ElevenLabs vs Google Cloud Voices for Faceless Shorts: What 396 Viewer Ratings Told Us
We compared how users rated 396 AutoShort videos by the voice that narrated them. The premium voice did not win. Here is the data, what it can and cannot prove, and how to choose.
The short answer
Across 396 videos our users rated, the ones narrated by Google Cloud voices were marked helpful 92.6% of the time. The ones narrated by ElevenLabs were marked helpful 69.4% of the time. That is the opposite of what we expected. AutoShort uses Google voices by default and offers ElevenLabs as a paid add-on, and this data is why we are comfortable keeping it that way. It is also not proof that Google sounds better, and the rest of this post explains why.
The data
After a video finishes, AutoShort asks the creator whether it was helpful: a thumbs up or down, with an optional comment. On 13 October 2026 we joined every one of those ratings to the voice that narrated the video. There were 396.
| Voice | Rated videos | Marked helpful | Share |
|---|---|---|---|
| Google Cloud (WaveNet / Chirp 3 HD) | 148 | 137 | 92.6% |
| Piper (old open-source voice, retired) | 136 | 110 | 80.9% |
| ElevenLabs | 108 | 75 | 69.4% |
Four more rows came from a Vietnamese voice provider we no longer use; four is too few to say anything about. The most useful number is not in the table: only 4 of the 396 comments mention the voice, the audio or the narration at all, and those four were about the video. One said, roughly, that the audio was quite good but the video was not.
What it can and cannot prove
Be careful with this table. We were, and here is what keeps it honest:
- The rating is for the video, not the voice. A thumbs down on a video with a great voice and the wrong pictures still counts against that voice.
- The groups are not random. Different voices were the default at different times, and the product improved over those months. Videos from a period with weaker scenes will drag down whichever voice was the default then.
- Who picks a voice matters. People who deliberately choose a premium voice may also have higher expectations of everything else.
- It is one product. These are AutoShort videos, mostly in English, made from AI scripts. Your channel is not our sample.
What the data does show clearly is the thing we most needed to know: viewers and creators are not complaining about voices. When people are unhappy with a faceless Short, they are unhappy with the visuals, the script or the pacing. Paying more for narration does not fix those.
Why the premium voice might lose
We have three working theories, none proven. First, the most expressive voices can be too dramatic for a 30-second explainer; a calm, clear read fits the format better than a movie-trailer one. Second, a monthly character quota runs out mid-month on a busy account, so a channel can switch voices halfway through, which sounds inconsistent. Third, the ratings are mostly about pictures, and the voice is just along for the ride.
The practical lesson is the same whichever theory is right: pick a voice that sounds clear and steady at the speed your script needs, then spend your effort on the hook and the scenes.
When ElevenLabs is worth it
There are real cases where we would still pick ElevenLabs:
- Fiction and storytelling. Pacing carries tension in a story, and a voice that pauses before the turn is worth paying for.
- A recognisable channel voice. If the voice is part of your brand, its voice library and cloning features give you something distinctive.
- Languages Google does not cover well. A few languages are only spoken well by ElevenLabs.
- Work outside AutoShort. For podcasts, audiobooks or ads, a direct ElevenLabs plan makes sense on its own. We pay for the Creator plan ourselves.
How to test voices on your own channel
Do not trust our table, or anyone's. Run a small test of your own:
- Pick one script you would publish anyway.
- Render it twice, changing only the voice. Same scenes, same captions, same length.
- Post them a few days apart, or to two similar audiences, and compare average view duration, not views. Views depend on the algorithm; duration tells you whether people kept listening.
- Repeat with two or three more scripts before you decide. One pair is a coin flip.
If the premium voice wins clearly on watch time, keep it. If the difference is noise, take the free voice and put the savings into more videos and better hooks.
Settings that matter more than the provider
Whichever voice you choose, three settings change how it lands far more than the brand name does. Speed: most voices sound rushed at the pace AI scripts are written for, so cut words before you speed the voice up. Punctuation: a full stop gives a real pause and a comma a short one, so edit the script for breath, not grammar. Consistency: use the same voice across a series, because a channel that changes narrator every few videos feels like a different channel.
What each option costs
- Inside AutoShort: Google Cloud voices are included in every video. An ElevenLabs voice adds 25 credits, so a standard Short goes from 40 to 65 credits. You do not need your own ElevenLabs account.
- Google Cloud directly: pay per character, with a monthly free allowance for WaveNet voices. The newest HD voices cost noticeably more per character, which is worth knowing if you build your own pipeline.
- ElevenLabs directly: monthly plans with a character quota; the Creator plan is what we use. Check their pricing page for current limits, since quotas change.
Our default is the cheapest voice that viewers did not penalise. If your own test says otherwise for your audience, follow your data, not ours.
Written by AutoShort Team