Interrogating the Script (Practical use Cases of Notebook LM)

One of the more practical AI Applications is Google's Notebook LM. This week I take a look at some practical application for editors AND I make an AI podcast on Goodfellas.

With the initial hype (and panic) surrounding the introduction of ChatGPT and other artificial intelligence tools beginning to settle, the novelty has worn off a little and a sense of practical use has started to take hold. The environment is now more open to real-world applications of large language models. One of the best examples I’ve come across is Notebook LM.

Notebook LM is Google’s attempt at letting you “have a conversation with your notes.” The tool allows you to collect information, files, audio, and videos on a single subject, and then use a large language model (in this case, Google’s Gemini) to interrogate the material. One of its most talked-about features is the ability to generate your own podcast, where two characters casually discuss your chosen data files.

Of all the AI tools that have emerged since ChatGPT first hit the market, Notebook LM is the first one that I feel has a practical real world use case for my day-to-day work. Recently, I’ve been exploring interesting ways of applying it to my work - in particular working with the script or screenplay.

Getting the Overview

While I always make time to read through all the scripts on a project, it can take a while to get through eight to ten hour-long episodes. Often, you don’t get access to all the scripts until a week or so after joining a project. So when you’re rushing to get everything in motion, it’s not easy to dive into ten 60-page scripts straight away.

One of my favourite tools for this is an app called Talking Scripts. Created by Suspicion director Stefan Schwartz, it uses AI to read your scripts aloud. While a bit unusual at first, you quickly get used to it. I often listen to scripts on my commute or during morning workouts—similar to a podcast. It’s great for quickly catching up on the series and a fantastic first pass of the script.

In a similar vein, the AI-generated podcast summaries that Notebook LM creates are an interesting way to digest story content. You can interrogate the material during mundane tasks or while on the go. It's not the most studious way of learning a script, of course, but using downtime to familiarise yourself with the material is a compelling additional method of consumption.

Below is a Notebook podcast that I “produced” based on the screenplay for Goodfellas.

Setting the Scene

At the beginning of a project, one thing I always create is a scene-by-scene breakdown of important data and information. This includes one-line summaries of each scene (which I use for scene title cards), and a spreadsheet listing all scenes along with character appearances and other key details.

In the past, this level of organisation was usually reserved for bigger productions—where you could delegate to a trainee who’d spend weeks building spreadsheets manually. With Notebook LM, I can now do this very quickly. With a bit of human tweaking, I’ve even set up tasks like assigning each character a colour and applying that colour to clips in each scene based on who is most prominent in that scene (something which Adam Gough talked about doing recently in interviews on Disclaimer)

---

Organising Character and Story Timelines

Timelines (character and story - not the edit tool) often get muddled. Many shows now jump back and forth in time, and during the edit, it can be challenging to keep track of the story order. This is something Notebook LM handles brilliantly. It can take the scripts and generate a sequence of events—and you can also ask specific questions.

For example, this question on an episode of Peaky Blinders:

Where is Tommy when Polly is first talking to Reuben?

Notebook LM’s response:
When Polly is first talking to Reuben Oliver in the drawing room, Tommy Shelby is visible in the crowd, engaged in conversation with Arthur. This is supported by the script, which mentions: “We drift on as we glimpse Tommy in conversation with Arthur through the crowd.” The episode summary also states that during the mingling in the drawing room where Reuben speaks to Polly, “…we glimpse Tommy in conversation with Arthur.” Later in the scene, Finn interrupts Polly and Reuben to tell Polly that Tommy wants her, further suggesting that Tommy was elsewhere in the house.

Notebook even links back to the relevant parts of the script.

Another example is from Constellation (Apple TV+), a show I worked on with a highly complex story timeline. Set in a world where not everything is as it seems, the story jumps back and forth through time and space. One of our key editing challenges was tracking which timeline we were in at any given point.

We developed a system of colour-coding different timelines and aspects of the story, with scene bins marked accordingly. Testing how easy it would be to set up on Notebook—it could easily give quick chronology breakdowns from the scripts and answer queries about what happens next in any particular timeline.

“Previously On” and “Next Time On”

There’s definitely an art to crafting the “Previously On” and “Next Time On” segments that bookend a series. They’re often treated as an afterthought, yet can be complicated and information-dense parts of the edit. You have to condense an episode—or sometimes an entire season—into a tight, compelling trailer, while preserving the nuance and detail that producers and directors want conveyed.

Quite a bit of the editing process for previously on and next time segments are finding good, short, examples of lines that can help explain in a very short period of time something which has already, or will happen. Notebook is great at searching these out for you from the script.

Here’s an example of how Notebook LM can help. I asked it:

Find me five quotes that show Jo misses her daughter.

Notebook LM’s response:

1. Scene 219A, JO: “Hello. Hello Alice.”
   – The opening of a voice memo Jo records, showing Alice is immediately on her mind.

2. Scene 237, JO: “Hi, darling. It’s me... It’s Mamma.”
   – A tender message from Jo to Alice, full of maternal affection.

3. Scene 239, JO: “Alice – I didn’t leave you. You were always the most important thing in the world to me.”
   – A powerful statement that directly conveys Jo’s deep love and longing.

4. Scene 237, JO: “I’m leaving the iPad here, in case you need to hear this at some point.”
   – Not a direct statement of missing Alice, but a touching gesture showing care and foresight.

5. Scene 219A, JO: “It’s nice to hear a voice, even if it’s just my own.”
   – This line, following Jo’s message to Alice and Magnus, hints at her loneliness and longing for their voices.

Notebook LM isn’t perfect. In all cases above you have to apply a certain amount of human oversight to get it right, a lot of its descriptions are weird or unbalanced, it misses things and the podcast can often miss the point of the thing its exploring (in the way that real human podcast on the subject wouldn’t)—but for the first time, I’ve found a tool that genuinely supports the work I do on a daily basis. From absorbing scripts and tracking character arcs to organising timelines and extracting emotional beats, it’s become a valuable assistant in the editing suite.