More than 619 million people will listen to podcasts globally in 2026. The market is on track to reach $17.59 billion by 2030. And right now, inside marketing teams at companies that care about content, there is a version of the same conversation happening: Can we use AI to do this faster? Should we?
The question isn't cynical. It's rational. AI is generating text, video, voice, and visuals at a scale and speed that would have seemed absurd three years ago. Nearly two-thirds of podcasters have already used generative AI in some part of their production, according to a survey by Descript, and 78% say they plan to continue. When your CFO sees those numbers, your podcast budget starts to look like a target.
So JAR decided to stop theorizing and run the experiment.
What Actually Happened When We Tested It
The setup was simple, but the work was serious. JAR formed an internal RED Team and immersed them in the contemporary AI toolkit: ChatGPT, DALL·E 2, Midjourney, Descript Voice Cloning, DeCIFR, Adobe Podcast Speech Enhancer, RunwayML, Melville, ElevenLabs.io, and Vidyo, among others. The team wasn't skimming — they were using these tools as deeply and honestly as possible to understand what they could and couldn't do.
From that research, the team produced two finished podcasts. The first was made entirely by humans, with full creative freedom. The second was built largely by AI tools, with minimum human intervention. Both were then played for a panel of unsuspecting listeners who had no idea which was which.
The results weren't close.
Audiences preferred the human-made podcast. Not slightly — meaningfully. Listeners found it more inspiring, more engaging, and more positive for the brand associated with it. The AI-generated version had real problems: structural coherence suffered, the emotional pacing felt off, and there was something underneath it all that audiences sensed even if they couldn't name it — an absence of genuine intent. The show had been generated, not made.
For Roger Nairn, CEO and co-founder of JAR, the results were a relief, but also a confirmation. In his own words: "I confess I had my doubts about whether others would feel the same" about human creativity mattering. They did.
The five takeaways from JAR's experiment are worth sitting with:
Human creativity still resonates differently. AI-made podcasts didn't connect with audiences the same way. The emotional and inspirational dimension was weaker — and listeners felt it without being told to look for it.
AI tools require more human intervention than advertised. Even with significant effort applied to the AI-built show, it needed restructuring and creative direction before it was listenable. The efficiency gains don't materialize if you're spending that saved time fixing structural problems downstream.
AI struggles with pacing and coherence at the show level. Sentence-level fluency is not the hard part. Knowing how a conversation should breathe, when to slow down, where the emotional peak of a story lives — that's where AI falls apart in a 30-minute format.
Voice cloning and synthetic audio carry real ethical weight. Authenticity isn't just a creative value. It's a trust variable. Audiences who discover they've been listening to a cloned voice don't extend grace.
AI's actual role in podcasting needs to be honestly calibrated. It has a place — but that place is not at the centre of the creative process.
The Question Brands Are Actually Asking
The noise around AI has shifted what should be a strategic conversation into an anxious one. Brands watching the tool landscape evolve aren't asking "Is AI magic?" — they're asking something more specific: "Are we leaving efficiency on the table by not using this more aggressively?"
That's a legitimate question. AI-driven podcast production tools are expected to increase production efficiency by over 40%, and podcasters using advanced AI workflows report production cost reductions around 30%. Those aren't numbers to dismiss. For a marketing team managing a show quarterly with a stretched content operation, that efficiency gap is real.
But there's a difference between using AI to eliminate the low-value time in your workflow and using AI to generate the show itself. Most of the hype conflates these two things. The first is smart. The second is where you lose the thing your audience is actually showing up for.
The strongest branded podcasts — the kind that build real audience relationships and deliver measurable business outcomes — are built on editorial point of view, genuine conversation, and an authentic brand voice. Those aren't inputs AI can provide. They're the outputs of human thinking, experience, and creative judgment. AI can support the infrastructure around those outputs. It cannot replace the source.
The brands that will pull ahead in 2026 aren't the ones that automate the most. They're the ones that understand which parts of their podcast are worth protecting from automation — and which parts are just logistics.
Where AI Actually Belongs in a Serious Production Workflow
Here's the position that the experiment, and honest daily use, supports: AI is a production accelerant. It belongs in the parts of your workflow that require consistency, repetition, and speed — not the parts that require judgment, taste, and voice.
Three places where AI earns its seat at the table:
Clip identification and social teaser generation. Tools like Descript's auto-highlights can pull compelling moments from a longer episode in minutes. This used to require a skilled editor watching an hour of footage to find three 60-second clips. AI gets you to a shortlist faster, and a human makes the final call. That combination — AI speed, human selection — is the right model. A well-structured episode dramatically increases the quality of what AI surfaces here; if you're not already thinking architecturally about how your episodes generate clips and secondary content, the companion post on how to structure podcast episodes that generate clips, posts, and sales content is worth reading before you set up your workflow.
Automated transcription as a starting point. Clean, accurate transcripts are both an accessibility baseline and an SEO asset. AI transcription tools get you roughly 90% of the way there — faster and cheaper than manual typing, accurate enough to be useful, and fixable by a human editor in a fraction of the time a full manual transcription would take. The critical word is starting point. Publishing raw AI transcripts without review is how you introduce errors into your content infrastructure at scale.
Content repurposing from transcript. Feed a clean transcript into a language model and you have the raw material for episode summaries, newsletter excerpts, blog drafts, and social copy — all grounded in what was actually said, not invented. This is where AI provides genuine leverage: the ideas, stories, and insights already exist. AI helps restructure them for different formats and contexts. But the editorial voice has to come from a human before anything goes out. Turning one episode into 20+ content assets only works when the source material is strong and the repurposing is curated, not automated end-to-end.
The pattern across all three is the same: AI removes friction from the structural work so that human attention can go where it actually matters. This is creative leverage, not creative replacement.
Why "AI-Assisted" and "AI-Generated" Are Not the Same Thing
The failure mode isn't using AI. It's publishing AI output without understanding what you've lost in the process.
A podcast episode generated primarily by AI — scripted by a language model, voiced by a synthetic host, edited by an automated pipeline — can be technically competent. It can hit the right duration, pass quality thresholds for audio, and cover a topic accurately. What it cannot do is make a listener feel understood, surprise them with a genuine insight, or communicate the kind of specific brand intelligence that comes from years in an industry.
That gap matters more in podcasting than in almost any other content format. Audio is intimate. Listeners are alone with the host, often during commutes, workouts, or tasks that create a low-distraction state where authenticity is harder to fake. The research backs this up: when AI tools lead the thought leadership and storytelling, shows become generic — and generic podcasts don't grow. They don't build trust. And for a branded podcast with a business job to do, that's the only metric that actually matters.
The post-production time savings are real. Podcasters using advanced AI workflows have reported per-episode post-production time reduced by 60-70%. That's a meaningful operational gain. But it only translates to a better show if the time saved goes back into the creative work — better guest research, tighter editorial direction, more thoughtful distribution strategy — rather than into producing more volume of content that doesn't connect.
More episodes made faster is not a strategy. A better show, made with discipline, is.
The Position Worth Defending
JAR's core philosophy has always been that a podcast is for the audience, not the algorithm. AI doesn't change that. What AI does is give teams more time to focus on what the audience is actually there for — real perspective, genuine conversation, stories worth following.
The brands doing this well in 2026 are the ones that have figured out where the human investment belongs: in editorial strategy, in host preparation, in story selection, in understanding what the audience actually cares about. They're using AI to handle the scaffolding so that the creative work doesn't get crushed by production overhead.
The experiment settled the question of whether AI can replace that creative work. It can't. What's worth asking now is whether your current production workflow gives human judgment enough room to do what it's actually good at — or whether your team is drowning in logistics that AI could handle instead.
That's the right version of the question. And the answer to it is probably worth more than any AI tool you could adopt.