DeepL Pro and ChatGPT translate a page from Italian to English in a few seconds, and clients still pay human translators. The reason is that the real work starts once the machine output is on your screen.
The first skill is a high level in both languages. You have to hear when a sentence is grammatically fine and still sounds wrong to a native reader, because without that ear you can't judge what the AI gave you.
Next comes editing machine translations. The tools hand you a fast first draft, and your job is to fix the tone, catch false friends and rework the passages that don't read naturally in the target language. Part of that is knowing when to give up on a draft. Some outputs are too far off to patch, and retranslating a passage is often quicker than fixing it line by line.
You also check quality on every single job. This service needs a lot of human involvement, and the check before delivery is much of what the client is paying for. And you need to be comfortable talking about price: explaining what you deliver and what it costs matters as much as the translation. For general work you don't need an industry background. Legal and medical texts are the exception, since they need specialist knowledge of their own.
The numbers in Kairo's data are small. Setup costs $0 to $20, running costs are about $10 a month, you need at least 3 hours a week and no equipment beyond the computer you already use. A first client takes around 14 days.
The margin is moderate and the competition is real, because IT/EN translators are already on Guru, Fiverr, Proz and other platforms. Speed won't set you apart when everyone has the same AI tools. The quality of your review will: natural phrasing, the right tone and no false friends in the final text.
Low cost and few weekly hours make this a realistic side project. The editing skill is what takes time to build, and it improves with each job you deliver.
