by Avi Gesser and Michael Pizzi

From left to right: Avi Gesser and Michael Pizzi (photos courtesy of Debevoise & Plimpton LLP)

In Part 1 of this series, we “vibe coded” a game designed to help train lawyers and professional staff on the firm’s new AI policy using Generative AI. In Part 2, we demonstrate how lawyers can use AI tools to rapidly transcribe and summarize audio recordings from congressional hearings, oral arguments, or depositions, and transform them into a variety of easy-to-digest formats. Specifically, we used NotebookLM and Gamma to transcribe a recent 30-minute Second Circuit oral argument and then create a two-page written summary, a “podcast”-style audio summary, and a PowerPoint presentation—all in under nine minutes.

Audio of Oral Arguments

Suppose you represent a client in a false advertising lawsuit. The case raises several legal issues that are currently being litigated in similar cases in courts around the country. To advise your client on strategy, you need to keep track of the arguments being made in these other cases, assess how they are landing with judges, and anticipate questions you might get asked at an upcoming oral argument in your own case. Many of the courts publish audio recordings of the arguments on their websites, but no transcripts.

How to Transcribe and Summarize with AI

For this demo, we used a recent oral argument in a false advertising case that is currently before the Second Circuit (Zesty Paws LLC v. Nutramax Laboratories, Inc.). Taking it step by step:

• We downloaded the audio recording as an .mp3.
• We created a new NotebookLM notebook and then dragged and dropped the .mp3 into the “Sources” window.
• In the “Chat” window, we instructed NotebookLM to create a complete transcript of the audio file with speaker tags and time stamps.

Example Prompt: “Please create a transcript of this recording. Put the speaker names in bold and add time stamps every one minute.”

• We then asked NotebookLM to create a summary of the argument.

Example Prompt: “I’m an associate at a law firm and I need to provide a client with an accurate two-page summary of the argument. My firm has no involvement in this case. Can you prepare that for me?”

• We also created an audio summary of the oral argument to listen to on our morning commute. To create the podcast, we started by generating a longer summary using the “Chat” window to give NotebookLM more content to work with.

Example Prompt: “Can you do a five-page summary instead?”

• We saved this five-page summary as a Source by selecting “Add” in the “Sources” menu, copying the text into “Paste text”, and clicking “Insert”. Back in the main notebook view, we unselected the .mp3 file so that NotebookLM would only summarize the five-page summary when generating the podcast. We then selected “Audio Overview” in the “Studio” menu and, in about five minutes, had a podcast-style, 17-minute summary of the oral argument that we could download or stream on the NotebookLM app.
• While the podcast generated, we turned this summary into a client-ready presentation. We opened Gamma, selected “Paste in text”, then copied in the five-page summary from Step 5. We chose “Generate” from the options below and, using the fields in “Text Content”, instructed the tool on our desired level of detail, format, language, tone and design. We made a couple adjustments to the default settings, including the following:

Amount of Text: Minimal

Write For: “Client, mix of business and legal professionals with limited knowledge of trademark/advertising litigation.”

• Finally, using the “Agent” feature, we instructed Gamma to add an additional slide, right after the title slide, that provides an overview of the case.

Things to Keep in Mind

We haven’t thoroughly QC’ed the accuracy of these summaries, and we did notice a few typographical errors in the transcript—e.g., “a stoppple” instead of the legal term “estoppel”. But a quick comparison of these materials against the recording suggests they captured the general thrust of the legal arguments at the hearing.

Of course, as lawyers, we need to carefully consider the risks inherent in using these or similar tools for client work and take several steps to mitigate those risks. In addition to carefully QC’ing any outputs for accuracy, you might also:

• consider adding a disclaimer stating that these summaries are preliminary, were generated using AI tools, have not been checked for accuracy, should not be relied on as an accurate summary of the argument, etc.;
• avoid uploading any confidential client information into these tools, unless your firm has an enterprise / licensed version, and such use is not contrary to any client instructions; and
• stick to trusted source material, such as an official unedited audio recording from the Second Circuit’s website

Avi Gesser is a partner and Michael Pizzi is an associate at Debevoise & Plimpton LLP. This post was originally published on the Debevoise Data Blog.

The views, opinions and positions expressed within all posts are those of the authors alone and do not represent those of the Program on Summer 2025 Compliance & Enforcement Blog Corporate Compliance and Enforcement (PCCE) or of the New York University School of Law. PCCE makes no representations as to the accuracy, completeness and validity or any statements made on this site and will not be liable any errors, omissions or representations. The copyright of this content belongs to the authors and any liability with regards to infringement of intellectual property rights remains with the authors.