What I’d Do Differently If I Started Over on ElevenLabs

I’ve created two voice clones on ElevenLabs. The first one I did everything wrong. The second one I did everything differently. Here’s what I’d change if I started over today, in the order it would have saved me the most time and money.

1. I’d skip the Blue Yeti and buy the AT2020 from the start

I bought a Blue Yeti for $100 because every YouTube tutorial recommended it. The voice I recorded with it peaked at €50/month and was earning €2.84 by month 8. ElevenLabs rejected it for the High Quality badge because of plosives and mouth sounds.

The Audio-Technica AT2020 is the mic ElevenLabs actually recommends in their documentation. It’s an XLR mic (not USB), which means you also need an audio interface – but the total setup is €160 and the audio quality difference is obvious.

My second voice, recorded with the AT2020, earned €76.50 in its first month and €263.42 in 3 months. The equipment paid for itself in month 2.

2. I’d buy a pop filter before recording a single word

An €8 pop filter. That’s what got my first voice rejected for the High Quality badge. Plosives – the harsh “P” and “B” sounds that make audio sound amateur. A pop filter between your mouth and the mic eliminates them.

I recorded my entire first voice without one. By the time I realized the problem, I’d already spent hours recording and editing audio that would never pass quality review. €8 would have prevented all of it.

3. I’d record 2 hours, not 30 minutes

My first voice: 30 minutes of audio. Lazy. I didn’t want to do more.

My second voice: 2 hours of clean audio across multiple sessions. Not 2 hours of work – 2 hours of usable material. The actual recording took longer because you keep making small mistakes, stopping, redoing sections. It’s exhausting.

But ElevenLabs’ quality dial goes from “Good” (30 min) to “Better” (1 hour) to “Best” (2 hours). More varied audio means the AI learns your full vocal range – different emotions, pacing, energy levels. The clone sounds more natural and handles more types of content. 30 minutes wasn’t enough.

4. I’d actually edit the audio properly

With my first voice, I did some basic noise removal and called it done. No compression, no normalization, no loudness check. The result sounded like a bedroom recording.

For my second voice, I spent 12+ hours editing. Listening back over and over. Running the 6-step Audacity workflow: noise reduction, truncate silence, amplify measurement, normalize, compress, final amplify. Then checking loudness in Youlean before uploading.

The compression step is the one that makes the biggest difference. It evens out your volume so loud words and quiet words are consistent. Without it, your clone reproduces that inconsistency. Most people skip it. Don’t.

5. I’d test the voice clone before publishing

With my first voice, I uploaded the audio, created the clone, and published it. Never tested it. Never listened to what it actually sounded like generating speech. Just hit publish and hoped.

With my second voice, I ran 5 test scenarios before publishing – conversational, energetic, storytelling, numbers/dates, and dialogue with emotion shifts. When I wasn’t happy with how it handled certain phrases, I went back and re-recorded those sections. Testing takes 15 minutes and catches problems before your users find them.

6. I’d train the voice for all models, not just the default

After ElevenLabs finishes creating your voice clone, there’s a popup that says you should also train it for other models – Multilingual v2, Flash v2.5, Turbo v2.5. I completely missed this popup on my first voice. Never saw it. Never did it.

This means my first voice only worked with one model. Users who wanted to use a different model (faster generation, different language support) couldn’t use my voice. That’s potential earnings I never got. On my second voice, I trained for all available models. It takes 5 extra minutes.

7. I’d set the notice period to 2 years immediately

The notice period is how long before your voice can be removed after you request it. Longer notice = higher earnings multiplier. The 2-year notice period gives you a 2.75x multiplier – nearly 3x earnings for zero additional work.

85% of the top 300 voices by usage chose 2 years. It sounds like a big commitment, but you can still request removal at any time – it just takes 2 years to execute. Your voice is passive income anyway. The multiplier alone makes it worth it.

8. I’d downgrade to Starter after publishing

You need the Creator plan ($11/month) to create your Professional Voice Clone. But once it’s published and live, you can downgrade to Starter ($5/month) and your voice keeps earning. Full monetization, full payouts, stays in search results.

I’ve been on Starter since my second voice published. It works. Saves $72/year per account. I don’t know why this isn’t mentioned in every ElevenLabs guide.

*This article contains affiliate links. If you sign up through one of these links, I may receive a commission at no extra cost to you. The price you pay stays the same. I only recommend tools I actually use.