How We Offered 50 Languages

This article was revised from an original version published on March 19, 2025.

A team of two

There were two of us, one website in English, and a world full of people who speak other languages. Hiring human translators for every page was never going to happen, so I went looking for another way to say hello to the world.

The call

Then came the idea: let an AI translation service do it. I built a simple setup that sent each English page to the AI and saved whatever came back. In the first version of this article, I admitted we were “hoping for the best.”

You might ask why we did not just let each visitor’s browser translate the page. A browser can do that, but only after the visitor has already found the page, and search engines like Google never see the result. They only see the English. We wanted each language to be a real page that search engines can read, index and show to people searching in their own language. That is how we hope to reach readers all over the world, not only the ones who already speak English.

The trials

Hoping was not enough. The AI almost always answered, and most answers were good. Some were quietly broken.

Malayalam, Tamil and Telugu pages came back with Chinese and Japanese characters sprinkled through them. Serbian arrived in Cyrillic letters, when our site uses the Latin alphabet.

That was the real danger. Nobody on our team can read 48 languages, so nobody would have noticed. The AI was a powerful ally, but it could not be trusted on its own, and I was the one who would have to find a way around that.

The gift

Since no person could proofread 48 languages, I built automatic checks that do the part a computer can do. Now every answer is checked by software before it is saved. The checks look for the kinds of problems we had seen, and others: repeated words, cut-off pages, untranslated copies, broken links, a missing sentence.

When one piece fails, only that piece is redone. If it fails again, a stronger second AI takes a turn. And anything that still fails is never saved. A bad page simply waits, rather than reaching a visitor.

The return

Before anything goes live, the software runs one more scan of every translated page, and a safety gate stops a broken page from reaching visitors. It cannot judge whether a translation is beautiful, only whether it is broken, so that is the limit of what we promise. Each page also gets a fingerprint, so when we change the English, only the changed pages are redone. That saves time, and it means the checks stay quick and easy to repeat.

The lesson

Let the AI translate, but never save its work without an automatic check in between.

If you are reading this in a language other than English, I would love your help. Did anything sound strange, awkward or just wrong? Please tell me. No one on our team can read every language, so your feedback is the one check the software cannot do.

David
Founder, EVnSteven.app
david@evnsteven.app
WhatsApp: +1 236 882 2034
LinkedIn