The Bench Goes Digital: How Judges Are Using AI to Read, Write, and Test Briefs

For the past year, most of the attention around AI in legal writing has focused on lawyers.
What happens when a brief is AI generated? What risks does it introduce? What disclosures are required?

All fair questions. But they miss the more important development.

Judges are using AI too.

A March 2026 random sample survey of federal judges from the New York City Bar found that more than 60% of responding judges had used at least one AI tool in their judicial work—most commonly for legal research and document review. That does not mean judges are handing decision making to machines. It does mean that AI is already entering the workflow in chambers across the country: helping with research, summarizing records, testing arguments, and speeding up the early stages of drafting. Quietly but unmistakably, the bench is becoming a technological participant—not just an observer.

AI as a judicial drafting partner

Start with writing.

Judges and clerks have always relied on templates, prior opinions, and internal memos to produce decisions efficiently. AI is a natural extension of that workflow. Used carefully, it can summarize long records and procedural histories, suggest clearer phrasing for dense passages, generate draft structures for orders or opinions, and surface weaknesses or unanswered questions.

Take US District Judge Xavier Rodriguez in Texas. He has publicly described using AI to turn large filings into timelines, summarize claims, flag weaknesses in arguments, and prepare questions for hearings—then, after reaching his own conclusions, sometimes using it to help generate draft language for later review and revision. As he told The Washington Post, “My law clerks would be wasting 30, 45 minutes, an hour, developing a chronology of events. This thing does it instantaneously.”

But here is the key: judges are not outsourcing authorship. They are accelerating iteration.

The best uses of AI on the bench resemble a disciplined editing process. Draft, test, refine. Not generate, paste, and file.

In that sense, AI is less like a ghostwriter than a highly responsive clerk—one that never tires, but also cannot be trusted without supervision. As Judge Scott Schlegel of the Louisiana Fifth Circuit Court of Appeal has put it: “Treat AI like a first-year law clerk and double-check everything.”

AI as a tool for interrogating briefs

If AI is changing how judges write, it may be changing even more how they read.

A judge reviewing a brief today has new options. With the right tools, judges and their clerks can check whether a cited case actually supports the proposition it is cited for, compare a quoted passage against the original text, identify when authority is stretched rather than simply cited, and surface relevant cases neither side emphasized.

This is a real shift.

Traditionally, verifying a brief required time: pulling cases, reading opinions, cross-checking quotations, testing whether a case really went as far as the advocate claimed. Now much of that work can be accelerated dramatically.

That does not make judges less careful. It makes careful scrutiny cheaper.

And when scrutiny becomes easier, it becomes more likely.

From “good law” to “truthful use”

Legal research tools have long answered one question: Is this case still good law?

AI-enabled workflows increasingly help answer a different one: Is this case being used correctly?

That distinction matters.

A case can be perfectly valid and still be quoted out of context, applied too broadly, or cited for a proposition it does not quite reach. The old problem was whether the authority existed and remained intact. The newer problem is whether the authority is being used faithfully.

Judges are increasingly equipped to catch those moves—not just through experience, but through tools that highlight mismatches between claim and authority.

In effect, AI is helping courts move from validity checking to fidelity checking.

Internal guardrails against hallucinations

Much of the public conversation around AI and hallucinations has focused on lawyers who submitted fabricated cases. The original warning shot was Mata v. Avianca, Inc. in 2023, when a New York attorney used ChatGPT to draft an opposition brief containing six entirely fabricated case citations. Since then, Damien Charlotin, a researcher at HEC Paris has documented more than 1,300 AI hallucination cases worldwide.

Courts took notice.

And in some instances, they learned the lesson the hard way. In 2025, two federal judges—one in Mississippi, one in New Jersey—were forced to withdraw or correct orders after staff used consumer AI tools like Perplexity and ChatGPT without adequate verification. One clerk reportedly used Perplexity as “a foundational drafting assistant.” An intern used ChatGPT for legal research in a draft opinion. The resulting work contained serious errors, including fabricated citations, misstated authorities, and factual mistakes. Both episodes drew scrutiny from Senator Chuck Grassley.

That is why guardrails matter.

Formal or informal, many courts are now building them around a few basic ideas: AI outputs must be verified against the record and the law; quotations and citations must be checked manually; and core judicial functions cannot simply be delegated to a machine.

The goal is not to ban AI. It is to ensure that anything touched by AI is still grounded in source material, reviewed by humans, and owned by the judge.

That discipline is shaping judicial norms in real time.

A more exacting—but less forgiving—process

There is a paradox here.

AI can make writing more efficient, more polished, and more accessible. But it also makes weakness easier to detect.

Judges who use AI in their own workflow understand both sides of that equation. They see how quickly text can be generated. They also see how quickly it can be tested.

That dual perspective changes expectations.

It makes courts less forgiving of sloppy citations, overread cases, imprecise quotations, and arguments that sound plausible on first read but weaken under closer scrutiny.

Not because judges are harsher. Because the cost of scrutiny is dropping.

And where chambers have access to tools that can compare briefs side by side and flag omitted authority, misquoted language, and stretched holdings, the standard of care shifts.

What lawyers should take from this

The practical implication is not that lawyers should avoid AI.

It is that they should assume courts are increasingly able to meet them at the same technological level—or higher. In April 2025, Thomson Reuters secured a multi-year contract giving the federal judiciary access to Westlaw Precision with CoCounsel, covering more than 25,000 legal professionals across the federal courts.

That means every citation should survive immediate verification. Every quotation should be exact. Every characterization of a case should reflect the actual holding. And every AI-assisted draft should be treated as a starting point, not a finished product.

The gap is no longer just between lawyers who use AI and lawyers who do not. It is between lawyers who review rigorously and lawyers who do not.

That was always true. AI just makes the consequences more visible.

Final thought

The story of AI in legal writing is often framed as a question of adoption. Who is using it?
How quickly? With what safeguards?

But from the court’s perspective—and from conversations with judges and clerks grappling with these questions—the more revealing question is different.

What does AI expose?

It exposes whether a brief holds up under pressure. Whether authority is used carefully. Whether the logic is sturdy enough to survive fast, systematic scrutiny. Whether the advocate has done the hard work of precision rather than the easier work of fluent prose.

Judges are not just adapting to AI. They are using it—carefully, unevenly, but increasingly—to sharpen the standards that define good legal writing.

And in that environment, the fundamentals have never mattered more.