Stop Using AI for Client Intake Before You Read NJ RPC 5.5

Automated client intake is one of the most attractive AI applications for a NJ solo or small firm. Set up a chatbot on your website, feed it a few prompts about your practice areas, and it starts collecting facts, answering preliminary questions, and even generating a draft engagement summary before you've had your first cup of coffee. The time savings are real. The ethical exposure, though, is something most attorneys haven't thought through carefully.

New Jersey RPC 5.5 prohibits the unauthorized practice of law. That rule is usually discussed in the context of out-of-state attorneys practicing in NJ without a license. But it has a second dimension that matters here: a licensed NJ attorney can't assist a non-lawyer in doing something that constitutes the practice of law. An AI tool that gives legal advice to prospective clients, even tentative, conditional, "this is not legal advice" advice, is a non-lawyer doing lawyer things. And if that tool is running under your firm's name, on your website, you are the one who set it up.

Where the Line Actually Sits

The hard question is what counts as "legal advice" versus "information gathering." Courts and bar authorities across the country have struggled with this distinction for decades, and AI doesn't resolve it. It sharpens it.

Collecting a prospective client's name, the county where their accident happened, and the date of injury is intake. Asking follow-up questions that apply legal standards to those facts, "Was the other driver cited at the scene? That affects whether you have a strong negligence claim", is starting to look like legal advice. An AI chatbot that dynamically responds to user input and starts narrowing the legal theory before a licensed attorney has reviewed anything is on thin ice.

The NJ Supreme Court's Committee on the Unauthorized Practice of Law has taken a broad view of what crosses the line. Guided questionnaires that generate legal conclusions, even with disclaimers, have drawn scrutiny nationally. The disclaimer problem is real: slapping "this is not legal advice" on an AI response doesn't change the functional character of what the tool is doing. If a prospective client is making decisions about their legal situation based on your chatbot's output, the legal system tends to treat that as legal advice regardless of the label.

What NJ Small Firms Are Actually Doing Wrong

Most of the intake automation I see in small NJ firms falls into one of two patterns. The first is an off-the-shelf chatbot (Tidio, Drift, or a legal-specific tool like Intaker or Crisp) configured with prompts that were clearly written by someone focused on lead conversion, not ethics compliance. The prompts are aggressive. They respond to user descriptions of their situation with output that sounds confident and legally specific. The attorney never reviewed the prompt logic.

The second pattern is a general AI assistant (often ChatGPT with a custom system prompt) embedded in a contact form or website widget. The attorney set it up in an afternoon because it seemed useful, gave it a persona and a description of the firm's practice areas, and didn't think through what the model would generate in response to edge cases.

In both patterns, the attorney hasn't documented the AI's behavior, hasn't tested it for the kinds of legally loaded responses it might generate, and has no supervisory mechanism to catch problems after launch.

A Practical Configuration Standard

If you want to automate intake without running into RPC 5.5 problems, the configuration logic matters more than the tool you choose. Here's a working standard:

The AI should collect facts only. It should not interpret them. If a prospective client describes a workplace injury, the tool should collect the date, the employer, the nature of the injury, and whether a report was filed. It should not tell the user whether they "have a case," whether the statute of limitations applies, or whether their employer's conduct sounds like negligence. Those conclusions belong to the attorney after reviewing the intake summary.

Every AI-generated intake summary should be treated as a raw document, not a legal assessment. Build your workflow so the attorney reviews the summary before any substantive communication goes to the prospective client. An auto-reply confirming receipt and setting an appointment is fine. An auto-reply that says "based on what you've shared, it sounds like you may have a valid claim" is not.

Test your intake tool against hard inputs before you launch it. Ask it about facts that could involve complicated statute of limitations questions, comparative negligence, or jurisdictional issues. See what it says. If it sounds like legal analysis, rewrite the prompt or disable the dynamic response features and replace them with a static form.

Finally, document your configuration. Keep a record of the system prompt you've given the tool, the date you last reviewed it, and any changes you've made. If a complaint ever lands at the OAE, you want to show you thought about this and built in guardrails.

Intake automation isn't inherently problematic. An AI that schedules appointments, collects basic facts, and routes inquiries to the right attorney is a legitimate efficiency tool. The problem is that the same technology, configured carelessly, can start doing legal analysis on your behalf without your knowledge. In New Jersey, that's your license on the line, not the vendor's.

Audit your intake flow this week. Pull up the prompts you've given whatever tool you're using, read them as if you were the OAE, and ask whether a prospective client acting on the output could claim they received legal advice. If the answer is yes, reconfigure before the next intake form comes in.