Making a podcast with AI comes down to three choices: what you feed in (a document, a script, or an idea), what you want out (an audio show or a short video clip), and how much of the middle you want to control. Pick one of the three lanes below, and you can go from nothing to a published podcast today, for somewhere between free and about $30 a month.
This guide walks through all three lanes step by step, with honest costs and tradeoffs, so you can pick the one that matches what you're actually trying to do.
TL;DR:
- Lane 1: document to audio. Tools like NotebookLM turn a PDF or article into a two-voice audio discussion for free. Fastest start, least control.
- Lane 2: DIY stack. An LLM writes the script, a text-to-speech tool voices it, an editor assembles it. Most control, most moving parts, roughly $5 to $50 a month.
- Lane 3: video podcast clips. A script plus an AI host becomes a short, captioned podcast reel for TikTok, Reels, and Shorts. Built for feeds, not RSS.
- Podcasting is worth the effort: 58% of Americans 12+ now listen monthly, a record high (Edison Research, Infinite Dial 2026).
What does "making a podcast with AI" actually mean?
An AI podcast workflow replaces recording and editing with generation: you supply the idea or the source material, and AI models produce the script, the voice, and optionally the on-screen host. That's the definition worth keeping. The microphone, the studio, and the edit timeline stop being requirements and become options.
In practice, people mean one of three different things when they say they want to make a podcast with AI:
- Turn existing material into audio. You have PDFs, articles, or notes, and you want a listenable discussion generated from them.
- Produce a traditional show faster. You still want an audio podcast with episodes on Spotify and Apple Podcasts, but you want AI to write, voice, or edit it.
- Make podcast-style video clips. You want short, feed-native clips of a host talking to camera, without filming anyone.
These are genuinely different products with different tools, so the first step is deciding which one you're making. If you're still fuzzy on the third category, what an AI podcast is covers it in plain terms.
Which lane should you pick?
Pick by output, not by tool. If your audience listens on podcast apps, you need lane 1 or 2. If your audience scrolls TikTok, Instagram, or YouTube Shorts, you need lane 3. Nothing stops you from running two lanes off the same source material later, but start with one.
| Lane 1: document to audio | Lane 2: DIY stack | Lane 3: video clips | |
|---|---|---|---|
| Input | PDFs, articles, notes | Your topic or outline | A script (written or AI-drafted) |
| Output | Audio discussion, 2 AI voices | Full audio episodes | 15-90s captioned podcast reels |
| Where it lives | Private or shared links | Spotify, Apple, RSS | TikTok, Reels, Shorts, LinkedIn |
| Control | Low | High | Medium-high |
| Typical cost | Free | ~$5-50/mo across tools | $1 first clip, plans from $29/mo |
| Time to first output | ~10 minutes | A few hours | Minutes |
Worth knowing before you pick: video is where podcast consumption is heading. YouTube is now the most-used podcast service in the US, chosen by 33% of weekly podcast listeners, ahead of Spotify and Apple Podcasts (Edison Research). Audiences increasingly expect a face on screen, which is exactly what lanes 1 and 2 don't give you.
How do you turn documents into an audio podcast with AI?
Upload your source material to a generator like Google's NotebookLM, click its audio overview feature, and wait a few minutes. The tool reads the material, pulls out the key ideas, and produces a conversation between two AI voices discussing it. You get a listenable summary without writing a word.
The steps, concretely:
- Open NotebookLM and create a new notebook.
- Add sources: upload PDFs, paste URLs, or drop in text.
- Generate the audio overview. A long document takes several minutes.
- Download the audio, or share the link.
This lane is free and genuinely impressive for what it is. Its limits are equally real: you don't control the script, the voices are the same two hosts everyone else has, the output isn't branded, and you can't publish it as a growing show without extra work. It's a consumption tool first and a production tool second.
If your starting material is documents and you want to go further than a private summary, we compared five document-to-podcast routes in PDF to podcast: 5 real ways to convert with AI, including where each one breaks down.
How do you build a DIY AI podcast stack?
Chain three tools: a language model for the script, a text-to-speech service for the voice, and an audio editor for assembly. This is the lane for people who want a real episodic show on podcast apps but don't want to spend evenings recording and cutting audio.
The stack, step by step:
- Script with an LLM. Give Claude or ChatGPT your outline, audience, and target length, and iterate until the draft sounds like you. Feed it transcripts of things you've actually said; it makes the output noticeably less generic. Our full walkthrough on how to write a podcast script with AI covers the prompt and the edit passes.
- Voice with TTS. ElevenLabs is the usual choice; paid plans start around $5 a month, and you can clone your own voice on the mid tiers. Read the script yourself instead if you want zero synthetic feel, the AI still saved you the writing time.
- Assemble and clean up. Descript or a similar editor lets you edit audio like a text document: delete a sentence in the transcript and it's gone from the audio.
- Host and distribute. A podcast host (Buzzsprout, Transistor, and similar run roughly $10-20 a month) gives you the RSS feed that Spotify and Apple Podcasts pull from.
Total damage: roughly $5 to $50 a month depending on tiers, plus a few hours per episode instead of a full production day. The tradeoff is operational: you own every step, which also means every step is yours to babysit.
How do you make a video podcast with AI?
Write a short script, pick an AI host, and generate. This is what we build at MakePodcast: you type what you want said, choose a host (a stock one, or a version of yourself built from three photos), and the system generates a vertical, captioned podcast reel of up to 90 seconds. No camera, no studio, no editor.
The full loop:
- Write one beat. A podcast reel carries one idea. If your script has three points, that's three clips.
- Pick the host and language. Same host every clip builds recognition. 70+ languages from the same script if you need them.
- Generate. The voice, the lip-synced host, captions, and branding render into a finished clip in minutes.
- Post. Download and publish to TikTok, Reels, Shorts, or LinkedIn.
We run this lane on our own channels daily: automated news-style shows where AI drafts the script from the day's stories, a clip renders with a consistent host, and a human approves it before it publishes to TikTok. One clip a day, every day, with the human involvement measured in minutes. That cadence would be impossible to sustain with filming, and it's the honest pitch for this lane: it turns consistency from a production problem into a writing problem. The full setup is documented in how to make an AI news channel.
This lane costs $1 for your first clip, then plans from $29 a month. What it doesn't do matters too: it won't give you an hour-long RSS show. It's built for feeds.
How do you publish and grow an AI podcast?
Match distribution to the lane. Audio episodes go to podcast apps through an RSS host; video clips go directly to short-form feeds. Then hold a schedule, because in every lane the algorithmic platforms reward consistency more than polish.
Three things that compound regardless of lane:
- A recognizable constant. Same host, same intro pattern, same visual frame. Recognition is what turns a scroll-past into a follow.
- A sustainable cadence. One good clip a day beats seven on Monday and silence after. AI removes the production excuse; the discipline is still on you.
- Honest labeling. Don't pretend a synthetic host is a live recording. Audiences forgive AI; they don't forgive feeling tricked.
The audience is there and still growing: weekly podcast listening in the US hit 45%, another record in the same Edison study. The catch is that they have more choices than ever, which is an argument for the formats where you can show up daily.
Frequently asked questions
Can ChatGPT make a podcast?
ChatGPT can write your script, outline episodes, and draft show notes, but it doesn't produce audio or video on its own. You pair it with a text-to-speech tool for the voice, or a clip generator for video. It's the writer in the pipeline, not the whole pipeline.
Which AI is best for making podcasts?
Depends on the lane. NotebookLM is the strongest free document-to-audio tool. For a DIY audio stack, ElevenLabs leads on voices and Descript on editing. For short video podcast clips with an on-screen host, that's what MakePodcast does. There is no single best; there's a best per output.
Can I make a podcast with AI for free?
Yes, in lane 1: NotebookLM generates audio discussions from your documents at no cost. The free route ends when you want your own voices, branding, or a published show; that's where the $5-30 a month tools come in.
Do I need a microphone to make an AI podcast?
No. In every lane above, the voice is generated from text, so there's nothing to record (voice cloning needs one short sample, recordable on a phone). We wrote up the no-equipment workflow in how to make a podcast without a microphone.
What's the best AI podcast setup for a beginner?
Start with the lane that matches where your audience already is. If that's podcast apps, begin with lane 1 to learn the format, then graduate to the DIY stack. If it's social feeds, start with a $1 clip in lane 3 and post it. The worst setup is the one that keeps you planning instead of publishing.
If your lane is short video, make your first podcast reel with MakePodcast for $1 and see what your script looks like with a host delivering it.