To celebrate its 50th anniversary, Microsoft hosted a special Copilot event where several new AI features for the Copilot chatbot were announced. Among them, Copilot Podcasts was the feature that caught my attention as I had tried out NotebookLM sometime ago, and it had impressed me with its AI-generated podcast feature. So, it was time to put Microsoft’s Copilot Podcasts feature to the test and see just how realistic the AI-generated podcast sounds.
Thankfully, Copilot Podcasts is already available for users to try right now, which can’t be said for other announced features at the event.
To get started, you can open the Copilot app on Windows, macOS, Android, or iOS, and enter the prompt “create a podcast on…” and Copilot will generate a 4-minute AI podcast within a minute. You can also generate a podcast on Microsoft Copilot’s website (visit).
Once the podcast is ready, you will get a notification on your device, or you can open Copilot’s homepage, where the AI-generated podcast will appear. Now, you can listen to the AI-generated Copilot Podcast. Note that there are no playback speed controls, and you can’t download the AI-generated podcast from Copilot’s interface.
Copilot Podcasts Put to the Test
In my testing, I asked Copilot to create a podcast discussing China’s emergence in AI. Since I cover all things AI at Beebom, I thought it would be a fitting topic to test Copilot’s podcast capabilities. First and foremost, unlike NotebookLM, which sounds very conversational and engrossing, Copilot Podcasts sound more like a text-to-speech conversion.
The Copilot AI voice has a robotic and synthesized tone, lacking natural pauses. There are no expressive cues like “umms” to make the podcast feel truly human-like. Even the interaction between the two AI hosts feels monotonous, at best. There are no interruptions or cross-talk, and sounds like a scripted daily briefing of events.
Sure, the voices do show some emphasis and shifts in tonality, but that’s all the realism you get on Copilot Podcasts. As for conversation depth, well, Copilot Podcasts only touched upon DeepSeek R1‘s development and didn’t go deeper at all. The AI podcast felt like word salad — overly verbose, lengthy monologue, and information density was next to nothing.
In comparison, you can listen to Google’s NotebookLM podcast, which is 20 minutes long (I have attached a shorter audio clip below), and it does a far better job. I took advantage of NotebookLM’s new Discover Sources feature to create a knowledge base on China’s emergence in AI and then generated the podcast.
As you can hear, the podcast sounds pretty natural, and there is a lot of cross-talk between the AI hosts. You can also notice expressive cues and natural pauses. In addition, NotebookLM now lets you interact with the AI hosts, which is awesome for real-time engagement. Not to mention, you can adjust the playback speed and download the audio clip as well.
All in all, Copilot Podcasts feels like a half-baked product in its current form. It lacks realism in voice tonality and is not optimized for natural, free-flowing conversations. Microsoft needs to improve Copilot Podcasts across the board by adding insightful information and enhancing the audio tonality for a more human-like experience.
