Law Firm Engagement Letter: The AI-Use Clause Clients Read

Add a short, specific clause that names the kinds of task where you use AI, says client information stays inside tools you have vetted, promises a lawyer reviews everything, and invites the client to object. Write it in plain words and talk it through, because ABA Formal Opinion 512 treats boilerplate consent as not enough.

The reason is simple. A clause nobody understood cannot make a client's consent informed. Opinion 512 is based on the ABA's Model Rules rather than on any one regulator's rules, and the conduct rules that bind your firm are your own regulator's, so treat what follows as drafting practice to check against those rules and your insurer's wording, not legal advice.

Follow me on Instagram@sagnikteaches

What the opinion asks for, in plain terms

Formal Opinion 512, issued on 29 July 2024, makes a handful of points a letter has to respect (read the full text on the ABA's website before relying on any paraphrase, this one included, and see the ABA Formal Opinion 512 summary for small firms for the parts beyond engagement letters). Lawyers must understand a tool well enough to judge its output. They must protect client confidences. They must keep clients reasonably informed about how the work is done. And before information relating to a matter goes into a self-learning tool, meaning one that may learn from what you put in and surface it to other users, the client's informed consent is required.

Connect on LinkedInSagnik Bhattacharya

The sentence that matters for drafting says that merely adding general, boilerplate provisions to engagement letters purporting to authorise AI use is not sufficient to obtain that consent. The opinion then lists what the client needs from you: your best judgement about why the tool is being used, specific information about the risk (including the kinds of client information involved and how others might use it against the client's interests), and a clear explanation of the tool's benefits to the matter. A line such as "the firm may use technology, including artificial intelligence, in providing its services" covers none of that. A client who signs it has learned nothing about what happens to their documents.

Subscribe on YouTube@sagnikteaches

Two further points shape the wording. The opinion accepts that, depending on the circumstances, telling a client about AI use may be unnecessary; but a lawyer must disclose AI practices when the client asks, or when the engagement terms or the client's outside counsel guidelines require it. And before charging a client for the use of an AI tool, the lawyer must explain the basis for the charge, preferably in writing.

That leaves two jobs. The letter should be honest and specific enough to inform. And, where client information really will go into an AI tool, the letter should be the start of a conversation rather than the end of it.

Decide what is true before you write a word

The most common failure is a clause that describes an imaginary firm. Before drafting, fill in the inventory below. It takes about an hour and it is the only part of the job that cannot be borrowed from a template.

If your firm runs on a legal practice platform, start with what it provides. Clio, Smokeball and similar platforms add AI features to the system that already holds the file; Smokeball, for example, describes its Archie assistant as working inside the matter to draft, review and answer questions. Their terms are written for law firms, unlike a consumer chatbot's, but do not assume they suit you. Open the help article for each feature you switch on and write down what it says about retention.

Three clause variants, from lightest to fullest

Pick the variant that matches the inventory, not the one that sounds most reassuring. Square brackets are for you to fill in.

Variant A: no client information enters an AI tool

Use this only if it is genuinely true, for example where AI is limited to administrative drafting or internal checklists.

How we use AI tools
We use AI tools for some internal tasks, such as drafting checklists and
tidying our own notes. We do not enter your documents, names or case
details into them. Everything we send you is checked by a lawyer here
before it leaves the firm. If you would like to know more about what we
use, ask [name] and we will explain.

Variant B: AI used on client material inside vetted tools

This is the realistic position for most small firms. It names the task types, the safeguards and the right to say no.

How we use AI tools on your matter
To work efficiently, we sometimes use AI tools to [summarise long
documents / draft first versions of letters and notes / search through
documents you have given us]. When we do this with your information we
only use tools we have checked, under business terms that [do not allow
the provider to use your information to train its models] and [keep it
for no longer than [period]].

A lawyer reads and takes responsibility for anything AI helps to
produce before it is used or sent. We do not use AI to decide legal
strategy or to give you advice.

If you would rather we did not use AI on your matter, tell us at any
point and we will note it on your file. If that would change how long
the work takes or what it costs, we will tell you before we go ahead.
You are not charged for time that a tool saves; our fees reflect the
work actually done.

Variant C: a fuller version for sensitive matters

For family, employment, immigration, regulated-sector or high-profile matters, add the specifics a worried client will ask about anyway.

AI tools and your confidential information
On this matter we expect to use [tool name], a [type of tool], for
[specific task]. We use it because [benefit to your matter, e.g. it
lets us review the 4,000 emails disclosed to us in days rather than
weeks]. The information involved is [kinds of information, e.g. your
emails and contracts with the other side]. In plain terms: your
documents are sent to the provider's servers, processed, and the
result returned to us. The provider's terms say [retention period /
not used for training]. The
remaining risks are that a provider could suffer a security breach
or change its terms, and that AI can produce wrong or invented
content, which is why a lawyer checks every output.

We will not use any AI tool on your matter that is not described here
without telling you first. You may withdraw agreement for AI use on
your matter at any time by telling [name]. Please tick one:
[ ] I agree to AI tools being used as described.
[ ] I would prefer that AI tools are not used on my matter.

Notice what is missing from all three: "may" used as a hedge, the phrase "state of the art", and any claim of accuracy. Each variant says what the firm does, what it checks, and how the client can say no.

Variant C is built to carry the elements the opinion lists for informed consent. Check a filled-in draft against this map before it goes out; any empty right-hand cell is a gap a client could fairly point to.

A five-lawyer firm drafting its letter

Take a five-lawyer general practice firm (illustrative) with a practice management platform, one paid chatbot subscription on business terms, and a transcription add-on for client meetings.

The inventory shows three uses. The platform's AI summarises documents already stored on a matter. The chatbot is used for first drafts of letters, always with names removed. The transcription tool records meetings with the client's permission. Those are three different risk levels, and one blanket sentence would misdescribe at least two of them.

The partner writes Variant B and trims it to fit. For the platform: "We use the AI features in our practice management system to summarise documents you have given us. Its provider is bound by business terms; retention is [as stated in its data terms]." For the chatbot: "We sometimes use a general AI assistant to draft routine wording. We remove names and identifying details first." For meetings: "We only record or transcribe meetings with you if you agree at the start of each meeting."

The result runs to about 140 words, roughly a minute's reading. That matters: clients skim letters, and a 600-word technology section gets skipped as legal noise. The partner then adds a line to the new-matter checklist: "AI paragraph discussed, client response recorded: yes / no / objects". That line turns the letter into evidence of an actual conversation.

The client conversation: a 90-second script

Because a signed letter alone is weak evidence of understanding, give the clause a spoken companion. Say this at the first meeting, before the client signs, and note on the file that you did.

"One thing in the engagement letter I want to point out. We use some
AI tools to speed up routine work, like summarising long documents or
drafting a first version of a letter. When we use them on your
documents they stay inside tools we have checked, and a lawyer here
always reads and signs off the result. The risks are small but real:
a provider could be breached, and AI can get things wrong, which is
why we check everything. You can say no to AI on your matter at any
time and it will not affect how we treat you. Do you have any
questions, or would you like it left out?"

Three details make that script work. It names a concrete task, so the client can picture it. It names a risk in one sentence, which is the bit boilerplate leaves out. And it ends with an invitation to say no, which makes consent a choice rather than a formality.

Four clauses that look fine and fail

The catch-all

"We may use technology, including AI, in the delivery of our services." It is true of every firm and informs no client. Replace it with a sentence naming at least one task and one safeguard.

The overpromise

"Your information will never leave our secure systems." If you use a cloud AI tool, that is false the moment a document is processed on the vendor's servers. A before and after:

Before: Your information will never leave our secure systems.
After:  Your information is processed by tools we have vetted, under
        business terms, and is not used to train the provider's models.

The opt-out buried in a schedule

A right to object that appears on page nine of the terms of business is not much of a right. Put it in the AI paragraph itself and repeat it in the conversation.

The clause that outlives the tool

A firm switches assistants in March and the letter still names the old one. Name tool categories in the body ("an AI assistant used under business terms") and keep a dated one-page appendix of named tools you can update without reissuing the letter. Versioned appendices also give you a record of what a given client was told.

Other places the clause has to match

A letter is only one surface. If the others disagree with it, the letter becomes a liability.

Website and marketing copy

If a page says "we use the latest AI to deliver faster results" while the letter says AI is used sparingly, a client can fairly say they were told two things. Read AI marketing for small law firms and keep public claims no bigger than the clause.

Fee wording

Say plainly that hourly fees reflect time actually spent. Opinion 512 gives the working rule: a lawyer billing hourly may charge for the time spent putting information into the tool and reviewing what comes back, but not for time the tool saved, and not for learning a tool the firm uses for every client. In figures (illustrative): a first-draft letter that used to take 90 minutes now takes 15 minutes of instruction and 30 minutes of checking, so the bill shows 45 minutes. If you would rather charge a fixed $250 for that task, give it its own sentence in the fees section, describe what it covers, and agree it before the work starts. If you intend to pass on any cost of the tool itself, explain the basis for that charge in writing first. A reader should never wonder whether faster work means a bill that stays the same.

Outside counsel guidelines and corporate clients

Some business clients send their own panel rules that ban or restrict AI. Those override your standard clause. Keep a register of such clients and tag their matters so tool access follows the tag.

Meeting recordings

Transcription raises its own consent and confidentiality points, covered in AI note takers in client meetings. Keep that consent separate from the general clause so a client can accept one and refuse the other.

Answers to the awkward questions clients ask

Expect five. Prepare short answers with the whole team so every fee earner says the same thing.

Do not improvise on retention or training. If you do not know, say "I'll confirm in writing", then check the vendor terms. A wrong answer given in a meeting is worse than a short delay.

Matching each task to the wording it needs

A clause gets sharper once you tie each AI use to the sentence that describes it. This table is filled in for a small firm doing mixed work (illustrative). Use it as a model for your own, then delete rows you do not recognise.

The last row is the one firms forget. Pasting a draft letter into a grammar tool is the same act as pasting it into a chatbot, and a clause that claims "no client information enters any AI tool" is false the first time someone does it.

Existing clients: sending the paragraph to people who already signed

New matters are easy; open matters and long-standing clients are where firms stall. Handle them in three tiers so you do not mail five hundred people at once.

1. Active matters using AI on client material. Contact these clients first, by phone or email, using the spoken script above. Record the response on the file.
2. Active matters not using AI on client material. Send the Variant A paragraph with the next routine letter. No reply needed.
3. Dormant clients. Add the paragraph to the engagement letter template and wait. Nothing needs sending until the client instructs you again.

Say, for a firm with 60 open matters (illustrative), perhaps 15 involve the document summariser. That is fifteen short calls, roughly five minutes each including a note on the file, so about an hour and a quarter of fee-earner time. Compare that with explaining a retention question to a complaints handler after the fact.

What the team needs to hear before the letter changes

A clause is only as honest as the behaviour behind it. Hold a twenty-minute briefing and cover four points, using real examples from your own matters if you can.

• Which tools are allowed, and which are not. Personal chatbot accounts on a client matter are the usual leak.
• What stripping names means. Replacing the client's surname with "the client" while leaving their address and the unusual facts of the case is not removing identifying details. Show one before and after.
• Where the objection flag lives. Everyone should know how to see that a client declined AI.
• Who to ask. Name one person who answers "can I put this into the tool?" the same day, so people are not tempted to guess.

The tutorial on what staff need to know about AI gives a longer syllabus if you want a half-day session later.

Where the insurer comes in

Some professional indemnity proposal forms now ask whether and how a firm uses AI, so check whether yours does. Your answer on that form and your letter should describe the same firm. Before you finalise the clause, read the AI questions on your latest renewal form, and if something in the clause goes beyond what you told the insurer, tell your broker. If you cannot find a clear answer on what the policy expects, ask in writing and keep the reply with the clause's version history.

Keeping the clause true after you send it

Add a quarterly routine. It costs perhaps thirty minutes.

1. List every AI tool in use again, including anything new that staff mention.
2. Re-read each vendor's current data terms for retention, training and processing location.
3. Compare the result with the clause and its appendix, and edit anything no longer accurate.
4. Check the matters flagged "client objects to AI" and confirm none of their documents sits in any tool.
5. Record the date and who did the review.

The tutorial on legal tasks a small firm should never hand to AI gives the flip side of step one: the uses the clause should rule out rather than describe.

Checklist before the letter goes out

• Every AI tool used on client matters appears in the clause or its appendix.
• At least one task type and one safeguard are named in plain words.
• The right to object is in the AI paragraph and in the conversation.
• The fee section says time saved is not billed as time worked.
• The new-matter checklist records that the conversation happened.
• A named person owns the quarterly review.
• Your insurer's and regulator's current guidance has been read, and a qualified adviser has looked at the final wording.

If you would rather work through the tool inventory with someone, that is part of what my AI implementation consultation covers. Either way, do the inventory first. If you are still deciding between a general chatbot and a legal-specific product, ChatGPT or a legal AI tool sets out how their data terms usually differ. The best clause is the one that is accurate on the day a client reads it and still accurate next year.

Questions firms ask before they change the letter

Do we need a clause if we only use AI for admin?

Opinion 512 accepts that disclosure may be unnecessary in some circumstances, but you must answer honestly when a client asks, so a short statement is still sensible. If no client information goes into a tool that could retain or learn from it, one paragraph on how you use AI is enough. Check your own conduct rules and your insurer's wording too.

Can the clause let us bill for the time AI saves?

No. Opinion 512 says lawyers who bill hourly must bill their actual time: minutes spent instructing the tool and reviewing its output count, the time it saved does not, and learning a tool you use for every client is not billable. If you move to a fixed fee for AI-assisted tasks, say so in the fee section first and explain what it covers.

What if a client refuses AI use entirely?

Record the refusal on the file and flag the matter in your practice system so nobody pastes its documents into an AI tool by accident. Tell the client plainly if that choice changes timing or cost, so the decision stays informed.

How often should the clause be reviewed?

Review it whenever you add, swap or change the settings of an AI tool, and at least twice a year. Tools change their data terms often, so a clause written for last year's settings can quietly become inaccurate.

Further reads

• Can Solicitors Use ChatGPT Without Breaching Confidentiality? — The confidentiality question that sits behind the clause.
• AI Readiness Checklist for Accountants, Solicitors, Consultants — Check the firm is ready before promising anything to clients.
• AI Implementation Plan for a Small Law Firm: The First 90 Days — A 90-day plan for rolling tools out behind the clause.
• How Small Law Firms Use AI to Answer Client Status Questions — Another client-facing use the letter may need to cover.
• AI for Small Law Firms: What to Automate First — Pick the first tasks the clause will describe.
• Conflict Check Software for Small Law Firms: Where AI Helps — Where AI helps in a conflict check, where it must not decide, and what a partner signs off, with a worked example and a 20-minute test of your own tool.
• Consulting Engagement Letter: Scope, Fees and AI-Use Wording — A clause-by-clause engagement letter for independent consultants: scope, fees, change control, confidentiality and plain AI-use wording you can adapt.
• AI for Personal Injury Lawyers: Seven Jobs to Hand Over First — Seven injury-matter workflows with sample outputs, review gates and a clear boundary around legal judgement and confidential records.
• Can AI Draft a Client Care Letter That Meets SRA Rules? — A matter-by-matter SRA review sheet for turning an AI draft into a clear, accurate client care letter.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ABA Formal Opinion 512, Generative Artificial Intelligence Tools (29 July 2024), full text; Clio help centre article AI Features in Clio Manage; Smokeball's Archie product description; the firm's own AI tool terms.