In an ordinary case, neither the attorneys nor the court have time for an in depth analysis of the potentially relevant jurisprudence.In the Germanic jurisdictions, they usually rely on a "commentary". A commentary summarises the doctrine of a statutory provision, as it has evolved in the jurisprudence of the competent courts. A large fraction of German legal academia is busy preparing and updating commentaries. A reasonably close analogue in the common law countries is code annotated.

In this project, Christoph Engel and Johannes Kruse use the guarantee of freedom of assembly, as protected by Art. 11 European Convention of Human Rights, for a proof of concept. They show that the entire process can be delegated to a large language model (GPT-4o, in this version of the commentary), starting from downloading the pertinent rulings of the European Court of Human Rights, over summarizing the doctrinal statements in each ruling, to compiling a structured summary of the discussion in all pertinent rulings, separate per element of doctrine. The result, i.e. the commentary entirely written by GPT, is made available on a dedicated website.

The result is not only much cheaper than competitive products prepared by human jurists. It is also comprehensive (while all human-written commentaries are selective). It is very precise (individual paragraphs of rulings are referenced) and convenient (each reference consists of a link to the cited paragraph). With little effort, the commentary can be updated, at much closer intervals than human-written texts. As this proof of concept shows, the process can be extended to statutes for which no commentaries exist, or to substantive questions covered by multiple statutory or jurisprudential rules.

In the companion paper, we explain the process of having GPT write a commentary in detail, compare its performance with the human-written competition, and put the project into the context of LLMs mustered for legal practice and legal research.