Court opinions: Use of AI leads to loss of remuneration

The integrity of the basis for decision-making forms the foundation of every constitutional procedure. If findings of fact are based on unreliable information, the desired legal peace is permanently jeopardized. A recent ruling by the Darmstadt Regional Court now makes it clear with remarkable consistency that delegating core expert tasks to artificial intelligence (AI) without appropriate disclosure leads to a complete loss of entitlement to remuneration.

When expertise becomes an illusion

In a civil lawsuit brought before the Darmstadt Regional Court, a professor of medicine was commissioned as an expert to prepare a medical report. He invoiced a fee of €2,374.50 for his services. However, the court refused to set this amount in full in the fee determination proceedings (decision of November 10, 2025, ref. 19 O 527/16).

The rejection was based on serious doubts about the authorship and quality of the submitted work. The expert had failed to examine the plaintiff in person, which is a basic requirement for the requested medical findings. Instead of providing well-founded reasons, the document merely provided brief answers to the court’s questions. When the court asked for clarification, the responses remained evasive and insufficient. The regional court ultimately concluded that the text had been generated primarily by artificial intelligence.

Identification of technological patterns in assessment

In its reasoning, the Darmstadt Regional Court explained in detail the characteristics used to identify AI authorship. The court recognized typical linguistic and structural patterns that are characteristic of current language models:

• Sentence structure and redundancy: A noticeable accumulation of identical word repetitions and a monotonous sentence structure consisting almost exclusively of main clauses with uniform sentence beginnings.
• Formal inconsistencies: In the report, the expert referred to himself using his full address within the body text, which is unusual behavior for a human author in this context.
• Artifacts of prompt processing: Particularly serious were formulations that read like written responses to a query as to whether a previous prompt (command) had been executed correctly. Such textual relics indicate a lack of editorial control over the AI output.

These cumulative deficiencies led the court to classify the expert’s own contribution as indeterminable or, at most, marginal.

The legal classification

Legally, the decision is based on a strict interpretation of civil procedure law. According to the Code of Civil Procedure (Section 407a (1) sentence 2 ZPO), an expert is obliged to disclose if he is unable to carry out the assessment alone or if he makes use of the assistance of third parties. The expert opinion must always be prepared personally (Section 407a (3) ZPO).

If an expert opinion is not prepared personally in violation of these obligations, whereby artificial intelligence is treated legally as an undisclosed third author, it is inadmissible in court proceedings. The consequences for remuneration are set out in the Judicial Remuneration and Compensation Act (Section 8a (2) JVEG). According to this, a service is only billable to the extent that it remains usable. Since the regional court completely rejected the usability due to the lack of investigation and the generated content, the claim was consistently reduced to €0. In addition, the Judicial Remuneration and Compensation Act (Section 8a (2) sentence 1 no. 2 JVEG) applies in the event of defective services. The omission of a physical examination in a medical expert opinion constitutes such a fundamental defect that the result becomes worthless for the judicial assessment of evidence.

The significance for out-of-court dispute resolution

Although the decision was made in the context of a state court case, the implications for mediation and other forms of dispute resolution are immense. In mediation, too, expert opinions often form the objective framework for making complex issues tangible for both parties. If facts are recognized or development processes are evaluated on the basis of expert knowledge, this expertise must be based on a valid method and personal
responsibility. An expert opinion that merely reflects the computing power of artificial intelligence, without a human expert taking responsibility for validation and liability, undermines the trust of those involved. In a consensus process based on voluntariness and transparency, such an approach would rob the entire agreement of its credibility and jeopardize the hard-won legal peace.

Professionalism in the digital transformation: A warning signal

The Darmstadt case is not an isolated incident. In the past, lawyers have already attracted negative attention for having AI models produce legal documents without checking them. Last year, the Cologne District Court sharply criticized such behavior. It argued that this made it more difficult to find the law, misled uninformed readers, and seriously damaged the reputation of the rule of law and the legal profession.
Both experts and legal representatives are not relieved of their personal responsibility for content by the use of technological aids. Artificial intelligence can serve as a tool for structuring or research, but it must never replace professional analysis and personal perception.

Significance for practice

The decision of the Darmstadt Regional Court should be understood as a clear signal for quality assurance in the field of expert opinions. Professional actors must be able to rely on expertise being provided and accounted for personally. For clients commissioning expert opinions, whether government agencies, companies, or private individuals, this means paying closer attention to formal and content-related AI indicators.
In the age of digitalization, true professionalism is characterized more than ever by transparency and irreplaceable human expertise. Only on this basis can conflicts be resolved sustainably and results achieved that stand up to legal and economic scrutiny.

Source: LTO, No money for AI-generated court reports, January 6, 2026