Stop Sending AI-Generated Client Emails Without Reviewing RPC 1.4 First

There's a workflow creeping into NJ solo practices that looks like efficiency but carries real professional responsibility exposure: an attorney prompts an AI tool to draft a client status update, skims it for two seconds, and hits send.

The email is grammatically clean. It sounds like the attorney. And it may be completely wrong about the status of the client's matter.

This is the RPC 1.4 problem that almost nobody is talking about — and it's arguably more urgent right now than the confidentiality concerns that dominate most AI ethics conversations.

What RPC 1.4 Actually Demands

New Jersey's RPC 1.4 requires attorneys to keep clients "reasonably informed about the status of a matter," to "promptly comply with reasonable requests for information," and to "explain a matter to the extent reasonably necessary to permit the client to make informed decisions."

That last clause is the one AI tools are most likely to undermine. Client communication isn't just about transmitting facts — it's about ensuring the client understands those facts well enough to exercise meaningful judgment. A fluent, confident AI-generated paragraph about "next steps in your litigation" can obscure strategic nuance that a client genuinely needs to grasp before authorizing action.

The NJ Supreme Court's disciplinary decisions make clear that RPC 1.4 violations don't require bad intent. Neglect — even unintentional neglect — routinely grounds formal discipline. An attorney who delegates communication drafting to an AI tool and rubber-stamps the output is not communicating with the client. The AI is. And the AI has no idea what happened in last Thursday's call, what the client's actual risk tolerance is, or that opposing counsel just sent a letter that changes the calculus entirely.

The Three Failure Modes to Watch

1. Stale or context-blind drafts. AI tools draft from what you feed them — a prompt, a prior email thread, maybe a document. They don't know what you know. If your matter status has shifted since the last file note you uploaded, the AI will confidently draft an update that reflects the old reality. Clients receiving stale information can make decisions — sometimes irreversible ones — based on it.

2. Oversimplification of legal risk. AI tools optimized for clear, readable prose have a structural bias toward plain-language reassurance. That's often good. But when a client needs to understand that their contract claim has a meaningful chance of dismissal on a threshold legal question, "clean" writing can shade into misleading minimization. RPC 1.4's duty of explanation requires accuracy, not just clarity.

3. Volume-driven dilution of judgment. The whole point of AI-assisted communication is throughput — drafting ten client updates in the time it used to take to draft two. The risk is that review quality degrades proportionally. If you're reviewing AI drafts at AI speed, you're not really reviewing them.

A Practical Review Layer for NJ Small Firms

The fix isn't to stop using AI for client communication — it's genuinely useful for routine status updates, acknowledgment emails, and appointment confirmations. The fix is to build a deliberate checkpoint into the workflow.

Before sending any AI-drafted client communication, run a quick three-question check:

• Is this factually current? Does the draft reflect the actual, present status of the matter as of today — not as of the last document the AI was trained on or the last prompt you fed it?
• Does this require a decision? If the communication is prompting the client to authorize something, approve a settlement range, or make a strategic choice, an AI draft is a starting point only. That communication needs to be written — or substantially revised — by you.
• Would this hold up in a grievance? Imagine the ACPE reviewing this email. Does it demonstrate that the client was meaningfully informed, or does it look like form correspondence that could have been sent to anyone?

If you're using a tool like Clio Duo, MyCase IQ, or a general-purpose LLM with your matter data, consider adding a simple file note step: before generating a client update, spend 60 seconds writing three bullet points about the current matter status. Feed those bullets to the AI as the first part of your prompt. You'll get a far more accurate draft — and you'll have forced yourself to actually think about the matter before the communication goes out.

The Bigger Picture

NJ's disciplinary infrastructure has not yet produced a published decision specifically about AI-generated client communications. That window won't stay open indefinitely. The Bar's general posture — as reflected in guidance tracking the ABA's Formal Opinion 512 framework — is that existing RPCs apply fully to AI-assisted work. RPC 1.4 was written for a world of handwritten letters. It applies just as forcefully to a ChatGPT output you approved without reading carefully.

The attorneys who get this right won't be the ones who avoid AI. They'll be the ones who use it deliberately — with a review layer that's proportionate to the stakes of each communication, not the speed of the tool generating it.