Google's new AI tool for podcasts has revolutionized the way writers share their content

I’m a prolific writer, aiming to produce over three articles each week. This practice has greatly enhanced my communication skills, improved my technical design documentation at work, and allowed me to share the many ideas swirling in my mind.

Until recently, I struggled to repurpose my articles effectively. I tried various text-to-video tools, but they all fell short.

Google’s new NotebookLM has completely changed the game for writers like me, generating engaging and realistic podcasts between two voices. Honestly, I’d listen to them for pleasure, and they don’t even sound AI-generated!

I then use Headliner to turn the audio into videos, which I share on platforms like YouTube and TikTok.

People might be skeptical, but I tried it out for fun last night, and I’m definitely going to use this as a way to learn about topics I find dull. It’s much more engaging than just standard text-to-speech conversion—I can actually listen and learn while I work!

It’s quite good and a genuinely impressive accomplishment from Google.

However, it does have some significant shortcomings. For instance, it often misses the main point of its source material, getting bogged down in irrelevant details and using hyperbole in ways that can feel surreal or inappropriate. As we listen more, the realism diminishes because we start to pick up on patterns and clichés. If they could introduce a greater variety of virtual hosts and allow us to customize their personalities or pair them up, I believe we’d have a truly game-changing tool.

Alright, this is quite clever. I fed it a bank statement since it was the only thing I had on hand, and it generated a podcast that broke down the payments and offered insights.

While it does come across as a bit robotic at times, the overall conversation felt pretty natural, complete with filler words, laughter, and casual dialogue.

It’s great, but it’s not quite production-ready. I’ve generated over 50 “podcasts” using this tool, which employs Gemini 1.5 for the transcript. Unfortunately, it experiences significant hallucinations, particularly with medium to long-form source content. Just a heads-up: it can be quite inaccurate, and there’s no way to correct those errors.

This is a game changer! I just conducted an interview with one of our alumni, and the summary it generated from the actual transcript—complete with my not-so-great questions—turned it into something that feels like a Rich Roll-style podcast.