You’ve seen the headlines: attorneys facing court sanctions and bar reprimands—resigning and losing coveted positions—over court filings plagued by AI-generated fiction. Technology is wreaking havoc, and all the newfangled AI hocus-pocus is the culprit.

Or is it?

In truth, AI has never been responsible for a single word in a court brief. Nor will AI ever be responsible.

The lawyer who creates and signs a brief—who decides whether any AI-generated text stays or goes—is responsible for every single word. Period.

The alarming headlines needn’t exist. If today’s lawyers simply do what lawyers routinely did back in 1985, 1995, 2005, or 2015, headline writers will have no fodder. The rules are simple:

• Lawyers must Shepardize¹ every case they cite.
• Before citing a case, a lawyer must read it.

That’s it. AI drama over. Headlines gone.

By going old school, new-school AI users can be safe, ethical, and accurate. And they’ll avoid the consequences seen in a recent Michigan Court of Appeals case, in which the court ordered an attorney to “personally” pay his opponent “actual damages, including reasonable attorney fees,” because he’d “relied on artificial intelligence without adequate verification.”² That lapse had “le[d] him to cite nonexistent cases and invoke real cases for propositions they do not support.”³

THE NEW CITE-CHECK TRIFECTA: REVERSED, OVERRULED, OR HALLUCINATED

For decades, lawyers have used citator services, including Shepard’s (LexisNexis) and KeyCite (Westlaw), for a variety of reasons. The main reason, though, has been to discover whether a case has been reversed or overruled—in other words, to ensure that a case “is still good law” before relying on it.

But in the past few years, a new motivator has emerged, and it’s a doozy: Does the cited case actually exist? As we’ve seen from the headlines about careless lawyers, the answer has been no too many times.

When AI gives you a phantom case and you type the citation into LexisNexis or Westlaw, you’ll suddenly find yourself swimming in the middle of a strange court opinion that has nothing to do with your issue (and that might not even be from your jurisdiction). This unexpected plunge into AI humbuggery is enough to shake any researcher—but better you than your judge.

Now more than ever, lawyers must check and verify every citation before filing a brief or submitting a memo to a supervisor. Every time. Every case. It’s an indispensable step. It always has been.

THE NEED TO READ

Lawyers must also read their cited cases because AI sometimes summarizes real cases in fake ways. In the past few years, I’ve read, for example, an AI-generated passage that turned a case involving a broken deck rail into a case involving a snowmobile accident. In another document, AI had transformed a young child retrieving a stray ball from a neighbor’s yard into a criminal-minded teen trying to break into the neighbor’s car. The cases were 100% real, but AI’s descriptions were pure fantasy.

Generative AI’s trustworthiness is illusory. Its speed, convenience, and cyberswagger lull some lawyers into believing it. But it isn’t trustworthy. I’ve heard it said that generative AI is a people-pleaser, and that’s why it’s so dangerous. It will give us what we want whether it exists or not. And there’s no magic prompt to guarantee that the information it generates is accurate.

So lawyers, beware. We must read every case we cite. No exceptions. Traps are lurking for those who take shortcuts—traps with consequences.

YES, THERE IS TIME

Cite-checking is no longer the library slog it was in the bound-volume era. Decades ago, law clerks and associates instructed to “Shepardize the cases” in a brief realized that they’d just been sentenced to hours of toil.

But with the advent of LexisNexis and Westlaw, those days are long gone. Today, we simply drag our mouse to copy a citation, paste it into the search bar, and click enter. This click reveals whether the case has been reversed or overruled—and whether the case exists.

Reading cases takes time, yes. But not so much time that you can risk the alternative: filing a court brief with false information, being sanctioned, and facing disciplinary proceedings. And if you’re a supervising attorney who doesn’t believe that your duties include reading or checking cited cases,⁴ you’d best ensure that someone else reliable and qualified reads the cases cited in your brief to verify its accuracy.

EFFICIENCIES, YES—SHORTCUTS, NO

I wish that my observations were more cutting edge. The truth is, I’m at least two years late. Back in 2024, for instance, the Second Circuit wrote, “At the very least, the duties imposed by Rule 11 require that attorneys read, and thereby confirm the existence and validity of, the legal authorities on which they rely.”⁵

During a recent Legal Writing Institute AI Committee webinar, one presenter noted that the number of cases in which lawyers have been sanctioned for AI misuse has eclipsed 1,400. And now there’s Michigan precedent confirming that AI hallucinations are grounds for severe sanctions—including sanctions against attorneys “personally.”

We can build efficiencies in the practice of law, but there are no shortcuts. So if you use AI to research or generate text for court briefs, remember that old practices have new value. They’re a lawyer’s best protection against making headlines for the wrong reason.