What Three NJ Solo Attorneys Learned the Hard Way About AI Hallucinations in Court Filings

By now, most NJ solo and small-firm attorneys have heard the cautionary tale: a lawyer submits a brief, the citations are fabricated, and the judge is not amused. What gets less attention is the practical question that follows. How does a solo practitioner, billing 50-hour weeks without a paralegal to double-check everything, actually build a reliable process for catching AI hallucinations before they become a bar complaint?

The answer isn't to stop using AI for legal research or drafting. That ship has sailed, and the efficiency gains are real. The answer is to treat AI output the way you treat a first-year associate's work: useful, sometimes impressive, and never submitted without verification.

Here's what that looks like in practice, drawn from conversations with NJ attorneys who have already been burned or come close.

The Hallucination Problem Is More Specific Than You Think

Most attorneys picture hallucinations as obviously fake citations: a case name that sounds plausible, a reporter citation that doesn't exist. But the more common problem is subtler. AI tools frequently get the case name and citation right while mischaracterizing the holding. A real case, cited for a proposition it doesn't actually support.

One NJ family law solo described drafting a motion to modify custody and asking an AI assistant to pull case law on the best-interests standard under New Jersey's governing framework. The citations were real. Two of the holdings were substantially distorted, walking back a nuance the actual court had emphasized. The attorney caught it because she pulled the full text. If she'd been in a time crunch and trusted the AI's parenthetical summary, she would have misrepresented the law to the court.

That's a candor problem under RPC 3.3, which prohibits making false statements of law to a tribunal. The attorney in that scenario would bear full responsibility. "I used an AI" is not a defense, and the New Jersey Rules of Professional Conduct don't carve out an exception for well-intentioned automation errors.

Three Points Where NJ Attorneys Are Getting Tripped Up

First, relying on AI tools that don't cite sources inline. General-purpose tools like ChatGPT don't retrieve live legal databases. They generate text that sounds like legal analysis, sometimes drawing on training data that's outdated or incomplete. If you're using a non-legal AI for research, you need to treat every citation as unverified until you've pulled it from Westlaw, Lexis, or a court's official repository.

Legal-specific tools (CoCounsel, Casetext, Harvey) retrieve actual documents and can show you the source. That's a meaningful difference, but even these tools can misread or mischaracterize what a case says. The source retrieval narrows the risk; it doesn't eliminate it.

Second, skipping verification on procedural citations. Attorneys are more vigilant about case law than about rules. But AI tools also hallucinate procedural rules, court filing requirements, and local rules. An NJ attorney preparing a Superior Court filing described getting a confident, wrong answer about page limits under a specific court's standing order. The tool cited a real rule number, but the limit had been amended. Local court rules change frequently, and AI training data has a cutoff.

Third, using AI for appellate work without a second pass. The Appellate Division and the Supreme Court of New Jersey take citation accuracy seriously. Several NJ appellate practitioners have adopted a two-pass rule: AI drafts the argument, a human verifies every citation independently before the brief is filed. That means pulling each case, reading at minimum the relevant section, and confirming the parenthetical the AI provided actually reflects what the court said.

A Practical Verification Workflow for a Solo Practice

If you're working alone, you can't delegate verification, but you can systematize it. Here's what works:

After AI-assisted drafting, run a citation extraction step. Either manually list every case and rule cited, or ask the AI itself to output a clean list of all authorities it referenced. That list becomes your verification checklist.

Then use Westlaw or Lexis to pull each citation. Don't just confirm the case exists. Read the paragraph the AI summarized. If you're tight on time, prioritize: verify every citation your argument depends on, and spot-check the rest.

For NJ-specific procedural rules, go directly to the New Jersey Courts website and the applicable court's standing orders. Don't trust any AI tool's representation of local rules without checking the primary source.

Finally, build verification time into your flat-fee or hourly estimate from the start. Attorneys who skip this step often do so because they didn't budget time for it. If AI drafting saves you two hours on a motion but you don't budget 30 minutes for citation checking, you've created a hidden liability, not an efficiency gain.

What the NJ ACPE Has Said (and What It Hasn't)

The Advisory Committee on Professional Ethics has addressed AI competence in broad strokes, and RPC 1.1's competence requirement clearly applies: you must understand the tools you use well enough to catch their errors. But the ACPE hasn't issued a formal opinion specifically on hallucination liability or on what a reasonable verification workflow looks like. That gap puts the burden on individual attorneys to define their own standard of care before a grievance or a judicial sanction defines it for them.

If you're a NJ solo attorney who uses AI regularly for research or drafting, document your verification process now, while it's proactive rather than reactive. A short written protocol that you actually follow is far more defensible than an informal habit that you describe from memory after the fact.

The attorneys who are getting this right aren't the ones who avoid AI. They're the ones who stopped treating AI output as a finished product and started treating it as a strong first draft that still needs a lawyer's eyes.