Many companies no longer buy dedicated translation tools: they piece together their own setup. Someone pastes text into ChatGPT, someone else writes a script around the DeepL API, and the marketing team keeps a glossary in a spreadsheet and a set of translation prompts in Notion or Confluence. It works fine at first, because it’s built to their own needs. Problems tend to come later, as the setup grows.
How an AI translation workflow grows
Month 1
- You paste text into ChatGPT. It’s fast, it’s free, and the result looks great.
Month 2
- You write a few prompts to keep the tone right.
- Someone from your team starts a glossary in a spreadsheet.
Month 3
- A colleague builds a small script around a translation API, and now you can call it a translation workflow.
Month 6
- The glossary is out of date, and nobody owns it.
- The prompts behave differently after the latest model update.
- The same fixes get made again and again, because past translations are scattered across chats, mailboxes and folders.
- And when an update or product launch comes up, nobody can say for sure if the translation is right.
The workflow was built exactly for your needs. But now someone has to maintain it.
AI made translation fast and cheap. It didn’t make it reliable. Reliability takes three things most DIY setups lack: a way to check quality, a memory that keeps improving, and an expert to call when it really matters. That’s the gap we’re building Lingwise for.
Found this article interesting? Whether you’d like to use our tool or collaborate on building your own setup, let us know.
