Using genAI to translate a game is as bad as using it to make assets – Whatever the Wind Brings

Using genAI to translate a game is as bad as using it to make assets

15 Jun, 2026

Man, I'm tired. I worked 7 days a week in May because I was dealing with some important projects, but also because I've got bills to pay and my rates have been mostly frozen for the last few years due to the rise of LLMs as a "translation solution". Count me lucky, because I'm one of the increasingly few people still working with only human-made translation and text, as a lot of companies still turn to machine translation and, now, LLMs as a way to save money. Which is funny, because as far as I know, the video game industry is the most profitable entertainment business in the world, so why does it need to be constantly cutting costs?

Anyway, I digress.

These "translation solutions" actually create more problems than they solve: not only companies using them pay less for the work, but they expect you to work more, not only to make it up in more earnings/income/returns, but because they are sold as a "faster solution". And, well, they are not: it takes as much time to review a machine-translated text as it takes to review a human-made one, with one extra detail: the cognitive load required to check and correct machine-translated content is way higher than human-made one, because you have not only to be constantly fixing mistakes that humans wouldn't make, but most of the time you have to rewrite the whole thing because there are so many "small" issues — and that's even if the machine got the context right — that it's literally easier to redo everything from scratch.

So, why am I rambling about this? Because I just saw this article on PCGamer being shared online. This paragraph was the part that made me angry (emphasis is mine):

Overall I was surprised by how many of these 120 games use generative AI for music and assets. I had assumed that the vast majority would be for translation and store page images. The former raises quality control questions and is obviously a disaster for actual humans who are employed to translate games, but it's arguably less egregious compared to machine-made music, textures and narrative.

And, like... no it's not? Just to be clear, I'm not calling out PCGamer or this writer specifically. Many people throughout the industry share this position, so it's not a unique point of conflict, it's a general attitude towards translation in video games, this idea of "oh, assets and music made by genAI are bad, but translation is kind of ok, it's not as if it's art or something". And I mean, it kinda is? Anybody who speaks more than one language knows you can't take things from one language and just swap the words around for another language. That's not how language works, and it's why translators still exist in a world where Google Translate has been around for decades. To translate, you have to rewrite the text in your language from scratch, taking tone, terminology, expression, and so on into consideration. It might not be original writing (in a sense), but it is creative writing nonetheless. It's why I translate games from English, but not into English. Using my own voice to write blog posts and marketing is one thing, but having to accurately reproduce the voice of another writer in a language and culture I didn't grow up in is way more harder to do accurately. So, having AI translate a game should be as egregious as having it generate art and assets for a game.

I mean, last month I proofread and edited around 90k words. 90 thousand words. If it were a manuscript, it would've been hundreds of pages long. A whole book! But it was just a small portion of this large project that's been going for a while, with multiple translators and 3 or 4 more proofreaders/editors taking care of other parts that I can't, because I'm just one person and the deadlines don't give me enough time to check everything. All the while, I have to keep an eye on terminology and voice consistency because, yes, it's a video game, but it's also telling a story and it needs a literary aspect to be pleasant, consistent, and enjoyable to read while playing. And that's a skill that takes years to develop and hone. And just like genAI images are a facsimile of human expression, genAI translation is a facsimile of human communication: it can mimic it, but not consistently, not with quality, and it can't generate new elements the text might need (like neologisms, original metaphors, and so on).

Think about it this way: if you take a genAI image that gets "close enough" and send it to an artist, asking "can you fix this and that parts to make it better?", they will probably say "I can, but it will be shit and have an uneven style, it's better if I draw a new one from scratch". And then you're paying for an original piece of art anyway. Text created with AI is the same thing. Everybody thinks it's easier to deal with because everybody knows...