The Dub Beat the Original by 24%. Now 80 Actors Want a Law.

TL;DR

Localization vendor 3Play Media published two months of YouTube data across eight languages showing its AI-voiced dubs held viewers 24% longer than the original English tracks, with more than 30 million hours watched. The distribution for that was already built: YouTube turned on auto dubbing in 27 languages in February, and says over 6 million people a day watch at least ten minutes of dubbed audio. Two days after the dubbing numbers landed, 80 British performers signed a letter asking the Prime Minister to make your voice your property. Both things are the same story.

The number, and who is holding the calculator

The claim is specific. Across a two-month window and eight languages, 3Play says its localized AI dubs recorded 24% higher average view duration than the English source tracks on the same channels, and that its dubs have now accumulated over 30 million hours watched on YouTube. Erik Ducker, the company's VP of Product Growth, framed it as audiences actively choosing the dub. Co-founder and co-CEO Chris Antunes put the takeaway more bluntly: YouTube creators are not dubbing on faith, they are following the data.

Two things to hold onto before you get excited. First, this is a vendor publishing a number about its own product three weeks before it demos that product at IBC in Amsterdam. Second, 3Play is explicit that these are not fully automated dubs. Trained linguists shape voice, tone, timing, and cultural adaptation throughout production, using speech-to-speech editing to retune multiple character deliveries in a session. So the result is not "AI dubbing beats humans." It is "AI dubbing with humans still in the loop beats not localizing at all," which is a genuinely different and much more boring sentence.

Why 24% is the wrong number to get excited about

Average view duration on a dub is not a fair fight against the source track, because the two are not competing for the same viewer. The English track carries the creator's home audience, including every casual click from the algorithm's biggest, loosest pool. The dub is served to someone who has set a language preference and is watching in their own language, which is already a higher-intent audience. You would expect the dub to win before a single model runs.

Then there is the survivorship problem, and it is the one that matters for your roadmap. Language tracks that underperform get dropped. What you are looking at after two months is the set of language and content pairings that survived, not the set that was attempted. It is the same reason every restaurant still open on a street looks like a good restaurant: the ones that did not work already closed, and nobody publishes their view duration.

None of this makes the number fake. It makes it a ceiling rather than a forecast. If you localize your best-performing evergreen video into a language where you already see search demand, something like this is plausible. If you bulk-dub your whole back catalog into 27 languages, you are running a different experiment, and the honest answer is that nobody has published its results.

YouTube already built the pipe

The reason a vendor can even measure this is that the platform stopped treating dubs as an exotic feature. On February 4, YouTube expanded auto dubbing to an expanded library of 27 languages, added an Expressive Speech mode in eight of them (English, French, German, Hindi, Indonesian, Italian, Portuguese, Spanish) that tries to carry the creator's original energy, and shipped a viewer-facing preferred language setting that defaults from watch history. It also quietly added filtering that recognizes when a video should not be dubbed at all, like music or a silent vlog. A lip sync pilot is testing, matching mouth movement to the translated audio.

The number YouTube gave in that post is the one worth writing down: in December, it averaged more than 6 million daily viewers who watched at least 10 minutes of auto dubbed content. Not impressions. Ten minutes.

The earlier signal was better still. When YouTube opened multi-language audio to millions of creators in September 2025, it reported that creators uploading their own tracks saw over 25% of their watch time come from views in the video's non-primary language. It said Jamie Oliver's channel amplified views 3x, and that Mark Rober uploads among the highest number of dubs, averaging over 30 languages per video. A quarter of watch time is not a growth hack. That is a distribution channel you either have or do not have.

The bill arrives on the other side of the microphone

On August 28, roughly 80 British performers signed an open letter to Prime Minister Andy Burnham backing Save Our Voices Now, a campaign co-founded by audiobook narrator Alice Sockett and performer Peter Caulfield. Signatories include Nicola Coughlan, Matt Lucas, Hugh Bonneville, Siobhan McSweeney, Luke Evans, Pearl Mackie, and Jen Brister. The campaign's premise is that three seconds of audio is now enough for a modern cloning system to produce a usable copy of a performer's voice, and that UK law has nothing specific to say about it.

The asks are narrow and, for anyone shipping a voice product, worth reading as a spec. The campaign wants a statutory right for every person to own their own voice as personal intellectual property; the burden of proving consent placed on the AI developer rather than the individual; and the right enforceable in UK courts against any platform serving UK users, wherever that platform is based. That third clause is the one with teeth, and it is the one your legal review will care about.

Note what is not in there. Nobody is asking to ban dubbing. The campaign's own framing is that it opposes theft, not the technology. Which tracks with the engagement data: the audience already voted, and it voted for the dub.

The petition math is not encouraging

Caulfield's Gov.uk petition, Strengthen copyright laws on our voices, is open until 6 January 2027 and sat at 2,844 signatures when we checked. A UK petition needs 10,000 for a government response and 100,000 to be considered for a Commons debate. Meanwhile the government's answer so far is a promised consultation, not a bill.

The uncomfortable precedent is that this exact ask has already lost once this year. A broader petition, Give UK citizens legal ownership of their face, voice, and likeness, closed on 3 June 2026 with 8,425 signatures. That is 1,575 short of the number that obliges a minister to write a single paragraph in reply.

What to do with this if you ship video

• Instrument before you localize. The useful artifact in 3Play's product is not the voice model, it is the analytics layer that tells you which video and which language to try. Per-language retention is the metric; total dubbed hours is vanity.
• Test one video, not the catalog. The 24% figure is what a surviving pairing looks like. Pick an evergreen piece where you already see foreign search demand, ship two languages, and let the numbers pick the third.
• Treat auto dubbing as the floor, not the product. It is on by default for eligible channels and costs nothing. If your paid workflow is not clearly better than free auto dubbing plus a preferred-language default, you do not have a workflow.
• Get consent in writing, in advance. If any part of your stack clones a real person, the UK ask puts the burden of proving consent on the developer and reaches platforms serving UK users regardless of where they are hosted. Denmark has already moved in this direction. Build the consent record now; retrofitting it onto shipped audio is miserable.

Key Takeaways

• 3Play Media reports 24% higher average view duration for AI-voiced dubs versus English source tracks, across eight languages over two months, plus 30 million hours watched on YouTube. It is vendor data with no published methodology.
• The comparison is flattered by selection: dubs reach viewers who already set a language preference, and underperforming language tracks get dropped before anyone measures.
• These are human-in-the-loop dubs with linguists shaping tone and timing, not fully automated output, so the result does not test automated dubbing on its own.
• YouTube's own numbers are the sturdier signal: 27 auto dubbing languages since February 4, more than 6 million daily viewers watching at least 10 minutes of dubbed content in December, and over 25% of watch time from non-primary languages for creators uploading their own tracks.
• Roughly 80 UK performers signed an August 28 letter to the Prime Minister seeking a statutory right to own your own voice, with the consent burden on the AI developer and enforceability against any platform serving UK users.
• The campaign's petition had 2,844 signatures against a 10,000 threshold; a broader version of the same ask closed in June at 8,425, and the government has offered a consultation rather than a bill.

Sources: 3Play Media, Sports Video Tech, YouTube Blog: auto dubbing, YouTube Blog: multi-language audio, Save Our Voices Now, UK Parliament petition 771831, UK Parliament petition 748841, Meath Chronicle, Martin Cid Magazine