BriefCatch 創辦人兼行政總裁 Ross Guberman 專訪系列

Ross Guberman, Founder and CEO of BriefCatch, is a leading authority on legal writing with more than two decades of experience training lawyers, judges, government agencies, corporations, and professional organizations. A former practicing attorney, he has conducted thousands of legal-writing workshops across three continents and has spent nearly a decade training new federal judges on opinion writing. Guberman holds degrees from Yale University, the Sorbonne, and the University of Chicago Law School and is the author of Point Made: How to Write Like the Nation’s Top Advocates and Point Taken: How to Write Like the World’s Best Judges, as well as co-author of Deal Struck. His contributions to legal writing and innovation have earned him recognition including the Legal Writing Institute’s Golden Pen Award and a place on the Fastcase 50.

BriefCatch is a legal writing and editing platform designed to help lawyers produce clearer, more precise, and more persuasive documents. Built on Guberman’s decades of research into effective legal writing, the platform works within Microsoft Word and uses AI and natural language processing to provide real-time editing suggestions, writing analytics, and guidance tailored specifically to legal professionals. BriefCatch is used by law firms, judges, courts, and government agencies and is designed to improve documents such as briefs, motions, opinions, memos, and contracts while also serving as an ongoing writing-development tool for attorneys.

You founded BriefCatch in 2018 after years of teaching legal writing to lawyers, judges, and law firms. What problems did you repeatedly see in legal writing that convinced you software could help, and how has that original vision evolved as AI has entered the profession? 

I kept seeing the same problems: sentences that pile on multiple introductory phrases, repetitive verbs and sentence patterns, and long quotations treated as self-explanatory. These habits are common in the work product of experienced lawyers, too. Under client and deadline pressure, even the best writers crave specific suggestions, not more mantras about being “clear and concise.”

In the one-off workshops that were my bread and butter, I shared tips, tricks, and techniques drawn from the work of the greatest legal writers. The lawyers and judges who attended seemed enthusiastic and even inspired, but then real life intervened. With the right software, I thought, I could offer similar help at any moment, day or night. The next thing I knew, I was an accidental founder of a start-up that turned thousands of patterns in legal writing into explainable (and codable) edits and examples.

GenAI has been an earthquake for my life’s work. It can generate seductively polished first drafts. More ominously, it can make a weak argument sound airtight before anyone has tested whether it holds up, fooling even the most astute readers. But harnessed in the right tools and workflows, GenAI can be a godsend for enhancing legal prose, verifying accuracy and integrity, and poking holes in arguments and analyses, whether a person or a machine wrote them.

BriefCatch is now being reintroduced as a unified legal writing suite rather than primarily an editing tool. What drove this evolution, and how much of it was a response to the rapid adoption of generative AI in legal workflows?

Generative AI sped things up, but these tools belonged together all along, no matter who, or what, writes the first draft.

Lawyers need to ask three related questions: Have I made my points well? Do the authorities support the argument? What can I learn from this draft for the next one? BriefCatch, CiteCheck, RealityCheck, and the Learning Hub are built to answer them. A sentence can carry a good argument but make the reader work too hard. Or it can sing off the page but lack support.

Access matters, too. Although much of my career has been spent serving the world’s biggest firms, I would love for solo practitioners to benefit from the same writing guidance and verification tools that Am Law 200 firms do. The idea is to make scrutiny and learning part of a lawyer’s everyday work instead of something you get once a year at a training workshop.

Generative AI can dramatically accelerate the first draft of a legal document, but BriefCatch is increasingly focused on what happens after that draft is created. Do you believe verification and quality control will ultimately become more important than generation itself in legal AI? 

Yes. I expect the lion’s share of legal AI’s value to come from refining, verifying, and deepening existing drafts. Writing faster is appealing, but it doesn’t help you discern which arguments will resonate, which authorities support them, or which sentences make logical leaps.

The hard part remains: how do you tell when a polished paragraph laden with citations is a house of cards?

RealityCheck goes beyond determining whether a case or citation exists by assessing whether the authority actually supports the proposition for which it is cited. Why is this distinction so important, and what types of errors are the hardest for AI systems—and lawyers—to identify? 

You can have a case that’s still good law, a citation with nothing wrong with it, and a quote that matches the opinion word for word, and the case still won’t back up the point the brief is making.

Judges don’t write in sound bites. A court will often lay out a party’s argument at length, in that party’s own words, and then spend three pages taking it apart. The line that seems perfect for your brief turns out to come from a concurrence. Analogies are worse. They can look fine at first glance, and you only see the problem when you go back and read the facts.

So RealityCheck reads the cited passage against what the brief says about it. Sometimes that’s a simple word-for-word comparison. Other times it means reading around the passage to figure out whether the court actually relied on it or was just passing through on the way to something else.

A verification tool also has to know its limits. Sometimes a reading is debatable rather than wrong, and sometimes the tool just couldn’t pull reliable data. “I couldn’t verify this” and “this is wrong” are very different messages, and a lawyer deserves to know which one they’re getting, with the source passage right there to check.

Much of the concern around AI in law has focused on completely fabricated cases, but subtler errors such as altered quotations, misstated holdings, or real cases cited for unsupported propositions may be harder to spot. Which AI-generated mistakes concern you most? 

The ones that worry me make a source look a bit more helpful than it really is. A fake case tends to get caught eventually. A small distortion can sail right through.

Say an appellate opinion holds that a trial court “may” do something under certain conditions. The draft turns “may” into “will.” Or it gets the quote right and quietly drops the condition.

I see the same problem with quotes that are accurate but asked to do too much. The brief says “courts hold,” and the cited passage turns out to be one judge’s aside. It isn’t even a holding.

So matching the words is just the start. We built RealityCheck to ask whether the brief has shifted what the source means or how far it reaches. And to be fair, lawyers were making these mistakes long before AI came along.

Where should the line be drawn between AI automation and human judgment in legal writing? Are there parts of researching, drafting, editing, and verifying a brief that you believe should always remain the responsibility of the lawyer? 

I’d put the line roughly where generating text ends and strategy begins.

I’m happy to let AI help with research, drafting, editing, and checking, and it’s good at finding the soft spots in an argument. Brainstorm with it. But the theory of the case is yours. So is what you concede, what you leave out, and how you handle the bad facts in front of a skeptical judge. Those calls depend on the record, on what the client wants, and on risks a model can’t fully see.

The skill I’d invest in now is reviewing. Lawyers need practice tracing a proposition back to its source, spotting the step an argument skipped, and noticing when a suggested edit changes what a sentence means.

BriefCatch can also be used to analyze opposing counsel’s briefs for questionable citations or unsupported legal assertions. How do you see AI-powered verification changing the way lawyers review and challenge one another’s arguments? 

I expect verification tools like ours to push opposing counsel to up their game. It’s a perverse sort of fun for our users when they catch their opponents citing an authority that simply isn’t up to the task.

The same tools can arm you, too. Rather than declaring an argument “meritless” or a case “inapposite,” you can identify the proposition, explain what the cited court actually decided, and highlight why the difference matters.

But don’t weaponize trivia. A minor pin-cite error doesn’t, by itself, justify a credibility attack. A debatable reading doesn’t prove dishonesty. Treat a flag as a reason to dig further before you put anything in a reply.

The same standard has to apply to your own brief. If all goes well, we’ll see fairer, sharper arguments rather than dueling briefs full of tech-generated hyperbole.

The expanded platform combines software with the Learning Hub and your legal-writing curriculum. Why was it important to build education into the product rather than simply giving users increasingly powerful AI-powered editing tools? 

Because improving the document and improving the lawyer are different goals, and we care about both. “Accept all” may be tempting, but it won’t make you a better writer.

Take a heading that says “Causation.” Replacing it with a heading explaining why causation is missing does more than change the wording. It forces you to pinpoint where the doctrine meets your facts.

That’s why we pair our software with explanations, examples from leading legal writers, and the Learning Hub. A suggestion fixes the draft before filing; the entire suite helps you internalize the underlying patterns.

BriefCatch emphasizes grounding its recommendations in real judicial opinions and legal filings. As large language models become more capable, how important will domain-specific data and legal context be in differentiating specialized legal AI from general-purpose AI assistants? 

Every time the general models improve, the bar rises for specialized products. I welcome that. A business built on the hope that those models will never learn certain things isn’t much of a business.

Legal-specific data matters, but it’s hardly enough. Most of all, you need to truly understand how lawyers and judges think and work and even feel.

In legal writing, smoother prose is the easy part. Did an edit preserve the meaning? Did a proposed heading advance the argument? Did a verification tool identify a material discrepancy—or did it turn a defensible reading into a false alarm? A specialized product has to get those calls right.

We draw on real briefs and opinions because legal writing has its own demands. But calling a product specialized proves nothing. What lasts is sound judgment, rigorous testing, and a workflow that lets lawyers check the results and put them to use.

Looking several years ahead, what does the ideal AI-assisted legal writing workflow look like? Do you envision AI primarily becoming a better drafting partner, an always-on reviewer and verification layer, or something considerably more autonomous?

For me, the centerpiece is an always-on reviewer, working alongside the drafting partner and leaving consequential decisions to the lawyer.

In the future workflow I imagine, a lawyer or AI revises a draft, and the system tracks the substance of the changes. If “some courts” becomes “all courts,” the support needs to be checked again. If an edit drops an important exception, the tool warns the lawyer. It should speak up when a change matters and otherwise stay quiet.

Before filing, the lawyer sees what was checked, which concerns remain open, and where the system couldn’t reach a reliable conclusion. A blanket stamp saying the document has been “verified” tells you very little.

A lot of the work can be automated. The final judgment can’t be. My ideal is not a lawyer who supervises a machine at arm’s length. It’s a lawyer with more time to decide what the brief should say—and more energy to say it just right.

Thank you for the great interview, readers who wish to learn more should visit BriefCatch.