Some court opinions arrive with a dramatic bang. Others stroll in quietly, carrying a stack of footnotes and the legal equivalent of a raised eyebrow. Zelma v. Wonder Group belongs to the second category. On the surface, it is a Telephone Consumer Protection Act case about two verification text messages. Underneath, though, it is something more interesting: a modern cautionary tale about what happens when shaky legal writing collides with artificial intelligence, questionable citations, and a judge who clearly was not in the mood to play “guess the real precedent.”
That is why this decision matters. The opinion is not just about TCPA pleading standards, consent, or whether a text message can be dressed up as a solicitation wearing a fake mustache. It is also about credibility. And in litigation, credibility is the currency that buys everything from patience to persuasion. Lose it, and even your decent arguments start looking like they showed up in a trench coat full of fake parentheticals.
What Happened in the TCPA Case?
The plaintiff alleged that Wonder Group sent him two unsolicited verification texts and sued under the TCPA and New Jersey law. The court did not buy every theory in the complaint. It dismissed a TCPA do-not-call claim after concluding the verification texts did not promote goods or services and therefore did not qualify as telephone solicitations. The court also rejected related state-law theories that depended on a private cause of action that was not actually there. Still, the opinion was not a total wipeout for the plaintiff. Counts II and III survived the motion to dismiss because the court found the allegations about the use of an automatic telephone dialing system were enough, at least at that early stage, to move forward.
That split result is important. This was not a case where the court simply tossed everything into the nearest legal dumpster and called it a day. Instead, the judge separated the weak claims from the potentially viable ones. That makes the opinion more useful, not less. It shows that even when a court is deeply skeptical of a party’s briefing, it can still do the careful work of sorting the case on the merits.
The Brief Became Its Own Side Plot
The real headline came from the court’s discussion of the plaintiff’s opposition brief. Judge Evelyn Padin noted that the filing contained inaccurate quotations and citations. Some quotations were fabricated from real cases. Other cited cases appeared not to exist at all. That is the sort of discovery that makes judges reach for stronger coffee and litigators reach for their malpractice carrier’s number.
The court ordered the plaintiff to disclose whether he had used generative artificial intelligence while drafting the opposition and to explain the discrepancies. The response did not exactly calm the room. According to the opinion, the plaintiff said he had seen a “wave of AI-based services” while doing research and had tested one platform for unrelated input, but insisted that the experience reinforced his decision to rely on his own TCPA archive and trusted legal databases. He also suggested some quotation marks were used where he intended to paraphrase and that some missing authorities reflected either misreading a source or summarizing it poorly.
The court was unconvinced. In one of the opinion’s sharpest lines, the judge said the explanation “strains credulity.” That phrase lands because the court did not view the plaintiff as a novice stumbling around the courthouse with a map upside down. The opinion notes that he had brought nearly two dozen cases in the district and demonstrated familiarity with caselaw, pleading rules, and even citation formatting. In other words, this was not the legal equivalent of a first driver clipping the mailbox on day one. The court believed the plaintiff should have known better.
Why This Matters in TCPA Litigation
TCPA cases are already technical before AI enters the chat. They turn on distinctions that look small until they suddenly become expensive: whether a message was promotional or informational, whether consent was given and revoked, whether a platform qualifies as an autodialer, whether a prerecorded or artificial voice was used, and whether state-law claims ride alongside federal ones or fall flat. In that world, legal authority does not merely decorate a brief. It does the heavy lifting.
That is exactly why AI misuse is so dangerous here. A fake citation in a generic contract dispute is bad. A fake citation in a TCPA brief can be catastrophic because the law is dense, fast-moving, and full of edge-case reasoning. One invented quote about consent, autodialers, or verification texts can bend the entire argument out of shape. Suddenly the brief is not analyzing the law; it is fan fiction with Bluebook punctuation.
The Zelma opinion illustrates the point. The court highlighted a quotation attributed to Van Patten v. Vertical Fitness Group that did not exist. That was not a minor typo or a dropped page number. It went to the heart of the plaintiff’s attempt to argue that verification texts still trigger TCPA liability when sent without consent. If the quote is fake, the theory starts wobbling immediately. Once that happens, the judge is not just evaluating your argument. The judge is evaluating whether anything else in the brief deserves trust.
AI Is Not the Lawyer, and the Court Knows That
One of the most useful lessons from this case is also the least glamorous: courts do not care whether a bad citation came from a chatbot, a rushed associate, a sleepy paralegal, or a cosmic burst of bad judgment. The filing has a name on it. That name belongs to a human being. The responsibility stays there.
That principle is getting louder across the American legal system. Courts have been confronting AI-generated hallucinations in filings since 2023, and judges are increasingly moving beyond stern warnings. Reuters has reported sanctions, fee awards, and growing frustration in federal courts when briefs include made-up cases or mangled quotations. The broader message is simple enough to fit on a sticky note: if you file it, you own it.
That trend is especially important because AI tools are not only being used to draft prose. They are also being used to summarize cases, generate research leads, produce parentheticals, and smooth out arguments. Those tasks sound harmless, even efficient. Sometimes they are. But efficiency becomes very expensive the moment a tool invents a case, misstates a holding, or turns a cautious precedent into a bold proposition it never actually supported.
The TCPA Is Already an AI Story
The irony here is that the TCPA is not just the setting for an AI-tinged briefing problem. It is also one of the statutes being reshaped by AI itself. The Federal Communications Commission has made clear that the TCPA restricts robocalls and robotexts absent prior express consent or an applicable exemption. The FCC has also clarified that AI-generated voices count as “artificial” voices under the statute. Meanwhile, the FTC has been updating telemarketing protections with AI-enabled scam calls very much in view.
So the legal system is managing two AI problems at once. First, AI is changing how businesses communicate with consumers through voice and text technologies. Second, AI is changing how lawyers and litigants write about those communications in court. That is a messy combination. It means the same technology that is complicating compliance is also complicating advocacy. If that sounds inefficient, congratulations: you have discovered law in 2026.
For defendants, this means compliance teams and litigation teams can no longer live on separate planets. If marketing uses AI voice tools, legal needs a rock-solid consent framework, documentation discipline, and a plan for how those systems are described in court. For plaintiffs, it means the temptation to let a generative tool fill in the research gaps is particularly risky in a statute where wording and precedent matter so much. The court may forgive a novel argument. It will not happily forgive an imaginary one.
What Lawyers and Legal Teams Should Learn
The first lesson is procedural humility. Use AI for brainstorming, organizing issues, or turning a chaotic research pile into a cleaner outline if you must. But do not let it be the final authority on law, quotes, or parentheticals. A machine that sounds confident is still not a substitute for pulling the case, reading the page, and checking whether the sentence in your brief exists in the real universe.
The second lesson is workflow. Good legal teams need a verification system that is mind-numbingly boring and therefore incredibly valuable. Every case cited should be opened. Every quotation should be matched to the source. Every parenthetical should be reviewed against the actual holding. If AI touched the draft, someone needs to know where, how, and what got checked afterward. This is not anti-technology. It is pro-not-getting-embarrassed.
The third lesson is strategic. Once a court spots one fake quotation, the damage spreads beyond the bad cite itself. Opposing counsel will start reading every line like it came from a suspicious chain email. The court may do the same. At that point, even a strong argument can lose force because the advocate has burned the trust required to make close questions break in their favor. In litigation, losing an argument is painful. Losing credibility is compound interest on pain.
Experience From Practice: What This Kind of Mess Looks Like in Real Life
In practical terms, these AI briefing blowups tend to unfold in a painfully predictable way. First comes the rush. A deadline is tight, the law is technical, the record is annoying, and someone decides a generative tool can help “speed things up.” That phrase should probably come with caution tape. The draft that comes back looks polished, confident, and impressively structured. It also may contain one or two little time bombs disguised as elegant prose. Nobody hears the ticking because the sentences sound smart.
Then comes the verification stage, except sometimes it is not really a stage so much as a vibe. One citation is checked. Another is skimmed. A quote looks plausible enough, and the brief gets filed. Then the other side reads it carefully. Or the judge does. Or a clerk, blessed with excellent eyesight and very limited patience, notices that the proposition in the brief does not seem to match the case being cited. That is when the floor starts to wobble.
From there, the experience gets ugly fast. Lawyers have to explain why a case cannot be found, why a quotation is not on the page, or why a parenthetical says something broader than the decision ever held. Even when the answer is “we relied on AI and failed to verify,” that explanation rarely helps much. Courts hear it as a confession that the filing process broke down at the exact point where professional judgment was supposed to show up and do its job.
In TCPA matters, the problem is even worse because the legal issues are so precise. A single made-up line about prior express consent, revocation, solicitation, or autodialer capacity can distort the entire frame of the case. Suddenly the team is no longer arguing about the statute. It is arguing about whether it can be trusted to describe the statute. That is a terrible trade. Nobody wants to spend oral argument defending a quotation mark.
What many litigators are seeing now is a shift in judicial patience. Early AI cases sometimes drew lectures and warnings. Increasingly, courts are reaching for sanctions, certifications, or detailed disclosure requirements. That shift makes sense. Judges are not beta testers for your workflow. They are not required to absorb the cost of your convenience experiment. If anything, the emerging message from courts is that legal writing can use modern tools, but only when the human being behind the filing acts like an adult with a bar card and a red pen.
There is also a quieter experience that gets less attention: the cost to the side that did nothing wrong. Opposing counsel has to waste time checking fabricated authorities. Clients pay for that time. Judges and clerks spend resources untangling arguments that should have arrived in usable form. The system absorbs friction because someone trusted a tool that was never designed to carry the ethical weight of a federal filing. That is why these episodes irritate courts so much. They are not just sloppy. They are contagious.
The smartest response is not panic and not a total ban on technology for every task. It is discipline. Use tools where they help, keep humans firmly in the loop, and treat every citation like it will be checked by the most skeptical reader in the building. Because eventually, it will be.
Final Takeaway
Zelma v. Wonder Group is a useful legal snapshot of the AI era: a TCPA case, a brief with serious citation problems, a judge unwilling to shrug them off, and a larger legal culture that is getting noticeably less forgiving. The opinion does not say artificial intelligence has no place in legal work. It says something more practical and more enduring. Courts expect filings to be accurate, quotable, and real. Revolutionary stuff, apparently.
That should be the takeaway for lawyers, self-represented litigants, compliance teams, and anyone else orbiting the TCPA. AI can save time, but it cannot borrow credibility for you. And when a court starts questioning whether your authorities exist, whatever time you saved has already become the most expensive shortcut in the room.