AI-Generated Legal Scholarship: Publishing and Authorship of AI-Assisted Research on Aaron Burr's Trial

AI-Generated Legal Scholarship: Publishing and Authorship Challenges

In my previous post, I tasked an AI model with comparing two transcripts from the 1807 treason trial of Aaron Burr. The resulting document raises critical questions: Should I publish it, and if so, how? How do I define my role in its creation? Am I a co-author, the sole author, or merely a prompter? To address these questions, I first explain how the AI memo was generated.

How the AI Memo Was Created

The transcripts I needed to compare were PDFs of two-volume books from 1807 and 1808, totaling hundreds of pages. However, I focused only on the sections discussing the privilege against self-incrimination. I used Claude (Opus 4.6 extended) and instructed it to compare the legal arguments about the privilege against self-incrimination to evaluate the accuracy of my 2021 article, which was based on the Robertson transcript, in light of the Carpenter transcript.

The process involved approximately 30 rounds of prompting over several hours. As I refined my instructions, I discovered the limitations of Claude and pushed it to improve its output. For example, my initial request to compare the full PDFs was declined, as the task was deemed too complex. I then simplified the task: "Read my 2021 article, understand its claims about the Burr trial, and compare them to the Carpenter transcript."

The first draft provided a useful starting point but fell short of my expectations. Over time, I developed a method to guide Claude effectively. Eventually, it agreed to directly compare the two transcripts based on the claims in my 2021 article. However, the output required extensive refinement, particularly for direct quotes and page references. Claude struggled with pagination, especially since one of the PDFs combined two volumes without clear separation. I had to manually verify and correct errors, which Claude then adjusted. A significant breakthrough occurred when I realized that Claude could compare screenshots of the transcripts.

Key Challenges in AI-Assisted Legal Research

• Task Complexity: Directly comparing lengthy historical legal transcripts exceeded the AI's initial capabilities.
• Refinement Process: Achieving accurate results required iterative prompting and manual corrections.
• Technical Hurdles: Pagination inconsistencies in source documents posed significant obstacles.
• Output Accuracy: Ensuring precise quotes and references demanded extensive verification.

Next Steps: Publishing and Authorship

With the AI-generated memo complete, the next critical step is determining how to publish and attribute it. Should it be shared informally online or formally through a journal? How should my role—as a prompter, co-author, or sole author—be defined? These questions will shape the future of AI-assisted legal scholarship and its acceptance in academic and professional circles.