Guide
Translating documents with AI: build a glossary, check the result
A quote goes out in German to a customer in Zurich. With a glossary, the heat pump has the same name on every page, and you know which passages you still have to read.
A building services company with an English-speaking sales team sends a heat pump quote to a customer in Zurich, in German. The machine translation reads smoothly. Even so, page 1 can say “Angebot” and page 3 “Kostenvoranschlag”, when Swiss customers expect “Offerte”. The product name “CellarTherm 9” can come back half translated. And “on-call service” can turn into a wordy description instead of the Swiss term “Pikettdienst”.
What the AI lacks is your company’s terminology, and that’s what a glossary is for. This guide shows how to build one and what to check in a translated quote before it goes out. The company and product are invented.
In short
- A glossary fixes how a term is translated in one language direction.
- 20 to 30 entries from your existing documents are enough to start.
- Always read numbers, exclusions and deadlines yourself, glossary or not.
What a glossary does
A glossary is a list of term pairs for a language pair, such as English to German. When you select one for a translation, DeepL uses the translations you’ve set, as the DeepL developer documentation describes. For the quote, a first version looks like this:
| English | German | Why it’s in the glossary |
|---|---|---|
| quote | Offerte | should be the same word throughout, and Swiss customers expect it |
| heat pump | Wärmepumpe | core term, must not change |
| commissioning | Inbetriebnahme | appears in the price and the warranty |
| maintenance contract | Wartungsvertrag | a separate product with its own price |
| on-call service | Pikettdienst | Swiss term, otherwise gets paraphrased |
| time and materials | Regiearbeiten | billing type, must be unambiguous |
| CellarTherm 9 | CellarTherm 9 | product name stays unchanged |
Enter product names with the same text in both columns, so nothing gets translated or rearranged.
Where the entries come from
You don’t need to write a technical dictionary. Good sources are:
- earlier quotes that a German-speaking colleague has already corrected. Each correction is a glossary entry.
- your price list and product data sheets, if they already exist in both languages.
- terms customers often ask about.
Choose one person to add new entries. Otherwise every department soon has its own glossary.
What you still need to read yourself
A glossary keeps terms consistent. It doesn’t check whether a sentence is right. In every translated quote, compare these passages with the original:
- Prices, quantities and VAT. Do the amounts match, and do the separators suit the reader? 12,450.50 in English is 12’450.50 in Switzerland and 12.450,50 in Germany.
- Exclusions. Does “not included” come out as “nicht inbegriffen”, and not as the opposite?
- Deadlines and how long the quote is valid, with dates.
- Multi-word terms like “heat pump water heater”. German makes one word of them, “Wärmepumpenboiler”, and an entry for “heat pump” won’t reliably cover it, so look at these separately.
- Form of address. A quote to a private customer in German uses the formal “Sie”.
If nobody in your team reads the target language, have someone who does review the quote before it goes out. Contracts, terms and conditions, and safety notes in installation manuals need someone confident in both languages in any case, because one wrong word can have legal consequences.
In onAI
The onAI translator runs on DeepL. Word files and PDFs come back with the layout intact. You define a glossary once and share it across your organisation, and it works for pasted text and whole documents. You choose formal or informal address before you translate.
To start, take the last three quotes you had translated and note every correction. That’s your first glossary, ready before the next quote goes out.
