Mata v. Avianca, cite by cite

On June 8, 2023, Judge P. Kevin Castel made a forceful point in Mata v. Avianca: every citation in the brief in front of him could have been checked, in minutes, against a free public database. Attorney Steven Schwartz agreed on the record. Two weeks later, Judge Castel sanctioned Schwartz, LoDuca, and their firm for the eight fabricated cases in the brief.

The brief was an opposition to Avianca Airlines' motion to dismiss a routine personal-injury action. Peter LoDuca filed it on March 1, 2023, but his colleague Steven Schwartz had drafted it using ChatGPT for legal research. Eight of the cases the brief cited did not exist. When the court ordered counsel to produce copies of the opinions, counsel went back to ChatGPT and filed its output as exhibits to a sworn affidavit. These "opinions" were fabricated too.

CitationClerk is built on the premise Judge Castel articulated: we extract citations from legal documents and verify them against public databases. Verification is done by straightforward, deterministic code, not AI. Briefs are never sent to Google, OpenAI, Anthropic, or any other third-party AI service, so your client's confidences stay with you.

CitationClerk checks whether a cited case exists and whether its quotations are accurate. It does not check whether a case is still good law.

All eight fabricated cases flagged

As an experiment, we ran the Mata brief through CitationClerk. All eight fabricated citations were flagged:

Note: the original document was a poor-quality scan, resulting in small typos when read. Verification is unaffected.

Real cite, wrong case

Most of the fabricated citations were inventions on every axis. One was subtler: Peterson v. Iran Air, 905 F. Supp. 2d 121 (D.D.C. 2012). The reporter is real, the volume is real, and the page exists, but on that page is United States v. Iss Marine Services, Inc., an unrelated decision. ChatGPT had borrowed the citation format from a real case.

A note on the spelling: the brief cited this as "Peterson v. Iran Air," so that is the spelling shown here. The fabricated opinion counsel later filed under the same citation was captioned "Petersen." The case does not exist under either spelling.

Here is the row from the CitationClerk report:

A tool that only asks "does this reporter, volume, and page exist?" would get the wrong answer. CitationClerk checks both the cite and the case name attached to it.

Real case, different name

A name mismatch doesn't always mean fabrication. Sometimes the case is real, and the database just stores it under a different form of the name.

For example, in the real case El Al Israel Airlines, Ltd. v. Tseng, CourtListener stored the case with the respondent's name spelled out: "Tsui Yuan Tseng". Less careful tools wrongly flag this Supreme Court decision as fabricated. Despite the mismatch, CitationClerk recognizes the abbreviated party name and verifies cleanly.

Real case, totally different name

The court actually ordered counsel to produce copies of nine cases, not just the eight fabricated ones. The ninth was In re Air Crash Disaster Near New Orleans, La., 821 F.2d 1147 (5th Cir. 1987), a real Fifth Circuit MDL case.

CourtListener stores the case under one of its lead-party captions: Trivelloni-Lorenzi v. Pan American World Airways, Inc. The brief's "In re" form and the database's Trivelloni-Lorenzi form are two names for the same case, with no shared words between them. A naive matcher would flag this Fifth Circuit MDL as fabricated.

But CitationClerk doesn't stop at the surface case name; it also matches against the full case name on file in the public record.

Checking the quotations

CitationClerk goes beyond just verifying that a case exists. It also checks the accuracy of quotes.

The Mata brief has a quote from a real case: Ashcroft v. Iqbal, 556 U.S. 662 (2009), for the proposition that a complaint must allege more than "threadbare recitals of the elements" to survive a motion to dismiss. CitationClerk retrieved the actual Iqbal opinion text from CourtListener, located the quoted phrase, and matched it.

Checking statutes and treaties

Some citations aren't cases. The Mata brief also invokes the Bankruptcy Code, the Federal Rules of Civil Procedure, New York state law, and two international treaties. CitationClerk verified all six:

U.S. statutes and federal rules were matched against Cornell Law's online code (LII), the New York statute against the state code's official publication, and the conventions against archived treaty text.

Conclusion

The Mata brief is a good stress test for verification tools. It has eight fabricated cases, one famous Supreme Court decision with an abbreviated party name, one real Fifth Circuit MDL stored under a different name, a quotation claiming to be from Iqbal to check against the Iqbal opinion, and citations to non-cases like statutes and treaties. On this brief, CitationClerk flagged all eight fabrications, verified both real cases, located the quote word-for-word, and matched every statute and treaty against its authoritative source. All with zero calls to OpenAI, Anthropic, Google, or any other third-party model. The brief was checked, in minutes, against a free public database, as Castel said any lawyer could do.