After Using AI to Draft Fee Agreements, NJ Attorneys Are Finding Problems That RPC 1.5 Doesn't Forgive

A solo attorney in a general practice firm told me recently that she'd started using an AI tool to generate her engagement letters and fee agreements. It saved her real time. The output was clean, professional, and sounded authoritative. She'd been doing it for months before a client pushed back on a bill and she pulled up the agreement to defend her position.

The agreement didn't say what she thought it said.

Not because the AI hallucinated anything. The facts were accurate. The problem was subtler: the language was generic enough to be technically compliant under a national standard while quietly missing several requirements that New Jersey's RPC 1.5 specifically imposes. The client dispute exposed the gap. A fee arbitration filing came next.

This is the failure mode nobody talks about with AI-generated legal documents: the output is often good enough to feel finished without being specific enough to be enforceable in your jurisdiction.

What RPC 1.5 Actually Requires in New Jersey

New Jersey's version of RPC 1.5 is worth reading closely, because it goes further than the ABA Model Rule in a few meaningful ways. Under RPC 1.5(b), any new client engagement or any new matter for an existing client requires a writing, communicated "before or within a reasonable time after commencing the representation", that covers the rate or basis of the fee, the scope of the representation, and the expenses the client will be expected to pay.

That much most attorneys know. What gets missed in AI-generated drafts:

The contingency fee requirements under RPC 1.5(c). New Jersey's contingency fee rules are governed partly by RPC 1.5(c) and heavily by Court Rule 1:21-7, which sets specific caps, requires itemization of litigation expenses before deduction, and mandates a written closing statement. A generic AI fee agreement template will produce language like "attorney shall receive one-third of any recovery" without flagging that NJ Court Rule 1:21-7(c) sets a sliding scale for personal injury matters and that deviating from it requires court approval. If your AI tool wasn't specifically trained on New Jersey court rules, it almost certainly won't catch this.

The requirement that fee-sharing arrangements be disclosed. If you're referring this client out or sharing fees with co-counsel under RPC 1.5(e), the agreement needs to address that specifically. AI drafts tend to omit this entirely unless you explicitly prompt for it, and most attorneys don't know to ask.

Clarity on what "expenses" means for your specific practice. NJ ethics opinions have addressed disputes over whether clients were adequately informed about costs like filing fees, expert witness fees, and electronic discovery expenses. A vague "client agrees to reimburse all out-of-pocket expenses" clause may not survive a fee arbitration challenge. AI drafts consistently produce this kind of catch-all language.

Why the AI Draft Feels Right But Isn't

The reason this problem is hard to spot is that the AI output reads like a real fee agreement. It uses the right vocabulary. It covers the obvious bases. The issue is that large language models produce text that reflects the statistical center of whatever they were trained on, and the statistical center of fee agreement language is a national average, not New Jersey's specific court rules, not the ACPE ethics opinions that interpret RPC 1.5 in this state, and not the IOLTA fund's guidance on handling advanced fee deposits.

When you read the draft, your pattern recognition as an attorney says "this looks like a fee agreement." That's working against you.

A Practical Fix That Doesn't Require Abandoning AI

The tool isn't the problem. The prompt is.

Instead of asking an AI to "draft a fee agreement for a personal injury matter," build a prompt that forces the output to account for NJ-specific constraints. Include a reference to RPC 1.5(b) and (c), Court Rule 1:21-7 for contingency matters, and any specific expense categories your practice routinely incurs. Then treat the output as a first draft that gets checked against a short manual checklist before you send it to a client.

That checklist doesn't need to be long. For a contingency matter in New Jersey, it might cover six items: the sliding scale from Court Rule 1:21-7 is correctly applied; the expense deduction order (before or after attorney's fee) is explicitly stated; the closing statement obligation is referenced; scope limitations are specific rather than general; advanced fee handling aligns with your IOLTA obligations; and the client's right to fee arbitration under the NJ Fee Arbitration program is disclosed if your retainer requires it.

For hourly matters, the list is shorter but still requires attention to how billing increments are described, what triggers additional fee letters, and how disputed invoices are handled.

The attorney who came to me after her fee arbitration filing had been using a good AI tool. Her problem was that she'd skipped the human checkpoint at the end. She's since built a one-page NJ-specific review sheet that lives next to every AI-generated fee agreement before it goes out. That's the right architecture: AI does the first draft, attorney does the jurisdiction-specific verification.

If you're using AI to generate fee agreements and you haven't built that checkpoint yet, building it this week is a more useful hour than almost anything else you could do for your practice.