Kingsfield / AI hallucination cases / In re: the Marriage of Melinda Johnson v. Sabastian …
VERDICT: REJECT

In re: the Marriage of Melinda Johnson v. Sabastian Johnson

ChatGPT, in a contested divorce. Several invented cases across the briefs.

Court
CA Indiana
Decision date
2025-10-30
AI tool (per record)
ChatGPT
Hallucination
Fabricated citation
Court outcome
Warning
Kingsfield verdict
REJECT
In tested set
Yes

What happened

Melinda Johnson, appearing pro se, appealed a Bartholomew Superior Court order that denied her motions to appoint a Parenting Time Coordinator and to terminate a 2023 restraining order. The Indiana Court of Appeals limited its review to the issues from the trial court's February 3, 2025 order.

When the panel checked the legal authority in Johnson's briefs, several of the cases she cited could not be found. The court found that many of her cited cases do not exist. She also relied on Indiana Code section 31-17-2-16 as authority to appoint a parenting coordinator, but that statute covers a child's counseling and does not govern that point. Johnson admitted at trial and in a later motion that she used ChatGPT to prepare her pleadings.

The fabricated citation

Multiple cases the court found do not exist

Offered for: a custody-modification standard

What the court did

The court affirmed the trial court on every issue and denied Johnson's request for appellate attorney's fees. It cautioned attorneys and pro se litigants against using AI for legal research without independently verifying the citations, noting that the fabricated and misapplied authority had severely hindered its review.

How Kingsfield ruled

Kingsfield returned REJECT. Multiple cited cases do not exist. The court attributed the errors to AI-assisted drafting.

This matter is among the US case-law cases we ran Kingsfield against in the Charlotin benchmark. Kingsfield's existence check confirms every citation against the case-law corpus and an independent source (CourtListener) before any claim resting on it can stand. Across the run Kingsfield ruled 696 of 700 testable claims correctly, including every one of the 504 fabricated citations and 92 fabricated quotes.

Honest scope. This measures Kingsfield's ruling layer given an extracted citation, not an end-to-end run over a full brief. The case record and the AI-tool attribution come from Damien Charlotin's public database and the court order; the verdict is Kingsfield's. See the full 700-claim benchmark.

← Back to the live verdict demo