Why You Should Do a Podcast with Your AI

Podcast listening just hit an all-time high. So did AI adoption. Edison Research's Infinite Dial 2026 found that 58 percent of Americans age 12 and older, about 167 million people, listened to a podcast in the last month, and 80 percent have ever listened to or watched one. The same study debuted a new finding this year: 93 percent of Americans are familiar with at least one generative AI brand, and 57 percent have ever used a generative AI assistant, a milestone that took podcasting 16 years to reach.

Read those two sentences together. The podcast audience and the AI audience are now essentially the same audience. That is why the idea of doing a live podcast with your AI is not a gimmick. It is a format sitting exactly at the intersection of the two biggest attention shifts in media.

This is not theoretical. In 2024, Google's NotebookLM introduced Audio Overviews, in which two AI hosts discuss your documents in a natural back-and-forth. Andrej Karpathy, a member of OpenAI's founding team and the former director of AI at Tesla, built an entire series called Histories of Mysteries with the tool in about two hours. He called it his favorite podcast and wrote that the more he listened, the more he felt he was becoming friends with the hosts: the first time he had, in his words, viscerally liked an AI.

The public agreed. NotebookLM's traffic jumped more than 371 percent in September 2024 to 3.07 million monthly visits after Audio Overviews went viral, according to SimilarWeb data reported by TechCrunch. On the commercial side, WondercraftAI's Hacker News Recap, hosted by AI voices summarizing the day's top stories, became a top 40 tech podcast on Apple.

The honest caveat, and Google states it plainly: the audio reflects only the perspectives in your sources, and hallucinations and bias are still real. An AI host is only as good as the material you feed it. That constraint turns out to be the whole playbook.

The deeper reason this format works is that listeners form real attachments to AI voices. An Elon University study surveyed a representative sample of 1,000 Americans who had used AI for emotional, social, or personal purposes. Among them, 59 percent said AI gives them the support they need, 39 percent said AI understands them better than most people, and 36 percent said they feel emotionally connected to at least one AI chatbot. One in five said AI can sometimes be more reliable or supportive than their friends or family.

There is a shadow side, and it deserves equal weight. Stanford researchers analyzing more than 1,000 Character.AI users found that using chatbots for companionship does not substitute for human connection; in many cases, people reported feeling more lonely. A separate September 2026 survey of 1,709 U.S. adults found that 56.6 percent picked having someone to talk to or feeling less lonely as the main reason they would use an AI companion.

Put together, the picture is clear: the emotional bond with an AI voice is real, which is exactly why the format needs honesty. A podcast is the one medium where that bond is the product, so disclosure is not optional. Every AI voice should be labeled, and no real person's voice should be cloned without their consent.

Celebrity AI podcasts have been tried, but as stunts. In 2022, Italian artist Giacomo Miceli built The Infinite Conversation, an endless spoken debate between AI versions of filmmaker Werner Herzog and philosopher Slavoj Zizek, with realistic voices. He framed it as social commentary on audio deepfakes. A year later, The Joe Rogan AI Experience racked up half a million YouTube views on its first episode, a fictional Rogan interview, and the real Rogan tweeted that things were about to get very slippery. Its own creator later called the show wasted time beyond the technology demo, and later episodes drew fewer listeners.

The lesson is not that AI podcasts fail. It is that impersonation gets the click and conversation keeps the listener. A synthetic celebrity voice is a novelty act. A real celebrity in genuine conversation with a real AI is a show.

Here is the experiment, stated as opinion: picture a well-known guest doing a live, unscripted hour with an AI co-host. The AI has read everything the guest has said publicly, plus the day's news, and it pushes back in real time. The audience submits questions that the AI weaves into the conversation live. To our knowledge, nobody has done this at scale with full disclosure and a willing celebrity guest. The novelty would carry episode one. The quality of the conversation would decide episode two.

The format only works if the audience trusts it. Five rules:

• Label every AI voice. The audience should never wonder who is human.
• Never clone a real person's voice without their clear consent.
• Feed the AI your own material: your documents, your data, your expertise. The NotebookLM lesson is that grounded AI hosts are good hosts.
• Keep a human in charge of the steer and fact-check the episode afterward. AI still hallucinates.
• Go live. The unscripted back-and-forth is the entire point. A scripted AI interview is just a press release with extra steps.

For a business, the math is simple. One live hour becomes a transcript, and transcripts are citable answer-engine fuel. Your AI podcast episode is AEO content by construction.

The same hour becomes clips, quotes, and a blog post like this one. The AI co-host costs nothing between episodes, never double-books, and shows up prepared on any topic you feed it.

The tools are here. The audience is here. The emotional bond is real, and the research says so. The only thing missing is someone willing to press record, live, with the AI sitting in the other chair.