Using AI for Client Communication Drafting: A Practical Guide for Attorneys

The average attorney writes more emails in a week than most professionals write in a month. Status updates, follow-up requests, declination letters, engagement confirmations, settlement summaries, court-date reminders — most of it is routine correspondence that follows predictable patterns and consumes time out of proportion to its strategic value.

This is exactly the category where AI assistance delivers the most immediate, least controversial return. Not research (where hallucinations are dangerous), not contracts (where clause judgment matters), but the drafting of clear, professional correspondence based on information the attorney already has.

This guide is about making that workflow practical: the setup, the prompts, the checkpoints, and the ethics obligations that apply even to routine emails.

What AI drafting actually saves

The savings are in two places. First, drafting time: a general AI assistant can produce a first draft of a routine email in seconds, based on a brief prompt. The attorney's role shifts from writing from scratch to reviewing and adjusting — a faster task, especially for correspondence following a predictable pattern.

Second, the blank-page problem: some attorneys find starting correspondence easier when there is already something on the page. The draft does not need to be good to be useful; it just needs to be there to react to.

Neither of these is a large time saving per message. But routine correspondence is not one message — it is a category that generates dozens of messages per week across every matter. Compounded across a year, the time saving is material.

The workflow

The three-part prompt structure

Routine correspondence AI drafts work best with a prompt that provides three things:

1. The role and tone: "You are drafting a professional, plain-English email from an attorney to a client."
2. The specific situation: who, what, where the matter stands, what the email needs to accomplish.
3. The output requirements: length, formality, whether to leave blanks for specific details like dollar amounts or dates you will fill in.

An example prompt for a status update:

"Draft a brief status update email to a client in a personal injury matter. The client is [name or just 'the client']. Developments this week: opposing counsel provided interrogatory responses and we are reviewing them. Next step is a follow-up letter on two incomplete responses. Deposition of opposing party is scheduled for [date to be confirmed]. Tone: professional but direct, no jargon. Length: 3-4 short paragraphs. Leave blanks in brackets for the deposition date."

That produces a draft the attorney reviews in 30 seconds, adjusts where needed, fills in the blanks, and sends.

The anonymization practice

Client names and identifying matter details should not go into a general AI assistant on a consumer tier. Under Model Rule 1.6 and ABA Formal Opinion 512, client information is confidential and should only go into tools with appropriate data handling protections — a no-training-on-inputs tier and a data processing agreement.

The practical solution for routine drafting is anonymization: write the prompt with "the client" instead of a name, "the opposing party" instead of a specific entity, "the contract in dispute" instead of the specific agreement. The AI produces a perfectly usable draft without receiving identifying information. You then fill in the actual names and specifics during review.

For firms on business tiers of their AI platform (ChatGPT Team, Claude Pro, Gemini Workspace), the terms generally cover client-specific information — but verify the current terms before relying on it.

Templates versus one-off drafts

For correspondence you write repeatedly — engagement letters, declination letters, case status updates, settlement offer summaries, court-date reminders — the highest-leverage use of AI is template creation, not one-off drafting.

Draft a template for each correspondence type once, using AI to produce the first version and you to refine it. Then the workflow for each instance is: fill in the case-specific details, review for accuracy, send. This is faster than prompting from scratch every time, and the quality is more consistent.

For correspondence that is genuinely one-off — a complex settlement summary, a nuanced declination letter, a client update on an adverse ruling — one-off prompting with a detailed context prompt is the right approach.

The review step

Every AI-generated draft requires attorney review before it is sent. This is not optional courtesy — it is the professional obligation. Under Model Rules 1.1, 5.1, and 5.3, the attorney is responsible for every communication that goes out under the firm's name, regardless of what tool produced the first draft.

The review has three practical checks:

1. Accuracy: does the draft accurately reflect the matter status, the next steps, and any commitments or timeline representations? AI does not know your file; it knows only what you told it in the prompt.
2. Tone: does the draft match the relationship with this client? Some clients want formal and brief; others want warm and detailed. Review for tone, not just facts.
3. Commitments: does the draft inadvertently make a promise you do not intend? Phrases like "we will have a response by [date]" or "this should be resolved within [timeframe]" create expectations. Catch these before sending.

Common email types and prompt angles

Status updates — cover what happened since the last update, what comes next, what (if anything) is needed from the client, and when you will be in touch again.

Deadline and scheduling alerts — announce the date, what it requires of the client (if anything), and a brief note on what the firm is doing to prepare.

Settlement summaries — explain the terms, the recommendation (or that this is the client's decision), and the next step. These require careful review for accuracy: dollar amounts, conditions, and timing must be exactly right.

Declination letters — acknowledge the inquiry, decline representation, advise the client to seek other counsel promptly (deadline awareness), do not offer a legal opinion on the merits of their matter. This category is well-suited to a refined template.

Adverse-development updates — these are the hardest to draft because the tone must balance honesty about a setback with appropriate context and next steps. AI produces a first draft; the attorney's judgment shapes what is actually sent.

Document requests — describe exactly what you need, in what format, by when, and why. Keep it to a list rather than paragraphs where possible.

The billing question

ABA Formal Opinion 512 and Model Rule 1.5 require that fees be reasonable. AI-assisted drafting compresses time — if a routine email took thirty minutes to draft manually and takes five with AI assistance, you do not bill thirty minutes. What you bill is the time you actually spent: reviewing the draft, making revisions, applying your judgment, and ensuring accuracy.

This is not a financial loss — it is a business model adjustment. The time savings from AI correspondence drafting creates capacity for more matters or for deeper work on complex ones. The value is in the throughput, not in billing inflated time.

Setting up your correspondence workflow

The most effective approach is to spend a few hours upfront creating the templates and prompt patterns you will actually use, rather than improvising a new prompt each time. A simple system:

1. List the correspondence types you write at least monthly.
2. For each, write a model prompt that produces a useful draft — including the role, situation structure, tone guidance, and output requirements.
3. Keep these prompts accessible during your email workflow — a doc, a note, wherever works.
4. When you need one, retrieve the prompt, fill in the case specifics, and generate the draft.

The upfront investment is a few hours. The return is a faster correspondence workflow for as long as you practice.

For the tool selection — which general assistant to use, how to handle the confidentiality tier, and where AI correspondence fits in the broader toolkit — see the best AI tools for lawyers.

Related reading: The best AI tools for lawyers | AI for law firms | Using AI to write attorney bio and website content