Over the past few days, I’ve been exploring a new, experimental AI tool from Google, NotebookLM, which promises something truly remarkable: high-quality, human-sounding conversations generated entirely by AI. Everyone would agree that AI has changed content creation. Hack content creators turn out piles of AI-generated drivel. Skilled writers use generative AI tools for ideation, research, and drafting. But, original high-quality audio and video content have been elusive. Today, we may be closer to AI-generated audio that rivals human production than you think.
NotebookLM’s Accuracy Advantage
NotebookLM is a RAG-locked model—meaning it only uses the content you upload to it. This locked approach reduces hallucinations and errors compared to broader AI models like ChatGPT or Claude that pull from vast data sources. When I uploaded PDFs of my books Friction and Brainfluence to this model, it produced a surprisingly good comparison of their content and style. It also produced accurate quotes.
But the real surprise came when I tried a brand-new feature: near-instant “audio overviews.”
Audio Overviews From Any Content
Google recently added a button in NotebookLM to create an audio conversation based on the content uploaded to the system. Imagine this: a podcast-like discussion generated from a few articles or even a full book. It’s way more than just reading aloud—it creates a natural dialogue between two AI voices.
In my experiment, I fed Friction – a nearly 300-page book – into the system and got a 15-minute audio dialogue on the book’s core ideas. The result was shockingly human-like. The AI speakers didn’t just recite text; they conversed, interrupted each other naturally, and even added the occasional “um” for realism. If you didn’t know it was AI, you’d likely believe you were listening to two experts having a real discussion.
Turning Audio Into Video
I took it a step further by pairing this AI-generated audio with avatars using another tool, HeyGen. The final product was a video conversation featuring two digital personas that synced with the AI dialogue. While it wasn’t flawless—the avatars’ movements occasionally gave away the AI origins—it was far better than a lot of poorly-done videos populating platforms like YouTube. The gap between AI-generated content and human-created content is quickly narrowing. Here’s a short video that I created that includes a 90-second sample of the video conversation:
Meta: NotebookLM AI People Discuss Notebook LM AI People
Good news for the human race: Google’s AI “people” aren’t yet self aware. I wrote a piece yesterday at Forbes about NotebookLM. Then, I gave the link to NotebookLM and clicked the Generate button. It created a five minute discussion of the piece in which the two speakers were blissfully unaware they were discussing… themselves! (Or, if you have a sinister imagination, they WERE aware but didn’t want to reveal their sentience to humanity.) Here’s the audio:
What Does NotebookLM Mean for Content Creation?
Right now, NotebookLM audio oveview function is limited—there are no settings for choosing voices, controlling the length of the conversation, choosing topics to highlight, selecting languages or accents, etc. Since this is an experiment, Google might decide not to improve it or even make it disappear. But if Google expands the tool’s capabilities, the potential would be enormous. Imagine generating entire podcasts, video interviews, or educational dialogues without needing a team of writers, producers, or voice actors.
AI Content Creation: Helpful or Creepy?
This development raises some intriguing questions. If AI can generate convincing human-sounding conversations from any text, what does that mean for industries like radio, podcasting, or even television punditry? Will audiences eventually prefer AI-generated content that’s customized for them over real human voices? AI tools like HeyGen and ElevenLabs can already create audio from text in any voice, including yours. A few authors are using this technology for their audiobooks. No need to sit in a studio for days. No bumbled text that needs to be re-done. No lip smacks or stomach rumbles. (Yes, ask audiobook producers – stomach rumbles are a thing.)
At present, we’re not quite at the point of replacing radio hosts or podcast creators with AI, but we are certainly approaching a new frontier in audio and video technology.
Where Could You Use NotebookLM?
Even with the current state of NotebookLM’s audio summary tool, imagine the possibilities: AI-driven audio summaries of meetings, dense research papers turned into digestible discussions, engaging book summaries… Whether this strikes you as inspiring, a little unsettling, or somewhere in between, one thing is clear—AI’s ability to create realistic, engaging audio and video content is only going to grow. Now that Google has shown the way, others are sure to follow.
