Five Things NJ Solo Attorneys Should Audit Before Using AI for Discovery Document Review

Discovery document review is one of the most time-consuming, margin-crushing tasks in solo and small firm practice. It's also the workflow where AI assistance looks most attractive—and where the gap between "this saved me hours" and "I missed a critical document" is paper-thin.

New Jersey attorneys are increasingly reaching for AI-powered review tools to handle large-scale productions in civil litigation, family matters with financial records, and employment disputes. But unlike document drafting, where a hallucinated clause is usually visible on its face, AI errors in discovery review can be invisible—a relevant email that never surfaced, a privileged document that slipped into production, a date range quietly misapplied.

Before you run your next document set through any AI review tool, work through this five-point pre-deployment audit.

---

1. Confirm the Tool's Training Data Is Legally Relevant—Not Just "Legal"

Many AI document review platforms market themselves as built for legal work, but their underlying models may have been trained primarily on transactional or regulatory datasets. If you're reviewing employment discrimination records, that's meaningfully different from reviewing M&A due diligence documents.

Ask your vendor directly: What corpus was this model fine-tuned on, and how does it handle NJ state court discovery standards versus federal ESI protocols? If they can't answer specifically, treat that as a red flag. NJ's discovery rules under R. 4:10 and the court's active ESI Working Group guidance have nuances that a generic legal AI model may not weight correctly.

---

2. Test Privilege Detection Before You Rely on It

Privilege review is where AI-assisted discovery carries its sharpest risk. Inadvertent disclosure of attorney-client communications or attorney work product in New Jersey litigation can trigger RPC 8.4 concerns if the disclosure reflects a pattern of recklessness—not just an honest mistake.

Before deploying any tool on a real matter, run a controlled test: seed a document set with known privileged materials (attorney-drafted memos, communications clearly within the scope of representation) and verify that the AI flags them correctly. Most tools will not achieve 100% recall. The question is whether the miss rate is acceptable given the stakes of the matter, and whether your human review layer is calibrated to catch the gap.

Document your test results. If a privilege issue surfaces later, your ability to demonstrate a reasonable pre-deployment validation process matters—both to a court and to a malpractice carrier.

---

3. Map the Tool's Output to Your Production Obligations

AI document review tools typically produce a relevance score, a privilege flag, or a category label. None of those are a production determination. That judgment still belongs to you.

Under RPC 3.4, NJ attorneys have an unambiguous obligation to not suppress evidence and to produce what is required. AI output that you treat as a final answer—rather than a first-pass filter—inverts the professional responsibility structure. Build an explicit workflow step where a lawyer (you or supervised staff) reviews the AI's near-threshold calls: documents it rated as borderline relevant, borderline privileged, or borderline responsive.

The tool does the lifting. The attorney makes the call. That sequence isn't optional—it's the whole point.

---

4. Verify Data Handling Against Your Confidentiality Obligations

When you upload a client's documents to an AI review platform, you are transmitting confidential client data. Under NJ RPC 1.6, that obligation doesn't pause because you're using a third-party tool—it extends through the tool.

Your pre-deployment audit should include a hard look at:

• Where the data is stored (cloud region, data residency commitments)
• Whether the vendor trains on your uploads (opt-out vs. opt-in language in the terms of service)
• What the breach notification window is under their contract versus NJ's 72-hour notification norm under the Identity Theft Prevention Act

If you signed up for a tool via a self-serve checkout flow and accepted default terms, there's a reasonable chance you haven't locked down any of these points. Revisit those terms before the next matter goes in.

---

5. Create a Written Record of Your Methodology

Courts in New Jersey and across the country are increasingly receptive to—and occasionally suspicious of—AI-assisted discovery processes. Being able to produce a clear, written description of your review methodology is not just good practice; it's becoming a competitive and professional differentiator.

Your methodology memo doesn't need to be long. It should capture: which tool you used, what version, how you configured the review parameters, what human review layer you applied, and how you handled edge cases. If opposing counsel or a court challenges the adequacy of your production, that document is your first line of defense.

It also reinforces something essential: that AI was your assistant, not your attorney of record.

---

The Honest Bottom Line

AI-assisted discovery review can legitimately transform the economics of small firm litigation practice. A solo attorney who once had to choose between an underfunded review and a disproportionate time investment now has a third path. But that path has a toll booth—a set of professional obligations that don't disappear because the technology is impressive.

Run the audit. Build the workflow. Document the methodology. The efficiency gains are real, and so are the consequences of skipping these steps.