I Cloned My Voice With AI and Made an 18-Episode Podcast: Here's What Happened

I wanted to know whether AI could remove enough of the work behind podcasting to make a small, useful show practical. So I cloned my own voice, created a short-form podcast for business owners, published 18 episodes, and watched what happened.

There are AI experiments you try for five minutes, say that's interesting, and never think about again.

Then there are the ones that accidentally become real projects.

Mine became a podcast.

I had been experimenting with AI in different parts of my work, and voice cloning caught my attention for a very practical reason.

I liked the idea of podcasting.

I did not particularly like the idea of having another recurring commitment that required me to find a quiet room, get behind a microphone, record an episode, stumble over a sentence, record it again, edit the audio, write everything needed to publish it, and then somehow remember to do the entire thing again next week.

So I started wondering:

Could I create a real podcast using an AI clone of my own voice?

Not a demo.

Not one novelty audio clip that I sent to a friend.

An actual show with a concept, episodes, artwork, descriptions, an RSS feed, and distribution on Spotify and Apple Podcasts.

And, perhaps more importantly:

Would anybody actually listen to it?

That question turned into Bite-Sized Small Business Owner Affirmations, a short podcast built around affirmation themes and encouraging reminders for people running businesses.

The format was deliberately simple. Each episode was generally about four to six minutes long. I used my own licensed AI voice rather than creating fictional hosts, and the episodes focused on specific situations business owners encounter: staying focused, networking, communicating with a team, navigating uncertainty, finding confidence, maintaining some semblance of work-life harmony, and more.

Eventually, I published 18 episodes.

And people did listen.

Not millions of people. This wasn't that kind of experiment.

But the 18 published episodes accumulated 2,011 listens, and what people chose to listen to taught me something I hadn't expected when I started.

More importantly, the project changed how I think about AI-assisted podcasting.

AI solved some of the problems I expected it to solve.

It didn't solve the hardest one.

That's why I wanted to go back and document the experiment.

Part 1: Why I Wanted to Try an AI-Voiced Podcast in the First Place

Starting a podcast isn't particularly difficult.

Continuing a podcast is.

There's a big difference.

You can come up with a name this afternoon. You can buy a microphone. You can make some artwork in Canva. You can probably record Episode 1 before dinner if you're determined enough.

Then next week arrives.

You have clients to deal with. Emails have accumulated. Something unexpected happens. You don't feel like recording. The room isn't quiet. You haven't decided what this week's episode is about.

Suddenly your exciting new podcast is another item on your list.

That was the part of podcasting that interested me as an AI experiment.

I wasn't trying to use AI because I believed people needed more artificially generated content in their lives.

I was interested in whether AI could remove enough production friction to make sharing useful ideas in audio more sustainable.

I needed a format that fit the experiment

At the time, I was also experimenting with affirmations for business owners.

Not just generic statements about success or positivity, but affirmations connected to situations that actually happen when you're running something.

Trying to concentrate when everything seems urgent.

Walking into a networking event and not feeling particularly confident about it.

Having a difficult conversation with your team.

Questioning whether the idea you're considering is actually a good one.

Feeling overwhelmed by uncertainty.

Trying to stop working long enough to have a life outside the business.

Those became natural episode themes.

The show eventually became:

Bite-Sized Small Business Owner Affirmations

The premise was straightforward: a short weekly podcast featuring affirmation themes and uplifting encouragement to help people start their week with more confidence, focus, and a positive mindset.

The short format mattered.

I wasn't trying to manufacture an hour-long AI podcast just because technology could produce one.

I wanted something a business owner could listen to in a few minutes.

And the recurring structure made it possible to create an identifiable show without reinventing the concept every time.

Business situation → mindset friction → targeted affirmations and encouragement.

That repeatable format turned out to matter much more than I initially realized.

Why I used my own voice

There's another choice I made early that I'd make again.

I didn't create two fictional AI hosts and have them talk to each other.

I used one voice: mine.

The audio was created with a licensed AI model of my own voice.

I've experimented separately with the increasingly common format where AI voices have a back-and-forth conversation.

My experience with that test wasn't nearly as positive.

People didn't particularly like the artificial banter.

That doesn't prove nobody wants conversational AI audio. It was a small experiment, not a scientific study.

But it changed how I approached this project.

If I had one useful idea to communicate, why create artificial conversation around it?

For this show, a solo voice felt simpler.

One topic. One voice. A few minutes. Done.

That simplicity also made the podcast a much cleaner test of the thing I was actually curious about:

Could AI make podcast production easier without making the finished content feel like an AI gimmick?

Part 2: What AI Voice Cloning Solved — and What It Didn't

The appeal of voice cloning is easy to understand.

Recording yourself takes time.

Even a five-minute episode isn't necessarily a five-minute task.

You have to get ready to record. Find an appropriate environment. Read the script. Stop when you stumble. Start again. Decide whether that weird pause bothers you enough to rerecord the sentence. Clean everything up.

And if you're creating a recurring show, the process repeats.

A usable voice model changes that equation.

Instead of every finished script creating another recording session, the script can become the basis for the audio.

That's significant.

But after doing this across 18 published episodes, I think it's important to separate content production from content creation.

They're not the same problem.

AI can remove production friction

An AI-assisted podcast workflow can help with quite a few parts of the process.

• AI can help you develop episode ideas.
• It can help organize an episode.
• It can turn rough thoughts into a usable first draft.
• It can help generate titles and descriptions.
• Voice technology can turn an approved script into audio.
• Other tools can help create artwork and supporting assets.

The podcast's hosting system can then make the finished episodes available through an RSS feed, which podcast directories use to access the show.

That's broadly how podcast distribution still works. An RSS feed contains information such as titles, descriptions, audio-file locations and artwork. Apple accepts shows through qualifying RSS feeds, and Spotify also supports RSS-based podcast distribution.

AI can make many pieces of that process dramatically faster.

But AI doesn't give you something worth saying

This became one of my bigger lessons from the project.

Generating words is easy now.

Generating audio is increasingly easy.

That means production itself is becoming less of a differentiator.

The more AI can create, the more important the question becomes:

Why should somebody spend five minutes listening to this particular thing?

That's why I don't think the best way to use AI for podcasting is:

Give me 50 podcast topics.

Then:

Write Episode 1.

Then:

Turn it into audio.

Technically, you could do that.

But you've skipped the most important work.

Who is this for?

What are they experiencing?

Why is this particular episode worth their attention right now?

The affirmation project made that especially obvious.

“Positive affirmations for business owners” is a category.

“Affirmations for a restaurant owner trying to create a great guest experience” is a situation.

“Affirmations for improving communication with your team” is a situation.

“Affirmations for navigating uncertainty in your business” is a situation.

Specificity gives the content somewhere to stand.

I eventually became interested enough in this part of the experiment that it deserves its own article: how I use AI to create affirmations around real situations instead of asking it for generic positive statements.

Voice cloning also creates a responsibility that generic AI voices don't

There's another side of using your own AI voice that I don't think should be ignored.

When the audio sounds like you, the words need to be words you're comfortable having attributed to you.

The fact that AI can turn a draft into something that sounds like you saying it does not mean you should stop reviewing the draft.

I think it means the opposite.

I want the final content to be something I actually approve.

My voice model was a production tool.

It wasn't the author.

That distinction matters to me.

And then there was the problem AI didn't solve

Consistency.

This was the surprising part.

The technology reduced several of the things that make podcast production inconvenient.

And yet the hardest part was still publishing the next episode.

My goal was generally a weekly cadence. At times it became closer to biweekly.

There was always another theme I could create.

There was always another script that could be produced.

The tools weren't the bottleneck.

Life and work still existed.

AI can shorten the path between an idea and a published episode.

It cannot make you decide that publishing the episode is important enough to happen this week.

That lesson eventually became one of the reasons I started thinking differently about the entire podcast workflow.

The system can't end with “AI can make an episode.”

It needs to answer: How does this show keep publishing?

Part 3: The 5-Step Process I Used to Turn an Idea Into 18 Published Episodes

I'm intentionally not going to turn this into a tutorial for duplicating my exact production system.

Some of the production workflow I use today is part of how I provide this as a service.

But you don't need my particular setup to understand the process.

If you wanted to build an AI-assisted podcast yourself, these are the five pieces I would solve.

Step 1: Design the show before you make the show

The first thing I'd use AI for isn't writing Episode 1.

I'd use it to make the show itself clearer.

At minimum, decide:

• Who is this for?
• What is the show about?
• Why would that particular person listen?
• How long should an episode be?
• Is it solo, interview-based, conversational or another format?
• What happens in a typical episode?
• How often can you realistically publish?
• What topics can you sustain for more than three episodes?

AI can be useful as a thinking partner here.

Give it information about your audience, expertise, existing content, services, FAQs, stories and ideas. Ask it to identify recurring themes.

Then work on the format.

For my experiment, the result was deliberately narrow: short episodes, one voice, one theme, one recognizable audience, and a repeatable progression from a business challenge into affirmations and encouragement.

That structure gave me constraints.

Constraints are useful.

If your show is simply “my thoughts about business,” every episode begins with a blank page.

If your format is one problem my audience encountered this week → what I noticed → three things I'd do about it, you've created a machine for generating episodes.

Step 2: Decide who is going to speak

Once you know what the show is, decide what role AI should have in the voice.

There are several legitimate options.

Option 1: Record yourself

AI can still help with research, outlining, scripting, titles, descriptions and production planning while you record the final audio normally. For many people, that's the right answer.

Option 2: Clone your own voice

Voice-cloning platforms such as ElevenLabs make it possible to create synthetic speech based on a voice, subject to the platform's requirements and permissions. This is closest to the route I chose. I wanted the show to remain connected to me without requiring a traditional recording session for every episode.

Option 3: Use an appropriate licensed synthetic voice

There are cases where a brand doesn't need the founder's voice at all. A synthetic narrator may be perfectly appropriate, provided you're using the voice within the relevant rights and terms.

What I would not automatically do is add more AI voices because more voices seem more sophisticated.

My separate conversational-audio experiment pushed me in the opposite direction. The AI banter wasn't adding value.

So my rule became: use as many voices as the content requires — not as many as the technology allows.

For Bite-Sized Small Business Owner Affirmations, that number was one.

Step 3: Build the entire episode package — not just a script

This is where AI becomes especially useful.

A podcast episode isn't just an MP3.

For every episode, you're potentially creating:

Topic → title → episode flow → script → audio → artwork → description → publishing metadata

If you approach every one of those as a separate creative project, the workload expands quickly.

Instead, I'd build one repeatable episode package.

For example, imagine you run a professional services firm and want a five-minute weekly show. Your episode workflow might start with:

A. Choose one real audience situation

Not: “Let's talk about leadership.”

Instead: “A good employee just made a mistake in front of a client and I need to address it without destroying their confidence.”

That's an episode.

B. Define what the listener should leave with

One useful shift? Three actions? A five-minute reset? One story and one lesson? Know the destination before generating the script.

C. Give AI a recurring episode structure

For example: hook → situation → insight → three practical ideas → next action. Your format will be different. The important thing is that it exists.

D. Develop the title after the idea is clear

AI is useful for title variations, but I wouldn't outsource the final decision. Ask for options built around different angles — curiosity, problem, outcome, mistake, contrarian idea, specific audience — then choose the one that accurately represents the episode.

E. Review the script as if you're about to say it yourself

This becomes especially important with a voice clone. Read it. Delete phrases you'd never use. Simplify sentences. Remove unnecessary introductions. Check factual claims. Ask whether the listener gets something worthwhile in exchange for their time.

Then create the audio.

F. Create the supporting assets

You'll need show artwork, and depending on your format you may want episode graphics as well. You'll also need an episode description. AI can help create a concise description from the approved script rather than forcing you to write another piece of content from scratch.

Step 4: Build the publishing pipeline before you launch

This is the part that can be confusing if you've never published a podcast before.

The simplified version looks like this:

Finished audio file → podcast host → RSS feed → podcast platforms

Your host stores or references the audio and maintains the feed containing the information podcast platforms need.

For example, Apple requires RSS-based shows to include required feed tags, artwork and at least one episode, and it validates the feed before publication.

Spotify similarly explains that an RSS feed contains the core information about a podcast, including titles, descriptions, audio links and artwork. If you're distributing beyond your host's native platform, you generally submit the feed to the other listening platforms you want to reach.

You do not need to become an RSS engineer to start a podcast.

A podcast hosting provider can handle the feed mechanics.

But you should understand the architecture because it changes how you think about publishing. You aren't normally uploading every new episode separately to every podcast app. You maintain the show through its publishing/hosting system, and the feed carries the episode information outward.

The thing I did that I'd recommend: make three episodes first

I didn't publish the podcast the moment Episode 1 existed. I created three episodes before launching it on the platforms.

I'd do that again.

Not because three is a magical podcasting number. Because Episode 1 enthusiasm is unreliable.

When you're launching something new, you're unusually motivated. You've picked the name. You've made the cover. You've finally created the thing. Of course you're willing to make an episode.

The better question is: will you still make one three weeks from now?

A small episode bank gives you breathing room. It also forces you to prove that your format works more than once.

If Episode 1 is easy to create but Episode 2 makes you wonder what on earth the show is supposed to be about, you have a format problem. Better to discover that before launch.

Step 5: Create the cadence AI can't maintain for you

This is the step I'd spend more time on if I were starting again.

Don't only design your episode-production system. Design your publishing system.

Those are different.

Ask:

• What day does a new episode publish?
• When does the topic need to be selected?
• When does the script need approval?
• When is audio generated or recorded?
• Who checks the finished episode?
• How many completed episodes should stay in reserve?
• What happens during a busy week?
• Do you have a list of future topics ready?

My original goal was a weekly show. Sometimes that rhythm stretched.

And that happened despite using AI.

That's worth emphasizing because AI marketing often focuses on how quickly something can be generated.

Speed isn't the same as consistency.

A five-minute episode that technically takes very little production time can still go unpublished because nobody owns the publishing rhythm.

For a solo project, that person is you. For a business podcast, it might be an employee, marketing partner or managed service.

But somebody has to own it.

What happened after 18 episodes?

This is where the experiment became especially interesting to me.

Across the 18 published episodes, the show generated 2,011 listens.

The average works out to roughly 112 listens per published episode, although that average hides one very large outlier.

Here were some of the stronger episodes:

And this is where I noticed something I hadn't designed the experiment to test.

The most specific episode became the biggest one

The restaurant-owner episode received 478 listens. The next-highest episode had 198.

Many of the more general topics — focus, innovation, gratitude, goal setting and similar themes — lived much closer to the middle of the pack.

I don't think 18 episodes are enough to declare that niche podcast episodes always outperform broad ones. That would be turning a small experiment into a sweeping marketing claim.

But I do think it's enough to say: specificity was worth paying attention to.

The episode that essentially said “this is for restaurant owners” stood out. A med-spa-specific episode also performed relatively well.

And that made me rethink the way I would approach a business podcast today.

Instead of asking “what should our podcast talk about?” I'd spend more time asking: who is this particular episode for, and what are they dealing with right now?

That's a much more useful content question.

And it's one of those lessons I probably wouldn't have learned from generating one AI demo and moving on.

I needed to publish enough episodes to see people make choices.

Part 4: How to Launch Your Own Podcast in 7 Days With Your First 3 Episodes Already in the Bank

If you've been thinking about starting a podcast, AI has made some parts of the process much more accessible.

But I wouldn't make “use AI” the goal.

I'd make the goal: seven days from now, have a real show, a repeatable format and three finished episodes — not just a podcast idea sitting in a document somewhere.

Here's what that week could look like.

Day 1: Define the show

Choose your audience. Define the promise. Pick your format. Decide whether it's solo, interview-based or another structure.

Use AI to pressure-test the concept and generate possibilities, but make the final decisions yourself.

By the end of Day 1, you should be able to finish this sentence:

This is a podcast for ______ who want ______. Each episode helps them ______.

Day 2: Build your episode engine

Create a list of at least 10 potential episodes. Then create your recurring episode structure.

Don't write ten scripts. Prove that the idea has enough depth to sustain a show.

Choose your first three episodes.

Day 3: Develop the first three scripts

Use AI to help turn your ideas into structured drafts. Then edit them.

This is especially important if you'll use your own cloned voice.

The final script should sound like something you're willing to say. Not something a language model thinks a podcaster should say.

Day 4: Create the audio

Record the episodes yourself or produce them through the voice workflow you've selected.

Listen to the finished files. Don't skip that step because the technology generated them.

Check pronunciation. Check pacing. Check the script. Make sure nothing sounds strange enough to distract from the message.

Day 5: Create the show package

Finish:

• show name
• cover artwork
• show description
• episode titles
• episode descriptions
• three finished audio files

At this point, the project should look like a podcast rather than a folder full of ideas.

Day 6: Set up hosting and distribution

Choose your podcast host. Set up the show. Create or enable the RSS feed. Test everything.

Then submit the show to the podcast platforms you want to reach.

Apple lets creators submit RSS-based shows through Apple Podcasts Connect and reviews submitted shows before they're made available. Spotify supports both Spotify-hosted shows and externally hosted podcasts distributed via RSS.

Allow for platform review and processing time rather than assuming every directory will make a new show publicly discoverable instantly.

Day 7: Publish — and don't use your entire buffer

Launch.

But here's the important part: don't immediately turn your three-episode head start back into zero.

Decide what your reserve should be. If you publish weekly, perhaps you always want two or three completed episodes waiting. Then build your production schedule around maintaining the buffer.

Because if my experiment taught me anything about the operational side of AI podcasting, it was this:

Making an episode easier does not automatically make publishing consistently easier.

You need both the production system and the publishing rhythm.

You can build this yourself

Everything I've described above can be assembled with tools available to individual creators and businesses.

AI can help you think through the show, develop episodes, create scripts and supporting assets. You can record your own audio or explore appropriate voice technology. A podcast host can handle the infrastructure behind your RSS feed. Apple and Spotify provide documentation for submitting and distributing shows.

You don't need my exact system to make a podcast.

But you do need to put all of those pieces together — and then keep putting them together after the excitement of launching wears off.

Or you can have the production system handled for you

That's the part I find most interesting now.

A business owner, consultant or subject-matter expert may have years of things worth talking about without wanting another production job.

You might already have the raw material sitting in client questions, articles, presentations, training materials, emails, videos, voice notes, opinions, stories, or simply what you explain to people every week.

A podcast can turn that expertise into another way for people to find you, hear you and get to know how you think.

AI can make the production side much lighter.

But somebody still needs to turn those ideas into a show, maintain the workflow and make sure episodes actually get published.

That's something I now offer through OnlineKix: done-for-you podcast creation and ongoing production for businesses that have something worth saying but don't want another weekly task.

My little affirmation podcast started as an experiment to see whether I could make a podcast with a clone of my voice.

Eighteen published episodes later, I think I was asking a slightly incomplete question.

The more useful question is:

Can AI remove enough of the production work that more people with something useful to say can actually keep saying it?

If you're running your own version of this experiment — or thinking about it — get in touch and let's talk it through.