Contents10
- 01Why translate your videos
- 02Subtitles, voice-over, dubbing: three levels
- 03What platforms already do
- 04The workflow of a translated video
- 05Where the machine gets it wrong
- 06A person’s voice: consent and transparency
- 07What remains for translators and voice actors
- 08Getting translated videos found
- 09Where to start
- 10Frequently asked questions
A product demo, a tutorial, a recorded webinar: many companies already have videos that sell well in their own language. For export, they often sit in a corner of the website, for lack of budget to translate them. Until recently, subtitling in several languages took one contractor per language, and dubbing a video took a studio and voice actors.
AI changes that calculation without removing all effort: transcription, translation and voice can be automated, review remains necessary. This article looks at what each level brings and what it requires. It complements our other articles on translation and new markets.
In short
Three levels, from cheapest to most demanding: translated subtitles, voice-over, full dubbing.
Subtitles are text like any other: glossary, translation memory and native review apply.
YouTube offers audio tracks in several languages and automatic dubbing, which you can review before publishing.
A cloned voice of a real person requires their written consent, and a clear statement that the video is dubbed.
Why translate your videos
A customer discovering your product in a language they do not master understands less well, and buys less. CSA Research’s survey of online shoppers measures this for product information in general:
of online shoppers prefer to buy products presented in their own language
never buy from websites in another language
A video is often the most convincing content on a website: it shows the product in action. Leaving it in a single language keeps your best proof for the customers who speak it.
Subtitles, voice-over, dubbing: three levels
Translated subtitles
The lowest cost, and the simplest review
Readable without sound, useful for accessibility
The text can also serve search engines
Voice-over or dubbing
More natural for a tutorial or a long demo
Viewers watch the screen instead of reading
A voice to choose, to review by ear and to have accepted
In most cases, start with subtitles: they cost little, are easy to correct, and show which videos are watched in each language. Dubbing is justified next for the most viewed videos, those where reading while watching the screen gets in the way, such as a software tutorial.
What platforms already do
YouTube has built in both levels. Its help on multi-language features describes adding audio tracks in several languages to the same video or Short: viewers hear their preferred language by default, and can switch. And its automatic dubbing generates translated tracks itself, marked as auto-dubbed in the description.
The same help warns that these dubs may contain errors of pronunciation, accent or translation, and offers to review them before publishing. That is the right reflex, and it applies to every tool, whatever the model: the platform does the volume work, you keep the decision to publish.
The workflow of a translated video
From one video to five languages
- Transcription
The source-language text, reviewed, with proper names correct.
- Translation
The subtitles translated with the brand glossary.
- Review
A native speaker checks meaning, terms and reading pace.
- Voice
Optional, generated or recorded from the approved text.
- Publication
One track per language, with translated title and description.
Step three is the one people skip, and the one that prevents visible mistakes. The rest follows the logic of any translated content: a brand glossary for product names and trade terms, a translation memory so that sentences recurring from one video to the next are not translated again, and a tool chosen on your own texts, following the method in DeepL, Google or an LLM: which translates your trade best.
From the field
Subtitles go through the same pipeline as the website
On Palmora Property, a real estate portal in six languages that we operate, translation runs through a tuned pipeline, with a glossary, a persistent cache and a parity check between languages. Switching provider divided the cost by 60, and 95% of segments now come from the cache without calling the AI again. Subtitles are segments like any other: the same pipeline handles them.
Where the machine gets it wrong
- Proper names and trade terms, mistranscribed and then mistranslated: the glossary and a review of the transcript solve most of it.
- Figures and units: a price, a dimension, a date in the wrong format. The rules of localize, don’t just translate apply to video too.
- Reading pace: a translated subtitle can grow longer and scroll too fast. It needs shortening, which a reviewer does better than a tool.
- Tone: a joke, an idiom, a voice that sounds wrong on a technical word.
To decide where to place the review without re-listening to everything, apply the method of our article on measuring translation quality: a reviewed sample per video, a full review for the videos that commit the brand. The split between machine translation and review follows translating your website with AI: automatic or reviewed?.
A person’s voice: consent and transparency
Dubbing a video with a neutral synthetic voice raises no particular question. Cloning the voice of an employee, a manager or a customer, to make them say in German what they said in French, raises two.
Consent. Plan a written agreement from the person, specific about the languages, the videos, the duration and what happens if they leave the company. The text and the voice also go through a provider: where do they go, how long do they stay, are they used to train a model? The questions to ask are those of translation and GDPR: where your translated texts go.
Transparency. A realistic voice making a real person speak a language they never spoke can pass for authentic. Article 50 of the AI Act requires whoever publishes generated or manipulated audio or video of this kind to disclose it; the details and the timeline are in our article on marking AI-generated content. An “AI-dubbed version” notice at the start of the video or in its description is often enough, and it prevents a customer from feeling misled.
Watch out
What should not be dubbed lightly
A customer testimonial, a statement that commits the company, a video with contractual value: these are the videos where a translation error costs most. For those, reviewed subtitles or dubbing approved line by line by someone fluent in the language. Content with legal weight follows the precautions of translating a contract with AI: where it stops.
What remains for translators and voice actors
AI takes the volume: dozens of tutorials, video answers to frequent questions, webinar clips. The brand film, the widely broadcast advertisement, the video that opens a market often keep a translator, an adapter and a recorded voice, because the right intonation sells there. Neither profession disappears: both concentrate on the videos where their work shows, and on reviewing what the machine has produced.
Getting translated videos found
A dubbed video nobody finds is useless. Translate the title, the description and the keywords for each language, starting from what customers in that country actually search for, and publish the transcript on the website page hosting the video, in each language. Start with the videos that already serve sales, the way you choose which blog articles to translate first. For a catalogue, demo videos accompany the product pages described in product pages for export, and for software, they follow interface updates, as translating software or an app explains.
The choice of languages comes first: which European market to open first, what changes beyond language in an English-speaking market, and the place of video commerce in some markets, discussed in China and Southeast Asia: website, payment, platforms. The budget is calculated with our article on website translation cost, and the questions from foreign customers that videos bring in are handled as described in translating customer support emails without hiring. Internally too, a subtitled video helps a multilingual team work together.
Where to start
Pick the five videos that serve sales most in your language, and the first foreign market you are targeting. Have them subtitled with your glossary, reviewed by a native speaker, then publish them and see which ones are watched. Dubbing will come for those that deserve it. To set up this workflow with your tools and your data rules, tell us about your videos and your markets.
Frequently asked questions
Subtitles or dubbing: which to choose for export?
Start with subtitles: they cost little, are reviewed like text and can be read without sound. Dubbing is justified for long or technical videos, where reading while watching the screen gets in the way, and for markets where audiences expect a dubbed version. First measure which subtitled videos are watched, then dub those.
Is YouTube’s automatic dubbing reliable?
It saves a lot of time, but YouTube itself warns that dubs may contain errors of pronunciation, accent or translation. Turn on manual publishing to listen to each track before it goes live, at least for videos that present your offer. For a video that commits the brand, have the translation approved by a native speaker.
Can you clone an employee’s voice to dub a video?
Yes, with their written and specific consent: which videos, which languages, how long, and what happens if they leave the company. Also check what the provider does with the recording. And tell your audience the version is AI-dubbed: a realistic voice in a language the person never spoke can pass for authentic.
Do you have to disclose that a video is AI-dubbed?
It is recommended in all cases, and the AI Act requires it for generated audio or video that could pass for authentic, such as the cloned voice of a real person. A notice in the description or at the start of the video is often enough. YouTube itself marks its automatic dubs in the description.
Sources
- YouTube Help, multi-language features (new tab)support.google.com
- YouTube Help, automatic dubbing (new tab)support.google.com
- AI Act Service Desk, Article 50 (new tab)ai-act-service-desk.ec.europa.eu
Jim · founder of Brain Plus AI
Also in translation and new markets
Practical guide8 min
Brand glossary: keeping translations consistent
The same product called three different things on three pages, and readers wonder whether there are three products. A brand glossary fixes the terms that are not up for debate, before the first translation, whatever the tool or the translator.
Understand8 min
Translation quality: measuring it without rereading it all
Nobody rereads every sentence of a website in six languages. Translation quality can still be measured: automatic checks that catch errors without reading, metrics that compare tools, and a sample reviewed by someone who reads the language, with errors counted by type and severity.
Choose and price7 min
Translating your website with AI: automatic or reviewed?
Having everything reviewed is expensive, and reviewing nothing lets errors through in languages that nobody at your company reads. The split is decided page by page, depending on what each page commits you to.