The short answer
To translate a document with AI, upload the whole file to a tool that keeps the layout. DeepL Translator has a Translate files mode for PDF, Word, PowerPoint, Excel and more, with one free file a month up to 5 MB; free texts may train DeepL's models, paid plans are never used for training. Microsoft Word has Review, Translate, Translate Document, which creates a translated copy. Then use a chatbot such as Claude or ChatGPT to check terms and tone.
What you need before you start
- The final source file. Translate the last version. Every change after translation has to be made twice.
- A list of key terms. Product names, job titles, legal phrases and anything your company always translates the same way. Ten to twenty terms is plenty to start a glossary.
- A decision about confidentiality. DeepL's privacy policy says texts sent to the free version are used for a limited time to train and improve its models, including corrections you make, and describes no opt-out switch. On paid (Pro) plans, texts are kept only as long as needed and never used for training. Contracts, HR files and anything under NDA belong on a paid plan or your company's approved tool.
- A tool. DeepL Translator (free version: 50,000 characters a month, one file a month up to 5 MB, one glossary with up to 5 entries per language pair), Microsoft Word with a Microsoft 365 subscription or Office 2021 on Windows (Office 2021 or 2024 for Mac also qualify), or a chatbot for short texts and checking.
- Someone who reads the target language, for anything public, legal or customer-facing.
How to translate a document with DeepL, step by step
- Go to DeepL Translator and choose Translate files (next to Translate text).
- Drag and drop your file. DeepL lists PDFs (scanned and digital), Word documents, PowerPoint slides, Excel spreadsheets, HTML, SRT and TXT files among the formats it handles, and says it keeps the original formatting and layout.
- Choose the target language. DeepL says it translates into over 100 languages.
- If you have a glossary, apply it so your key terms are always translated the same way. Free and Individual plans get one glossary; Team gets five with up to 1,000 entries per language pair; Business and Enterprise get unlimited glossaries.
- Translate and download the file, then open it and check the layout: long words in German or Finnish can overflow text boxes and table cells.
- Review the text (see the check below) before you send it.
File size limits depend on the plan: up to 5 MB on the free version and up to 100 MB on Business. Team adds bulk file translation.
How to translate a document in Microsoft Word
These steps follow Microsoft's support page.
- Open the document in Word.
- Select Review, then Translate, then Translate Document.
- Select your target language.
- Select Translate. A machine-translated copy opens in a separate window.
- In the original window, select OK to close the translator.
You need an internet connection and Office connected experiences switched on. Word's translator is handy because the copy keeps your Word formatting and you never leave the app, but it has no glossary, so check your key terms by hand.
How to use a chatbot for translation and checking
Chatbots such as Claude, ChatGPT and Gemini translate short texts well and are good at adapting tone, but they rebuild the layout of a file rather than keep it. Use them for emails, web copy and checking, and use DeepL or Word for formatted files.
- Upload or paste the text. In Claude, click the + button in the lower left of the chat box, then Add files or photos. In Gemini, click Add files, then Upload Files.
- Give it your glossary in the prompt and the audience ("formal German for a client's legal team").
- Ask for a back-translation into the original language, to spot meaning that drifted.
Since September 2026 DeepL also works inside Microsoft Copilot, ChatGPT and Claude through MCP, after you sign in with a DeepL account, so you can ask the assistant to translate text and documents with your DeepL glossaries and style rules.
A worked example: a supplier contract summary into German
Marta, an operations manager in Dublin, needs a three-page supplier summary in German for colleagues in Munich. It has a table of delivery terms and the company's product names.
- Because the summary names prices, she uses her company's paid DeepL Team plan, not the free version.
- She creates a glossary with six terms: the three product names (kept in English), "service level agreement" as "Service-Level-Vereinbarung", "lead time" as "Lieferzeit" and "purchase order" as "Bestellung".
- She chooses Translate files, drops in the Word file, selects German and applies the glossary.
- In the downloaded file, one table column is too narrow for "Service-Level-Vereinbarung"; she widens it.
- She pastes the German text into Claude with the prompt "Back-translate into English and list any sentence whose meaning differs from the original I pasted below." Claude flags one sentence where "within 30 days" has become "after 30 days". She fixes it.
- A German-speaking colleague reads it in ten minutes and changes one word for tone. Done.
Common mistakes and how to fix them
- Putting confidential files into a free tool. Fix: use a paid DeepL plan, your company's licensed tool, or remove sensitive details first.
- Inconsistent terms. The same product name gets three translations. Fix: set a glossary before translating.
- Translating fragments. Single lines out of context are more likely to go wrong. Fix: translate whole paragraphs or the whole file.
- Broken layout. Fix: check every table and text box after download, and leave space in templates for longer languages.
- Numbers, dates and units changed. Fix: ask a chatbot to list every number and date in both versions side by side.
- Skipping the human check. For contracts, medical or safety text, machine translation is a draft. Fix: have a qualified speaker review it.
- Running out of the free allowance. One file a month goes fast. Fix: paste short documents as text, or move to a paid plan if you translate often.
Which AI tool to use to translate a document
| Your situation | Use | Why |
|---|---|---|
| Formatted files, regular work, terms that must stay consistent | DeepL (paid plan) | Keeps layout, glossaries, never trains on paid-plan text |
| A Word file, occasionally | Word's Translate Document | Built in with Microsoft 365, makes a translated copy |
| Emails, web copy, tone changes | Claude, ChatGPT or Gemini | Good at adapting style and explaining choices |
| Checking any translation | A chatbot with a back-translation prompt | Finds meaning that drifted |
Our opinion: DeepL is the best default for documents because of glossaries and layout, and the paid plan is worth it the moment you translate anything confidential. For a broader ranking see the best AI for translation, DeepL vs ChatGPT and DeepL alternatives. Tool guides: DeepL, Claude, ChatGPT and Copilot. Learning the language yourself? Try how to learn a language with AI.
Prompts to copy
Translate with a glossary
Translate the text below into [language] for [audience]. Use these terms exactly: [term = translation, ...]. Keep product names in English. Keep every number, date and unit exactly as written. Formal register. [paste text]
You get: A consistent translation that respects your terms
Back-translation check
Here is an original text and its [language] translation. Back-translate the translation into [original language], then list every sentence where the meaning, a number, a date or an obligation differs from the original. Original: [paste]. Translation: [paste]
You get: A list of errors to fix
Numbers and names audit
List every number, date, amount, unit, name and product name in the original and in the translation, side by side in a table, and mark any pair that does not match.
You get: A quick table that catches the costly mistakes
Build a glossary
Read this document and list the 20 terms most important to translate consistently into [language]: product names, job titles, legal or technical terms. Suggest a translation for each and mark the ones that should stay in the original language.
You get: A starter glossary for DeepL
Adapt, do not just translate
Rewrite this [language] translation so it reads naturally to a [country] reader at [audience level]. Keep the meaning and all facts. List the three biggest changes you made and why.
You get: A translation that sounds local