3 Months of AI Voiceover Gigs — Real Numbers and What Actually Worked

so there’s this story circulating in a few private slack groups about someone who accidentally built a decent side income from AI voiceover gigs. not a course. not a guru thing. just a person with a laptop and too much free time who decided to actually test the theory instead of just reading about it.

the backstory: 3 months ago, scrolling reddit at 2am, saw some guy talking about AI voiceover gigs on Fiverr. thought it was bullshit at first — figured the market was flooded. but curiosity won and the experiment started.

turns out the market IS flooded, but 90% of the people selling are horrible at it. like genuinely bad. robotic delivery, bad audio quality, can’t follow basic instructions. the bar is so low it’s basically a tripping hazard in hell.

the setup (total cost ~$40)

ElevenLabs Turbo v2 — $22/month for the creator plan. this thing sounds genuinely human now. not even close to the old robotic TTS. if you haven’t heard v2, go listen to a demo. it’s creepy good.

Audacity — free. for cleanup and noise reduction. tour guide ads, internal training videos, YouTube narration — all of these need someone to strip out background noise and normalize volume levels.

a $60 USB mic that was already available. honestly not even necessary for AI voiceover since the AI generates the audio and the editing is what matters. but useful for client calls.

Canva free tier — for quick video+voice combos. some clients want a talking head video with voiceover. canva does that now without premium.

that’s the whole stack. $22/month for the AI. everything else free or already owned.

how clients were found (the real trick)

didn’t make a Fiverr gig. that’s a race to the bottom. everyone and their mom is selling “AI voiceovers” for $5 on Fiverr and delivering garbage.

instead went direct to businesses that OBVIOUSLY need voice work but don’t know they need it yet.

went through Google Maps searching for “real estate agent [city]” — picked a mid-sized city nearby. went through the first 50 results. found 28 agents who had property tour videos on their websites but the audio was them recording on an iPhone in a noisy room. traffic in the background. kids watching Peppa Pig in the next room. the whole deal.

sent each of them a super short email:

“hey [name], was watching your [property name] tour video. great staging btw. noticed the audio had some background noise — a common issue with on-site recordings. can clean this one up and re-voice it as a free sample if you want. no strings.”

out of 28 emails: 8 replied, 4 took up the free sample offer, 2 of those became paying clients.

one is a luxury agent paying $150/month for voiceovers on all listing videos. the other is a small brokerage paying $80/month for 6 tour voiceovers per month + cleanup.

monthly recurring from those two: $230. roughly 10 hours of work to set up everything including recording the first batch of samples.

then the approach was expanded to other niches.

the expansion

went after:

– elearning courses on Udemy. found instructors with great content but terrible audio. offered to re-do their course audio for a flat fee + small royalty. got 1 yes — $200 setup + 5% of course revenue.

– local restaurants with social media video ads. the ones doing “daily special” videos with AI voiceover are painful to watch. offered to do better versions. 0 bites from 30 emails. this niche is a dead end.

– YouTube automation channels. you know those “top 10 facts about…” channels with AI voice? most use the free TTS that sounds like a robot from 2015. offered to upgrade audio quality for a flat fee. 2 clients from 50 emails. one pays $60/month for 8 videos, the other $40/month for 5 videos.

running total: $230 + $200 (one-time) + $100/month = $330/month recurring + $200 one-time. not life changing. but 3 months in, roughly 25 total hours invested.

lessons learned the hard way

1. do NOT oversell the AI part. clients don’t care if it’s AI or human. they care if it sounds good. the AI is never mentioned in the pitch. the phrasing used is “professional audio processing tools.” when clients hear the result and it sounds clean, they don’t ask questions.

2. sample quality matters more than anything. the first samples were mediocre because the voice settings weren’t tuned properly. spent a whole weekend just testing voice parameters — stability, clarity, style exaggeration — until the sweet spot was found. now there are 3 saved voice presets for different content types (real estate warm, elearning professional, YouTube energetic). makes delivery 10x faster.

3. the most overlooked skill is AUDIO EDITING. not voice generation. elevenlabs outputs good audio. but clients need someone who can: strip silence, normalize loudness to broadcast standard, remove mouth clicks, add intro/outro music, match pacing to video. THAT’s what they pay for. the AI is just the engine. the editing is the product.

4. background noise in the original video is the best sales tool. when a prospect sends a video with terrible audio, that’s an opening. show them what’s possible once, then pitch the service. if they don’t bite, move on. there’s infinite supply of noisy videos on the internet.

5. upwork and fiverr are not worth it at the start. too much competition, too low prices. direct outreach to specific businesses in specific niches is 100x more effective. send 10 personalized emails and you might get 1-2 clients. list on fiverr and you’ll be fighting for visibility against 5000 other sellers charging $5.

the math so far

month 1: $0 (setup + learning + free samples)
month 2: $230 (real estate + brokerage)
month 3: projected $330 + $200 one-time

monthly costs: $22 elevenlabs + $5 domain + $0 everything else = $27 total

scaling is happening through a simple referral system. when a real estate client is closed, they’re asked if they know other agents who do video tours. got 2 referrals just last week from the luxury agent alone. one is a commercial real estate guy who wants 15 property videos done by next month.

the takeaway for anyone trying this

the AI is the easy part. everyone has access to the same models. the moat is:

– knowing which specific businesses have this problem right now
– being able to pitch without sounding like a spam bot
– delivering consistent quality that beats 90% of what’s out there
– having a system for scaling without burning out

the window on this specific niche might be 6-12 months before saturation. but by then there’ll be another AI niche to target. there’s always another niche.