page drafts, social posts, review reply templates.

8am's list of marketing uses for law firms covers exactly this ground, blog drafts, social posts, About pages, translated web pages, newsletters and marketing emails, with a warning not to put sensitive client data into consumer-grade tools (8am, 2025).

What must stay out of a public AI tool?

Anything that identifies a client or their matter must stay out of a public AI tool: names, contact details, case numbers, facts, documents, opposing parties, and details that could identify someone even without a name. That covers information you received from the client and information you created about the matter. When in doubt, it stays out.

Red list (never in a public tool):

  • Client names, addresses, phone numbers, emails, dates of birth.
  • Case numbers, court file references, opposing party names.
  • Documents from the file: contracts, wills, medical records, financial statements, correspondence.
  • The facts of the matter, even with names removed, if someone could recognise them.
  • Your strategy, advice or notes on a specific matter.
  • Intake form submissions and conflict check results.

The "names removed" point catches people out. A unique set of facts in a small town identifies a family as surely as a surname. North Carolina's wording, avoid "client-specific information", is the safer test than "no names".

When does the answer change?

The answer changes when the tool is not a public one. Client information can go into a tool the firm has vetted, reading its terms of use and privacy policy, and in many cases only with the client's specific, informed consent. A general clause in the engagement letter is not enough for self-learning tools under Opinion 512.

Amber list (only with a vetted tool, and consent where required):

  • Summarising a client document inside a legal research or practice management software tool whose terms you have read.
  • Drafting from client facts in an enterprise AI tool the firm has approved for that purpose.
  • Transcribing a client meeting with an AI tool, after getting the client's permission.

California's 2026 guidance says reasonable efforts may include reviewing a vendor's terms of use and privacy policies, rather than relying on its marketing (State Bar of California, 2026). North Carolina adds that disclosure is generally not required for ordinary tasks such as research, while delegating substantive work is "akin to outsourcing" and needs informed consent (North Carolina State Bar, 2024). Which amber tasks need consent in your state is a question for your state bar's opinion.

Example (illustrative, not a real case): A solo family lawyer's assistant has a client's financial disclosure open and wants a quick summary for the lawyer. Her rule sheet says financial statements are on the red list for public tools. The firm's practice management software has an AI summary feature, approved by the lawyer after reading its data terms, and the client's engagement letter has a specific paragraph about it. She uses that feature instead, and the lawyer checks the summary against the document before relying on it. For the firm's newsletter that afternoon, she uses the free tool, because the input is the text of a new court rule.

Why must every output be checked, whatever went in?

Every output must be checked because AI tools produce confident text that can be wrong, including invented citations and misstated rules. The inputs decide confidentiality; the review decides accuracy. Both are the lawyer's duty, and neither can be delegated to the tool. Build the check into the workflow, not into good intentions.

The standard example is Mata v. Avianca. The court sanctioned lawyers who filed briefs containing non-existent case citations generated by ChatGPT and imposed a $5,000 penalty, while stating there is "nothing inherently improper" about using a reliable AI tool (Mata v. Avianca, 2023). The lawyers' role as gatekeepers under Rule 11 was the point.

A short office routine for outputs:

  1. Every citation and quotation is opened and read in the original.
  2. Every statement of law is checked against the current source.
  3. Anything going to a client or the public is read by a lawyer.
  4. If anything from the red list was pasted by mistake, tell the responsible lawyer the same day.

Print the three lists on one page and pin it next to the screen. When the firm turns it into a formal rule, the one-page law firm AI policy template gives the wording. For marketing work the green list covers most of what you need; email marketing for a law firm shows a full newsletter workflow on that basis, and the small law firm marketing guide applies the same test to every channel.

If you want your AI tools to start from the firm's own written facts, including this rule sheet, instead of whatever was pasted last, the Business Brain Starter Kit is built for that. Get notified when it launches.

FAQ

Can a lawyer put client information into ChatGPT?

Not into the free, public version, according to the ethics guidance we checked. North Carolina says to avoid client-specific information in publicly available AI tools, and California requires informed consent before confidential information goes into a tool with material confidentiality risks. Client information belongs only in a tool the firm has vetted, with specific consent where your state requires it.

Is anonymised client information safe to put into AI?

Often not. Removing names does not stop facts from identifying a client, especially in a small community or a narrow practice area. North Carolina's test is client-specific information, not just names. If someone who knows the client could recognise the situation from the prompt, treat it as client information and keep it out of public tools.

Does a clause in the engagement letter allow me to use AI with client data?

A general boilerplate clause is not enough under ABA Opinion 512 for self-learning tools. Consent should be informed and specific: which tool, for what task, and what happens to the information. Some ordinary uses may not need disclosure in some states, so check your state bar's opinion for the exact line.

What can lawyers safely use ChatGPT for?

Tasks that need no client information: drafting marketing text, rewriting your own website pages, summarising a published statute or court rule, preparing generic letter structures, and brainstorming questions. Even then, a lawyer should check every statement of law and every citation, because AI tools can produce confident text that is wrong.

Sources

Related articles

Products from Miriad

Business brainBusiness Brain Starter Kit

Stop explaining your business from scratch. Write it down once, with a source next to every line.

49 USD

Get notified

See all prompt packs