What ABA Formal Opinion 512 requires, the privilege risk after United States v. Heppner, and a checklist for drafting client email with AI.

Many lawyers now use AI in some form. Drafting client email is one of the most common uses, and one of the easiest to get wrong. Here is what the ethics rules say, where the privilege risk sits, and how to use AI drafts safely.
You can use ChatGPT, Claude, or another AI tool to draft client email. ABA Formal Opinion 512 allows it, with conditions.
You must review every draft. Relying on AI output without appropriate review can violate your duty of competence.
Putting client information into a self-learning AI tool requires the client's informed consent. Boilerplate in an engagement letter is not enough.
A 2026 federal ruling, United States v. Heppner, held that a client's chats with a public AI tool were not privileged. Read the tool's terms before any client information goes in.
The numbers depend on how you ask.
Clio's 2024 Legal Trends Report found AI use among legal professionals jumped from 19% in 2023 to 79% in 2024. (Clio 2024, via LawNext) Clio's 2025 solo and small firm report found 71% of solo firms use AI. (Clio 2025, via 2Civility)
The ABA's own survey measures firm-level adoption and comes in lower. 30% of respondents used AI tools in 2024, up from 11% in 2023. Among solos it was 18%. (ABA 2024 Legal Technology Survey, via LawNext)
Two findings matter more than adoption:
More than half of legal professionals (53%) say their firm has no AI policy, or they do not know of one.
78% of clients want lawyers to disclose AI use. Yet 35% of legal professionals rarely or never do. (Clio 2025, via 2Civility)
AI is strong at the routine writing that eats your day:
Acknowledging a new inquiry
Status updates on an active matter
Scheduling and document requests
Turning your notes into a clear summary
A polite reply to a question you have answered a hundred times
It is weak at anything that needs judgment. Do not let it give legal analysis, promise outcomes, or cite law without you checking every word.
Thomson Reuters' 2025 survey found legal professionals expect AI to free up about five hours a week, or roughly 240 hours a year. Note the word expect. These are anticipated savings, not measured ones. (Thomson Reuters, Future of Professionals 2025)
The ABA issued Formal Opinion 512 on generative AI in July 2024. For client email, five duties matter most. (ABA Formal Opinion 512)
Competence (Rule 1.1). Relying on AI output without "an appropriate degree of independent verification or review of its output" could violate your duty of competence. How much review you need depends on the tool and the task.
Confidentiality (Rule 1.6). For self-learning tools, "a client's informed consent is required prior to inputting information relating to the representation." General boilerplate in an engagement letter "is not sufficient."
Communication (Rule 1.4). Whether you must disclose AI use depends on the facts. You must disclose if the client asks. You must also consult the client when AI use is relevant to your fee.
Supervision (Rules 5.1 and 5.3). Firm leaders "must establish clear policies regarding the law firm's permissible use" of AI. That includes staff who draft email for you, and the AI vendor itself.
Fees (Rule 1.5). If you bill hourly, bill your actual time. That means time writing prompts and reviewing drafts, not the time AI saved. You may not charge a client to learn a tool you will use regularly.
State bars have issued their own guidance too. Our guide to AI and attorney-client privilege lists the main state opinions.
In February 2026, Judge Jed Rakoff of the Southern District of New York decided United States v. Heppner. A defendant had used a public AI tool, on his own, to prepare notes on his defense. The court held those AI documents were not protected by attorney-client privilege or work product.
The court found no attorney, no confidentiality under the tool's privacy policy, and no request for legal advice. It also said that sharing privileged advice with the tool could waive privilege over it.
Heppner involved a client acting alone, not a lawyer drafting email. But the lesson carries over. What the tool's terms say about retention, training, and disclosure matters. Our full guide to Heppner and ABA Opinion 512 covers what the court decided and what it left open.
In 2023, lawyers in Mata v. Avianca filed a brief citing court opinions that did not exist. ChatGPT had invented them, along with fake quotes. Judge P. Kevin Castel of the Southern District of New York imposed a $5,000 sanction on the two attorneys and their firm. (Mata v. Avianca, 678 F. Supp. 3d 443 (S.D.N.Y. 2023))
A court filing is higher stakes than a client email. But the failure was the same: sending AI output without checking it.
Before you adopt a tool, check these:
Read the terms. Does the provider retain your inputs? Train on them? Disclose them to third parties? ABA 512 says every lawyer should understand this, or consult someone who does.
Prefer business agreements. Enterprise and API terms often differ from consumer terms. Confirm what yours say in writing.
Require review before send. The tool should draft. You should decide. Nothing should reach a client automatically.
Label AI drafts. You and your staff should always know what the AI wrote.
Limit what it stores. The less client data sits in a vendor's database, the less can leak.
Write a one-page policy. Who can use which tools, for what, and who reviews. That covers your supervision duty.
Get real consent. Explain the tool to the client in plain terms when their information goes in.
LegalContext drafts client email and intake replies for your review. Every draft is labeled, and nothing sends to an existing client without your approval. AI runs under enterprise agreements with zero data retention, so client data is not used for training. Full email bodies are fetched when needed and not stored in our database.
No tool can guarantee privilege. That depends on the facts and on your role as counsel. But the design targets the confidentiality problem the Heppner court pointed to. See pricing or book a demo.