What Do You Do With AI-Generated Legal Scholarship?: An April 2026 Question
Part 1 of 2.
I have a question about how to present the results of legal scholarship generated in part with AI. I pose it as "an April 2026 question" because what AI can do is changing quickly. I would guess that how we think about AI assistance in legal scholarship will change over time, too. But I wanted to explain why I ask, and then open it up for feedback. I'm very interested in your thoughts.
I'm going to present the question in two posts. In this post, I'm going to explain why I turned to AI for help with a scholarly problem I had. In my next post, I will explain what AI was able to do and present my question about what I should do with what AI produced.
Here's the context. A few years ago, I wrote a law review article, Decryption Originalism: The Lessons of Burr, 134 Harv. L. Rev. 905 (2021). The article sought to understand the original public meaning of the Fifth Amendment privilege against self-incrimination and its possible application to unlocking cell phones. It was based on a fascinating historical coincidence: In 1807, in the treason trial of Aaron Burr, there had been an extensive oral argument and then subsequent opinion by Chief Justice Marshall on how the privilege applied to obtaining testimony from Burr's private secretary about an letter in cipher that Burr was thought to have sent.
I wrote my 2021 article based in large part on a transcript of the proceedings made in shorthand by a lawyer in the courtroom. The lawyer, Mr. Robertson, had written everything down: Every argument, every legal source, even all the pincites, in what he claimed was a verbatim reconstruction of the proceedings. The idea of the article was that, given the prominence and experience of the lawyers in the case, the details of the 1807 arguments would likely reflect the Founding-era understanding of the privilege. So my article presented a very detailed reconstruction of what the lawyers relied on, what sources they looked to, and what arguments they made, all based on the Robertson transcript.
That article came out in 2021, and I moved on to other projects.
Just last year, however, I became aware that there is a second and independent transcript. Another lawyer, one Mr. Carpenter, claimed to have done the exact same thing that Robertson claimed to have done. Like Robertson, Carpenter claimed to have written down the whole trial in shorthand, including the legal sources and pincites. Both Carpenter and Robertson had published their transcripts as books shortly after the trial ended. The Robertson transcript is much better known. It is the one referenced in histories of the Burr case, and it was the one that was cited as the report of the trial in 19th Century caselaw. Those references had pointed me to the Robertson transcript, and I had studied it in great detail. I hadn't known the Carpenter transcript even existed.
This created a problem. The premise of my 2021 article is that the Robertson transcript accurately presented the arguments made in the Burr case about the privilege against self-incrimination. But a quick skim of a few spots in the Carpenter transcript suggested that they were not identical. There were things that appeared in one or not the other, or arguments presented somewhat differently, or parts summarized in different ways. If Robertson and Carpenter independently reported the same things, I could be pretty confident that it happened that way. But what if they reported key moments and arguments differently? In that case, I couldn't be confident that my 2021 reconstruction of the privilege arguments in the 1807 Burr trial was accurate.
My scholarly obligation, it seemed to me, was to conduct some sort of comparison of the two transcripts to alert readers to any meaningful discrepancies between them that might relate to my 2021 article. But this would also take a lot of time, as I would first have to go back and re-familiarize myself with the very long Robertson transcript, and then go through all of it and compare everything relevant from my 2021 article with the Carpenter transcript. It's certainly doable, but also pretty time-consuming. It's been on my list of scholarly things-to-do since last year.
And then in March 2026, I wondered: Hmmm, is this something that AI can do for me? These days, AI is really good at going through large documents and summarizing them, comparing them, and the like. And it just gets better and better as the weeks pass. Maybe, instead of going through the two transcripts myself, I can save time by asking an AI service to go through the two transcripts and compare them. Maybe AI can tell me quickly if there are substantive disparities between what Robertson says the lawyers argued and what Carpenter says the lawyers argued.
At least, I figured, it's worth a try. In my next post, I'll say how it went, and ask what I should do with the document AI produced.