What to tell clients about the software that reads your firm’s email comes down to three things. What the assistant reads. Where that text goes. Whether a lawyer sees every reply before it leaves. Some clients are owed a consent conversation before their mail goes in. Most are owed a short, specific explanation. A few have already set terms in their outside counsel guidelines. This page helps you tell which client is which, and gives you wording for each.
- Consent depends on the tool. ABA Formal Opinion 512 (July 29, 2024) requires informed consent before client information goes into a self-learning tool whose output could disclose it. Read the vendor’s terms to know which side yours falls on.
- An explanation is often owed anyway. Without a consent duty, Rule 1.4 may still ask you to explain. You must answer if a client asks.
- Be specific. A general clause in the engagement letter isn’t enough for consent. Name the tool, what it reads and who else processes it.
- Only promise what your settings do. If you tell a client a lawyer reviews every reply, keep the assistant set so that’s true.
What to tell clients, depending on the tool and the client
Start with one fact you already know. An email assistant reads the whole mailbox it’s connected to. That means every client’s mail, opposing counsel’s mail, and anything else that lands. It’s a different shape from a drafting tool you paste one document into.
That shape sets up three conversations. Most firms will have all three, with different clients.
| The situation | What the client is owed | Where the rule comes from |
|---|---|---|
| The tool can learn from client mail in a way that could surface it elsewhere | Informed consent, before their mail goes in | Opinion 512 on Rule 1.6 |
| The tool doesn’t, but the client would want to know | A plain explanation, usually at intake | Opinion 512 on Rule 1.4 |
| The client asked, or their guidelines require it | A direct answer, and compliance with their terms | Opinion 512 on Rule 1.4(a)(4) |
Underneath all three is the duty to keep client information protected while the assistant handles it. That part is set out in what an email assistant reads and keeps. This page is about the conversation.
Whether this client needs a consent conversation
Opinion 512 draws the consent line around one kind of tool:
Accordingly, because many of today’s self-learning GAI tools are designed so that their output could lead directly or indirectly to the disclosure of information relating to the representation of a client, a client’s informed consent is required prior to inputting information relating to the representation into such a GAI tool.
The worry is mail from one client turning up in work for someone else. So the question to ask the vendor is narrow. Does anything learned from your mail feed a model that other customers use? The answer is in the vendor’s terms of use, usually in a section on training. If it does, you need consent from every client whose mail runs through that mailbox.
A second question matters inside your own firm. Most email assistants learn your habits from your own mail, like which senders matter and how you write. That learning stays in your account. It also means a draft for one client is written by a tool that has read your mail with others. Inside one mailbox, there’s no wall between clients. Weigh that for matters where it would matter, such as a conflict screened by an ethical wall.
California’s bar has spoken to email directly. Its 2026 Practical Guidance, put to the State Bar’s Board of Trustees for approval on May 14, 2026, describes AI systems connected to firm email:
Unlike isolated prompt-based tools, such systems may have persistent or automated access to large volumes of confidential client information, raising acute confidentiality considerations under rule 1.6 and Business and Professions Code section 6068(e)(1).
The same guidance says a lawyer “must not input any confidential information of the client into a generative AI solution that may present material risks to confidentiality or security, absent informed client consent.” If you practice in California, that’s your standard. Elsewhere, it’s a careful bar’s reading of the same problem, and your own state’s rules govern.
Opinion 512 adds that its consent conclusion reflects the tools “as of the publication of this opinion” and may change. Check your own state bar for anything newer before you rely on either.
When a plain explanation is owed instead
Where consent isn’t required, Opinion 512 says you still have a decision to make:
Where Model Rule 1.6 does not require disclosure and informed consent, the lawyer must separately consider whether other Model Rules, particularly Model Rule 1.4, require disclosing the use of a GAI tool in the representation.
Rule 1.4(a)(2) asks you to “reasonably consult with the client about the means by which the client’s objectives are to be accomplished.” The opinion gives the factors that decide it:
Potentially relevant considerations include the GAI tool’s importance to a particular task, the significance of that task to the overall representation, how the GAI tool will process the client’s information, and the extent to which knowledge of the lawyer’s use of the GAI tool would affect the client’s evaluation of or confidence in the lawyer’s work.
Applied to an email assistant, those factors come out like this.
- Importance to the task. Sorting and summaries touch everything but decide little. Drafting matters more, because the words go to the client.
- Significance to the representation. Scheduling replies are minor. A reply that gives advice or answers opposing counsel is central.
- How it processes the client’s information. All of it. Every message the client sends you goes through the tool.
- Effect on the client’s confidence. This is the one you know best. A client in a divorce, a criminal matter or a trade-secret dispute may feel very differently from a landlord with a lease question.
Two triggers settle it outright. Opinion 512 says you must disclose “if asked by a client how they conducted their work,” and when the engagement agreement or outside counsel guidelines require it. It also says you must consult the client when the tool “is relevant to the basis or reasonableness of a lawyer’s fee.” How the hourly rules apply to time an assistant saves is covered in billing for time the assistant saved.
California’s guidance points the same way. A lawyer “must consider disclosure to their client that they intend to use AI in the representation, including how the technology will be used, and the benefits and risks of such use.”
In practice, most firms find it calmer to tell every client. The explanation is short. A client who hears it at intake has nothing to discover later.
What a good explanation covers
When consent is owed, Opinion 512 sets out what makes it informed:
For the consent to be informed, the client must have the lawyer’s best judgment about why the GAI tool is being used, the extent of and specific information about the risk, including particulars about the kinds of client information that will be disclosed, the ways in which others might use the information against the client’s interests, and a clear explanation of the GAI tool’s benefits to the representation.
That’s a good outline for a plain explanation too. Here’s what each part means for an email assistant, and where you find the answer.
| What the client should hear | For an email assistant | Where to find it |
|---|---|---|
| Why you use it | Faster replies, nothing missed, more time on the work itself | Your own judgment |
| What information it handles | Every message in the connected mailbox, including theirs | The vendor’s description of what it reads |
| Who else processes it | The AI providers that write summaries and drafts | The vendor’s subprocessor list |
| Who could see it at the vendor | Staff access, and when it’s allowed | The vendor’s privacy policy |
| Whether it trains anything shared | Yes or no, in the vendor’s words | The vendor’s terms, training section |
| How long it’s kept | While the account is open, and after you leave | The privacy policy and data processing addendum |
| Who sends the replies | A lawyer, after review, if that’s how you’ve set it | Your own settings |
| The benefit to their matter | Quicker answers, deadlines caught, a briefing each morning | Your own judgment |
Opinion 512 is clear that a general clause won’t do this work:
To obtain informed consent when using a GAI tool, merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient.
So when consent is owed, have the conversation, then confirm it in writing. When it isn’t, the engagement letter is a fine place for the explanation. The opinion calls it “a logical place to make such disclosures and to identify any client instructions on the use of GAI in the representation.”
Wording you can adapt
These samples assume a tool like the one described at the end of this page. Replace every bracket with what your vendor’s documents actually say. Leave out any line that isn’t true of your setup.
For the engagement letter
How we use an email assistant. Our firm uses [Point], an AI email assistant, on our firm email accounts. It reads the email we send and receive, including yours, so it can sort messages, summarize long threads and prepare draft replies. A lawyer reads every draft before it’s sent to you or to anyone on your matter. To do this work, [Point] sends excerpts of messages to [Anthropic, OpenAI and Google], which provide the AI. [Point]‘s terms say our mail won’t be used to train general AI models. Staff at [Point] can see our mail only in limited cases, such as security or legal compliance. Our mail is kept while our account is open and [deleted within 60 days if we leave]. Like any service that holds email, the provider could be breached or required by law to produce data. If you’d prefer we didn’t use the assistant on your matter, tell us. We’ll explain what that changes before we go ahead.
When a client asks by email
Thanks for asking. Yes, we use an assistant called [Point] on our firm email. It reads incoming mail so it can sort it and draft replies, and a lawyer reviews every reply before it goes out. Its terms don’t allow our mail to be used to train general AI models. I’m happy to send you the details, or to keep your matter out of it if you’d rather.
In the intake meeting
We use an AI assistant on our email. It reads what comes in, including your messages, and drafts replies that I check before they go. The company that makes it doesn’t train its AI on our mail. If you’d rather your matter stayed out of it, that’s fine. Just say so.
A line about the client’s own AI use
Clients use public chat tools too. In February 2026, a federal court held that a client’s chats with a public AI tool weren’t privileged. The case is covered in what happens to privilege when software reads the mail. One sentence protects them:
Please don’t paste our advice or your messages to us into public AI tools. Doing so may affect whether those communications stay privileged.
Promises to leave out
A client hears your explanation as a promise. A few phrases sound reassuring and aren’t true of most email assistants. This is the one place on this page where the difference costs something.
- “Your mail is kept separate from other clients.” Inside one mailbox, the assistant reads everything together. Some tools keep separate businesses apart, but that line runs between organizations you operate. It’s not an ethical wall.
- “Nothing is ever sent without a lawyer.” That’s only true while every kind of message is set to wait for review. Say it only if you’ve set it that way, and keep it that way.
- “Anything can be undone.” A sent email sits on the recipient’s server. Nobody can pull it back.
- “Nobody else ever sees it.” The AI providers process excerpts, and vendor staff have access in defined cases. Say who, and when.
If a client says no
Write the instruction down. Opinion 512 names the engagement agreement as the place for “any client instructions on the use of GAI.” Then work out what honoring it takes.
An email assistant reads whatever lands in the mailbox it’s connected to. So keeping one client out usually means routing their correspondence through an address you haven’t connected. That might be a separate mailbox for that matter, or the personal mailbox of one lawyer who doesn’t use the assistant. Tell the client which address to use, and tell your team.
The rest of the firm can keep working as before. Ask whether the client’s mail will still reach a connected mailbox through cc’s, forwards or co-counsel. If it will, tell the client that plainly.
California’s guidance adds that a lawyer “must also follow any applicable client instructions or guidelines that may restrict or limit the use of AI.” That’s a sound rule wherever you practice.
Corporate clients and their guidelines
Corporate clients often set their own terms for outside counsel. Read them before you connect the mailbox their matters run through. Look for these points:
- whether AI tools may be used at all, or only with notice
- whether the client must approve a specific tool
- whether client data may leave approved systems or regions
- whether you have to report AI use on invoices
- whether a breach at a vendor triggers a notice duty to them
If the guidelines are silent, Opinion 512’s duty to answer when asked still applies. A short note to the client’s legal department, with your engagement-letter paragraph attached, is usually welcome.
When to tell a client again
A client’s consent or understanding covers the setup you described. Revisit it when something material changes.
- You change tools, or add one that reads mail.
- The vendor changes its terms. Terms carry a version and an effective date. Note both when you write your explanation.
- You give the assistant more room. If replies of some kind start going out without review, a client told otherwise is owed an update.
- A matter becomes more sensitive. A routine file that turns into litigation may call for the consent conversation it didn’t need at first.
California’s guidance also asks lawyers to “evaluate their communication obligations throughout the representation.” An annual look at the engagement-letter paragraph keeps it accurate.
What Point’s documents let you tell a client
Point is an AI email client that sits on the Gmail or Microsoft 365 account your firm already has. Point reads the mail, sorts it, briefs you each morning and drafts replies. Here’s what you can tell a client about it, with the document behind each line.
- Training. Point’s terms, version 1.0 effective July 7, 2026, say at section 6.8 that Point won’t use your content to train generalized third-party or foundation models unless you expressly authorize it.
- Who processes excerpts. Anthropic and OpenAI write summaries and drafts, and Google’s embedding API builds search. All three are named on the dated subprocessor list.
- People at Point. Section 6.8 of the privacy policy lets staff see your mail in named cases only. Those are support you asked for, security and abuse work, legal compliance, troubleshooting, running the service, or your consent.
- Leaving. If you stop using Point, the data processing addendum says your data is deleted or returned within 60 days. You choose which, where your agreement specifies it.
- Who sends. How far Point acts on its own is set for each kind of action, and every kind starts on review. A reply is drafted and waits for you. A gray auto-drafted mark shows Point wrote it. You read it and press send. You can raise a setting later, and at the top of the dial Point acts without asking. For client mail, review is the setting that keeps your engagement letter true.
- A record. Everything Point did is in the activity log, with a time on each line. If a client asks what happened to a message, you can show them.
The rest of what Point does is on the benefits page.
Section 5.1 of the terms leaves attorney-client privilege and client confidentiality with you. That’s where it belongs, and it’s why the conversation with your client is yours to have. Retention details and the regulated-data limits are in what an email assistant reads and keeps.
Point keeps separate businesses you run apart from each other. That line doesn’t divide clients inside one mailbox, as keeping one business isolated from another explains.
Common questions
Do I have to tell clients I use an AI email assistant?
Sometimes, and often it’s wise anyway. ABA Formal Opinion 512 requires informed consent if the tool is self-learning in a way that could disclose client information to others. Where it isn’t, Rule 1.4 may still ask you to explain. A client who asks gets an answer, and so does one whose guidelines require disclosure.
Is a clause in the engagement letter enough?
For consent, no. Opinion 512 says general boilerplate provisions authorizing AI use aren’t sufficient. Consent needs a conversation covering why you use the tool, what it reads, the risks and the benefits. Where only an explanation is owed, a specific engagement-letter paragraph works well.
What should the explanation say?
Name the tool. Say what it reads, which outside AI providers process excerpts, whether it trains on your mail, how long mail is kept, and whether a lawyer reviews replies before they go. Offer the client a way to opt out. Use the vendor’s own documents for each fact.
What if a client doesn’t want AI reading their email?
Record the instruction and honor it. An email assistant reads whatever reaches the connected mailbox. So route that client’s mail to an address you haven’t connected, and tell the client and your team which one to use.
Can I tell a client their mail is walled off from other clients?
Not with most email assistants. Inside one mailbox, the tool reads all of it together. Some tools separate businesses you operate, but that isn’t an ethical wall between clients.
Do I need to tell a client again if something changes?
Yes, when the change is material. That includes a new tool, new vendor terms, or a setting that lets replies go out without review. A matter that becomes more sensitive may also need a fresh conversation.