Is this human or does it just sound human? – Listening closely to the AI in podcasts

Recently, I came across a podcast episode during my research on the use of AI in podcasting. The podcast’s description mentioned that the host had uploaded a sermon they’d preached, and Google’s NotebookLM had generated a podcast episode based on it. Intrigued, I clicked play, and it started like any other podcast I’ve listened to. I was waiting for a segue with some music introducing the show and the AI part to kick in. I was expecting a noticeable robotic voice, a flat tone, and anything to differentiate the human portion. But, it didn’t – and that’s when it hit me; it was AI from the beginning.

[The podcast episode] wasn’t merely mimicking human speech; it was mimicking human imperfection, and it did it well enough that I didn’t notice it until I consciously searched for it

I was surprised by how uncannily human the voices were, with intonations and pauses before continuing to speak. After a while, I started wondering if every ‘um’ was deliberately programmed? If every overlap in dialogue was intentionally designed? If every repetition of a phrase was specifically added in? The experience was fascinating but unsettling. It wasn’t merely mimicking human speech; it was mimicking human imperfection, and it did it so well that I didn’t notice it until I consciously searched for it. If you’re interested, the podcast episode in question is by AI and the Gospel titled ‘Google NotebookLM Deep Dive Podcast Demonstration’.

AI has permeated virtually every industry, ranging from healthcare and finance to creative sectors like writing, designing and for a while now, podcasting. The use of AI is increasing every day, and it has reached a point where many wonder how they lived and worked in a world without AI. My research led me down a rabbit hole of the AI podcasting tools on offer right now, like Google’s NotebookLM, Podcastle, Descript, Adobe Podcast, and Spotify’s Voice Translation, which are already reshaping the landscape. Podcasters are using AI for voice generation, editing, and scriptwriting, and an Adamfard blog mentioned the use of AI for automated transcription services, SEO enhancement, and repurposing content. Descript offers podcasters tools to create a digital clone of their voices, which they can then use to generate scripts. Spotify’s Voice Translation allows listeners to listen to their favourite shows in multiple languages, in the voice of the podcaster. These tools increase speed and accessibility, but they also prompt a deeper question: when the voice sounds human, but is not human, what happens to authenticity?

But not all tools are made to support human voices; some are replacing them altogether

In his article on AI in Podcasting, Romain Chauvet described how Laura Sibony, a French writer and teacher at HEC who specialised in AI, used Google’s NotebookLM as her online research assistant. Faced with weekly readings from multiple authors for her show, she relied on the tool to help her analyse because she “wouldn’t have been able to read everything”. In this instance, her use of AI didn’t replace her voice; it just enhanced her show by increasing efficiency. But not all tools are made to support human voices; some are replacing them altogether.

Chauvet also mentions that  Google’s NotebookLM creates audio that “mimics the speaking cadence of podcasters,” and AI software doesn’t stop there. It also turns out that entire podcast episodes can be researched, written, voiced, and edited by AI, and PodGenAI does exactly that. PodgenAI’s Spotify description reads that it “generates an informational single-speaker episode on any topic known to OpenAI GPT-4. Both the text and the audio of the episodes are AI-generated, and inaccurate or unintended content may exist.” A remarkable feat of automation, yes, but it is also a reminder of its limits.

As these tools become more sophisticated and widespread, there is an urgent need for ethical oversight to prioritise truth, creativity and human connection. Overuse and overdependency on AI risks eroding trust and originality in podcasts as a platform

Pablo Sanguinetti, an AI and Humanities research and adjunct professor at the Spanish IE University, mentions that studies have shown people prefer AI-generated content over human-generated content as long as they don’t know the source. He explains, “As soon as they find out that it is a machine and not a person, they tend to prefer the latter.” That discomfort is deeply human, because when there is no human behind the mic, what and who are we listening to?

AI has its place: using synthetic voices for translation, accessibility, and supporting production workflow. As these tools become more sophisticated and widespread, there is an urgent need for ethical oversight to prioritise truth, creativity, and human connection. Overuse and over-dependency on AI risks eroding trust and originality in podcasts as a platform. When trust is on the line, one question will go off in everyone’s mind: Is this human, or does this just sound human?