Stop Sending AI-Generated Client Emails Without a Review Protocol — NJ RPC 1.4 Has a Longer Reach Than You Think

Solo attorneys in New Jersey are quietly automating one of the most sensitive parts of their practice: talking to clients. AI tools now draft status update emails, intake follow-ups, deadline reminders, and even settlement summaries — often with minimal attorney involvement before the message lands in a client's inbox.

It feels like a productivity win. It probably is, most of the time. But there's a structural vulnerability buried inside that workflow that most small firms haven't confronted yet, and it runs directly into NJ RPC 1.4.

What RPC 1.4 Actually Requires (And What Most Attorneys Underestimate)

RPC 1.4 imposes two related but distinct obligations. First, an attorney must promptly inform the client of any decision or circumstance requiring the client's informed consent. Second, the attorney must keep the client reasonably informed about the status of the matter and promptly comply with reasonable requests for information.

Most attorneys read "reasonably informed" as a quantity standard — send enough updates, and you're fine. But the New Jersey courts and the Office of Attorney Ethics have consistently treated it as a quality and accuracy standard too. A communication that misstates a deadline, omits a material development, or creates a false impression about case posture doesn't satisfy RPC 1.4 just because it was sent promptly. It may actually make things worse.

This is exactly where AI-assisted client communication creates risk.

The Specific Failure Mode Nobody Talks About

When an attorney uses an AI tool to draft client-facing emails, the failure mode isn't usually a hallucinated case citation — those are caught more easily. The real danger is contextual omission: the AI drafts a plausible, professionally worded message that is technically accurate but missing something only the supervising attorney would know.

Consider a common scenario: your AI drafting assistant generates a case status email based on your matter notes. The notes mention a deposition rescheduled for next month. The AI dutifully reports the new date. What it doesn't know — because it wasn't in the notes — is that opposing counsel informally indicated last week they may seek to depose a second witness, and the client may need to make themselves available twice. That context lives in your head, not your file. The AI sends a clean, confident email. The client makes plans. The second deposition notice lands, and they're blindsided.

Under RPC 1.4, that sequence isn't just an inconvenience. It's a potential failure to keep the client "reasonably informed" about a circumstance that required their attention.

What a Real AI Communication Review Protocol Looks Like

This isn't an argument against using AI to draft client communications. It's an argument for building a lightweight review protocol that makes the automation actually safe. Here's what that looks like in practice for a NJ solo or small firm:

1. Gate on matter complexity, not just message type. Routine scheduling confirmations carry low risk. Any message touching case strategy, procedural posture, deadlines, financial exposure, or settlement should require attorney eyes before sending — regardless of how well the AI drafted it.

2. Create a "what does the client not know yet?" checkpoint. Before approving any AI-drafted status communication, ask yourself one question: Is there anything I know about this matter right now that the client hasn't been told and that this email should include? That question doesn't require re-reading the draft. It requires you to surface your own working knowledge — which the AI will never have.

3. Log your review, even briefly. If you edit an AI draft before sending, note it. If you approve one without changes after review, note that too. A one-line file entry — "reviewed and approved AI-drafted status email, 7/14/25" — creates a record that demonstrates your supervisory engagement. In an OAE grievance context, that paper trail matters.

4. Build client-type templates with standing disclosure language. For clients in active litigation, consider a footer or standard paragraph that sets accurate expectations: case developments can move quickly, this update reflects information as of today, and clients should call with questions. This doesn't waive your RPC 1.4 obligations — nothing does — but it frames the communication as one piece of an ongoing dialogue rather than a complete picture.

The Supervisory Layer Underneath This

There's an additional wrinkle for firms where a paralegal or legal assistant manages the AI drafting queue. If a non-lawyer is the one reviewing and sending AI-generated client emails, you've now stacked a RPC 5.3 supervision issue on top of an RPC 1.4 accuracy issue. The attorney remains responsible for those communications, and "I didn't see it before it went out" is not a defense — it's the problem.

The fix isn't more technology. It's a clear internal policy: no AI-drafted communication that touches substantive case information leaves the firm without attorney sign-off. That policy should be written down, shared with every person in your office, and followed consistently.

The Efficiency Argument Doesn't Disappear — It Just Needs a Guard Rail

None of this undercuts the legitimate value of AI in client communication. Automating first drafts of routine updates, intake confirmations, and appointment reminders saves real time. For a solo attorney managing 40-plus active matters, that matters enormously.

But the efficiency gain is only durable if it doesn't create a grievance liability on the back end. A single RPC 1.4 complaint — especially one where the client can show they acted on an incomplete AI-generated communication — costs far more than the hours you saved.

Build the protocol once. Run every AI-assisted client email through it. The automation still works. It just works safely.