My podcast is spoken by my cloned voice. This is how every episode gets made, week after week. Three tools are enough. You still check the script yourself, and that is what makes the difference.
Last updated:
Click a step to jump to its explanation below.
I give ChatGPT the topic and the source material, often a long English research paper, and have it write a script. Tell it the text will be read aloud. What looks fine on paper often sounds stiff when spoken.
This is the prompt I start with. Replace the pink parts with your own details.
Write a podcast script of about [LENGTH] minutes on [TOPIC], based on the attached material. The text will be read aloud by a voice. So: - short sentences and spoken language - everyday comparisons instead of jargon - no lists, no headings in the text - talk to the listener directly as "you" This is how I talk: [two or three sentences on how you sound, or a sample of your own writing] Flag every place where you are not sure about a number or a fact.
Skip this step and you will hear it in the episode. I read every script myself, ideally out loud. If a sentence doesn't sound like me, I rewrite it. Only then does it go to the voice.
Language models like to invent numbers and sources that sound plausible. I look up everything stated as fact. Whatever I can't back up gets cut or rephrased as an opinion.
In ElevenLabs you create a voice clone from recordings of your own voice. The cleaner the recordings, the better it sounds. Record in a quiet room and speak at your normal pace.
Cloning needs a paid ElevenLabs plan. The free plan doesn't include it (as of November 2026).
Paste the finished script into ElevenLabs and have it read in your cloned voice. Add pauses and regenerate single sentences until the emphasis is right. At the end you download the episode as an MP3. Jingles and background music can be generated in ElevenLabs too, so everything comes from one tool.
Mispronounced names and abbreviations only show up when you listen. Write them in the script the way they should be spoken, especially foreign technical terms.
I designed my cover with ChatGPT. On Spotify it often shows up thumbnail-sized, so keep the text short and large. The image has to be square. Make it big, 3000 × 3000 pixels is a safe size. A ChatGPT image may need upscaling for that.
Check every letter on the cover. Image generators like to twist or drop one, and a typo in your podcast name shows up again with every episode.
Since 2 August 2026, Article 50 of the EU AI Act applies. If you use AI to create a voice that sounds deceptively like a real person, you have to disclose that it is artificial. A cloned version of your own voice counts. That's why I disclose it in several places, so nobody can miss it.
Every episode opens with a note that the voice is AI-cloned. On Spotify it sits right under the podcast name: „KI geklonte Stimme von Olga Evangelou“, German for AI-cloned voice of Olga Evangelou. There is also a note in the description and on my website.
This is my practical take, not legal advice.
One sentence at the start of the episode is enough: “Quick note: the voice you're hearing is mine, but cloned with AI.”
In the Spotify for Creators app you set up your podcast with a name, description and cover. Then you upload your episodes as MP3s. Each new episode simply gets added.
Pick a fixed day and stick to it. A new episode of mine comes out every Sunday, so regular listeners know when to check.
Hear the result →The o.KI Podcast (in German), a new episode every SundayHave a language model such as ChatGPT write the script and check it yourself. Then clone your voice in ElevenLabs, generate the episode there as an MP3, design a cover and publish the podcast through Spotify for Creators. Disclose the AI voice in the episode and in the description.
For my podcast, three are enough. ChatGPT writes the script and designs the cover. ElevenLabs produces the voice, the episode and the music. Publishing runs through the Spotify for Creators app.
Yes, as soon as the voice imitates a real person, which includes your own cloned voice. Article 50 of the EU AI Act has applied to this since 2 August 2026. Each of my episodes opens with a note, and on Spotify the disclosure sits right under the podcast name.
Only with their explicit consent. Without it, you will usually infringe their personality rights. The simplest option is your own voice.
It can write a good draft. The script isn't finished at that point. Language models like to invent plausible-sounding numbers and write in a tone that isn't yours. I read every script out loud and check the facts before it gets voiced.
With the Spotify for Creators app. You set up the podcast with a name, description and cover, then upload each episode as an MP3.
Whether it's just you or your company: I'll show you how to set up a podcast like this with AI, from script to cloned voice. Or I'll produce it entirely for you.
Book a call