How I Created a Full, Co-hosted Podcast in Just 4 Hours with NotebookLM

and why it matters for creators

Last week, I read an article by friend and Canadian podcaster, Al Grego, titled “I Think… I’m Done”. In it, Al shared his frustration / admiration / fear / thoughts around Google’s NotebookLM, an AI tool from the giants of the web. This was inspired by an example from James Cridland, founder of leading industry publication Podnews.

Now, primarily, NotebookLM is a note taking and research assistant powered by AI. In Google’s own words:

However, while it’s aimed at research and maybe academia, there’s also a component that should make podcasters and creators sit up and take notice (as it did Al), and that’s their Audio Overview feature.

Because this is where NotebookLM moves from a note taking tool into a podcast creation tool. Essentially, NotebookLM takes a look at the “source” you’ve provided – which can be a web page, a document, a PDF upload, or even copy pasted text – and then creates an audio file where two people talk about the source in question, but with their own viewpoints.

And this is where things get interesting, since those two “people” aren’t real – they’re AI generated voices that have been created by Google to talk about the content you’ve just uploaded. But listening to the results, you’d be forgiven if you did actually think they were real, that’s how good the results are.

After experimenting with it for a bit, I was curious how well NotebookLM would cope if I tried to create a brand new co-hosted podcast, and how long it’d take from ideation to completion. The answer on both fronts was super impressive.

AI is No Longer Super Easy to Spot

As I mentioned earlier, NotebookLM takes a source and then creates notes around that. However, if you use the Audio Overview feature, it creates a conversation between two AI-powered “co-hosts”, if using podcasting vernacular. And it was this feature I was really curious about.

So I came up with an idea – let’s create a brand new co-hosted podcast, using this publication as “the source”, and see what happens. So I looked at some of my most popular articles, and used those as the starting point for the podcast, with each article being a singular episode of this fledgling podcast.

And it was equally scary and impressive at how well the result came out. But before looking at that, here’s the flow on how each episode was created.

• find an article on my publication

• copy the text part into Google Docs

• create a new “notebook”

• add sources (Google Drive, a link, or copy paste text) – I used Google Drive/Docs

• click “insert source”

Once done, NotebookLM will display your “Notebook Guide”, which gives you an overview of the note you’re creating, as well as what you’d like to create from it:

• FAQ

• study guide

• table of contents

• timeline

• briefing doc

Now, while all that is fun and relevant to someone needing notes and study guides, etc, as a podcaster you want the audio conversation. So you’ll click the Generate button on the Audio Interview option:

After clicking that, you’ll have a short wait until the voices and conversation have been generated. And this is where it gets really good.

At first, I wasn’t really sure what to expect, since most AI conversations and attempts at replicating the human voice that I’ve tried, and all the inflections, tone, and emotion that contains, have fallen flat. Not so with NotebookLM.

Listening to the result, this actually sounds like two people talking to each other. Sure, if you’re an audio person and listen closely enough, there are minor tell-tale signs – occasional breaths between talking that don’t sound “quite human”, and the odd word that is missing as the AI co-hosts look to fill space.

But overall, it’s really damn good. Here’s an example – to create the trailer for the show, I prompted NotebookLM with this:

And here’s the result: