ChatGPT for NJ Discovery Drafting vs. Purpose-Built Legal AI — Which One Actually Belongs in a Small Firm?

Solo and small-firm attorneys in New Jersey are increasingly reaching for AI tools to help draft interrogatories, document requests, and discovery responses. The question isn't really whether to use AI for this work anymore — it's which kind of AI, and whether the choice you make on a Tuesday afternoon is one you'd be comfortable defending to a client, an adversary, or the Office of Attorney Ethics.

Two categories dominate the conversation: general-purpose large language models (LLMs) like ChatGPT, Claude, or Gemini, and purpose-built legal AI platforms like Harvey, Clio Duo, or CoCounsel. They are not interchangeable. Understanding where they diverge is one of the most practically useful things a NJ solo practitioner can do right now.

---

What "General-Purpose" Actually Means in Practice

When you open ChatGPT and paste in a set of facts to generate interrogatories, a few things are happening beneath the surface that matter to your practice:

• Your prompt may train the model. Unless you're on a paid enterprise plan with data retention turned off, OpenAI's default terms have historically allowed user inputs to improve future models. That's a data governance problem when your prompt contains client facts.
• The model has no legal memory. It doesn't know whether NJ court rules require discovery responses within 30 days (they do, under R. 4:17-4), whether a specific interrogatory format is expected in your venue, or what local customs look like in Mercer County Superior Court.
• Hallucination risk is real and domain-specific. General LLMs regularly fabricate case citations. For discovery drafting this risk is lower than for brief research — but boilerplate that mischaracterizes the scope of a document request or misstates a privilege standard can cause real problems downstream.

That said, for a solo attorney without budget for premium legal software, a carefully structured ChatGPT workflow — with enterprise privacy settings enabled, a detailed system prompt, and mandatory attorney review — can be legitimately useful. It's a scalpel in untrained hands, not a toy.

---

What Purpose-Built Legal AI Gets Right (and What It Doesn't)

Platforms designed specifically for legal work address several of these gaps. The better ones offer:

• Data isolation by client matter, so your discovery draft for Client A doesn't bleed into Client B's context
• BAA (Business Associate Agreement) or equivalent data processing agreements — critical if you work on any matter with HIPAA-adjacent facts
• Jurisdiction-aware templates that reflect NJ-specific procedural rules
• Citation grounding, where outputs are anchored to actual documents rather than generated from statistical patterns

The tradeoff is cost. Purpose-built legal AI platforms typically run $100–$500/month per seat at the entry tier — a real line item for a solo operation billing under $400K annually. And not all of them live up to their marketing. Some "legal AI" products are simply ChatGPT with a law-themed interface and a premium price tag. The underlying model is identical; the data governance may be no better.

Before signing up for any legal AI platform, ask the vendor directly: Does your product use a proprietary fine-tuned model, or is it a wrapper on a third-party LLM? Who holds the data, and where is it stored? What happens to my firm's data if I cancel?

---

The NJ Ethics Layer You Cannot Skip

New Jersey's RPCs don't name specific AI tools, but they establish a framework that makes the general-vs.-purpose-built question an ethics question, not just a preference.

Your duty of competence under RPC 1.1 — as interpreted through the NJSBA's emerging guidance and the broader ABA commentary on technological proficiency — requires that you understand the material limitations of the tools you use. Choosing a general-purpose LLM for discovery drafting without understanding its data handling defaults, hallucination tendencies, or lack of procedural grounding isn't just a workflow risk. It's a competence question.

RPC 1.6 adds the confidentiality dimension. Submitting client facts to a general-purpose AI with default data-retention settings on a free or consumer plan is almost certainly a confidentiality problem. The fix isn't necessarily switching platforms — it's knowing and controlling the settings of whatever you use.

---

A Practical Decision Framework for NJ Solo Attorneys

Ask yourself these four questions before committing to either type of tool for discovery work:

1. What are the actual data handling terms? Read the privacy policy and terms of service. Look for a DPA (Data Processing Agreement). If the vendor won't provide one, treat that as a red flag.
2. Does the output require heavy manual correction? If you're spending 45 minutes cleaning up a draft that a paralegal could have done in 30, the efficiency gain is an illusion.
3. What's your fallback review process? Every AI-drafted discovery document should clear a checklist that includes: procedural compliance with NJ R. 4:17 and R. 4:18, privilege log triggers, and client-specific fact accuracy.
4. Can you explain your tool choice to your client? If a client asked how their matter facts were handled during the drafting process, you should have a clear, honest answer ready.

---

Neither category of tool is universally superior. A well-configured general-purpose LLM can outperform a poorly designed "legal AI" product. A purpose-built platform with strong data governance and genuine NJ procedural grounding can meaningfully reduce your risk exposure and improve output quality — if you vet it properly.

The attorneys who will get the most out of AI-assisted discovery drafting aren't the ones who picked the flashiest tool. They're the ones who understood what they were choosing, and why.