The short answer
The best AI for translation in November 2026 is DeepL, from DeepL SE in Germany, because it translates whole files with their layout, keeps terms fixed with glossaries and, since 15 September 2026, translates live speech in more than 30 languages. ChatGPT is the runner-up for adapting and replying, with ChatGPT Translate covering 40+ languages. The best free option for files is Google Translate, which takes documents up to 10 MB.
| Our pick | DeepL (DeepL SE, Cologne) |
|---|---|
| Runner-up | ChatGPT, with ChatGPT Translate |
| Best free | Google Translate for files up to 10 MB |
| Best for live meetings | DeepL Voice desktop app for Zoom, Teams and Meet |
| Free DeepL limits | 50,000 characters and 1 file up to 5 MB a month |
| Watch out for | DeepL's free version may train on your text, with no opt-out |
The best AI for translation in November 2026
Translation at work is rarely one job. Sometimes you need a contract back in German with its numbered clauses intact. Sometimes you need a Spanish reply to a customer that sounds human. Sometimes you need to follow a meeting in Japanese. No single tool wins all three, so here are our picks, based on what each maker documents today and the checks behind our DeepL vs ChatGPT comparison.
| Tool | Pick it if | The catch | |
|---|---|---|---|
| Our pick | DeepL | You translate documents, need fixed terminology or run multilingual meetings | The free version may train on your text, with no opt-out switch |
| Runner-up | ChatGPT | Translation is one step of a bigger job: adapting, replying, summarising | No glossary engine of its own; training is on by default on personal plans |
| Best free | Google Translate | You need whole files translated at no cost | Text in images is not translated, and document translation is not on mobile |
Our pick: DeepL
Why. DeepL, from DeepL SE in Cologne, Germany, is built for translation and nothing else, and it shows in the details professionals care about. Document translation returns your PDF, Word or PowerPoint file translated with the layout kept. Glossaries fix how a product name or technical term is always translated: one glossary with up to five entries per language pair on the free version, five glossaries with up to 1,000 entries per language pair on Team, and unlimited glossaries plus translation memory on Business and Enterprise. DeepL covers more than 100 languages, about 70 of which were added in beta for enterprise customers in November 2025.
DeepL also moved into speech this year. Voice-to-Voice launched on 16 April 2026, and on 15 September 2026 DeepL released Voice AI models that keep each speaker's voice recognisable in 14 languages, made voice-to-voice translation generally available in more than 30 languages, and added a Windows and Mac app that shows live translated subtitles in Zoom, Microsoft Teams and Google Meet. Since September DeepL also works inside Microsoft Copilot, ChatGPT and Claude through MCP, so an assistant can call DeepL and apply your glossaries.
Who it suits. Legal, HR, support and technical documentation teams, anyone translating files every week, and companies with a house style in several languages. Business adds files up to 100 MB and translation memory.
The catch. DeepL's privacy policy says texts sent to the free version, including your corrections, are processed for a limited time to train its models, and it describes no opt-out. The same policy says the free version must not be used for texts that contain personal data. Paid plans delete texts after translation and never use them for training. So the free version is fine for a menu or a public article, and wrong for client documents. The free limits are also tight: 50,000 characters and one file up to 5 MB a month. More in our DeepL guide.
Runner-up: ChatGPT
Why. ChatGPT, from OpenAI, is the better choice when a literal translation is not the goal. ChatGPT Translate at chatgpt.com/translate covers more than 40 languages and adds one-click options to make a translation more fluent, more professional or simpler. In a normal chat you can translate a headline, then ask for a version under eight words that works for German small business owners, and ask which idioms do not carry across. Replying to a foreign-language email, summarising three customer messages into a decision or drafting in two languages at once all happen in one conversation.
Who it suits. Marketers adapting copy for other markets, support teams replying to customers, and anyone who already pays for ChatGPT. If you also have a DeepL account, connect DeepL to ChatGPT as a custom connector and you get DeepL's translation and glossaries without leaving the chat.
The catch. ChatGPT has no glossary engine underneath. Paste a term list into the chat or a project's instructions and it will usually follow it, but usually is not always. On Free, Plus and Pro, Improve the model for everyone is on by default; switch it off on the web under your account menu, Settings, Data controls. Business and Enterprise content is not used for training by default. See our ChatGPT guide.
Best free option: Google Translate
Why. For whole files at no cost, Google Translate is the most generous option we could confirm. Its help centre says you can translate .docx, .pdf, .pptx and .xlsx documents up to 10 MB, with PDFs up to 300 pages, and you can translate a website by entering its URL under Websites. Compare that with DeepL's free version, which allows one document a month up to 5 MB.
Who it suits. Occasional translation of non-confidential material: supplier documents, research papers, a foreign website you need to read.
The catch. Text inside images and scanned PDF pages appears in the output but is not translated, and document translation is not available on mobile or smaller screens. Website translation is not supported in all regions. As with any free tool, read the terms before you upload anything confidential. For more free and paid options, see DeepL alternatives.
Other good options for specific translation jobs
- Speech on your phone: On 9 June 2026 Google brought Gemini 3.5 Live Translate to the Google Translate app on Android and iOS. It detects more than 70 languages and keeps the speaker's intonation, pacing and pitch.
- Translation plus careful rewriting: Claude from Anthropic is strong at adapting tone in long documents, and it can call DeepL through the Claude connector directory. Free Claude chats can be used for training unless you turn off Help improve our AI models under Settings, Privacy.
- EU data hosting: Vibe, Mistral AI's assistant formerly called Le Chat, keeps data in the EU by default, though individual plans train on your chats until you opt out. See our Vibe guide.
For a ready-made combination of translation, writing and meeting tools, see /stacks.
How to choose the right AI for translation
- Is it a file, a message or a conversation? Files with layout point to DeepL or Google Translate. Messages that need a reply or a rewrite point to ChatGPT or Claude. Live meetings point to DeepL Voice.
- Is the text confidential? If yes, use a paid DeepL plan, a business plan of ChatGPT or Claude, or switch training off before you paste anything. The free DeepL version has no switch, so keep client material away from it.
- Do the same terms have to come out the same way every time? If your product names, legal terms or brand phrases must never vary, you need DeepL glossaries, ideally with translation memory on Business.
A translation workflow from source file to approved copy
Here is a realistic job end to end: a 12-page English customer guide that needs to go out in German and French next week.
- Build the glossary first. List product names, feature names and terms that must stay fixed, and add them as a DeepL glossary for each language pair.
- Translate the file. Drop the .docx or .pdf into DeepL's document translator on a paid plan, choose German, and repeat for French. The files come back with their layout.
- Check the risky parts with a second tool. Paste the headings, warnings and anything legal into ChatGPT or Claude with the original, and ask it to flag mistranslations, missing sentences and inconsistent terms. Ask for a back-translation into English of anything it flags.
- Adapt the marketing bits. Headlines and calls to action often need a new idea, not a translation. Ask ChatGPT for three natural options per headline with a note on tone.
- Get a native speaker to read it. Anything public, legal or carrying your brand needs a human check. AI makes the reviewer's job faster; it does not replace it.
- Save what you learned. Add new fixed terms to the glossary, so the next document starts better.
The full step-by-step is in how to translate a document with AI, and you can learn DeepL from the basics on /learn.
Prompts for translation you can copy
Use these in ChatGPT, Claude or Copilot, with or without the DeepL connector. Replace everything in square brackets.
Prompts to copy
Translate with fixed terms
Translate the text below from [source language] into [target language] for [audience]. Use [formal or informal] address. Keep these terms exactly as written: [term list]. After the translation, list any sentence where you were unsure of the meaning. [paste text]
You get: A translation that respects your terminology, plus a list of doubtful sentences
Quality check a translation
Here is an original in [source language] and its translation in [target language]. Compare them line by line. Flag mistranslations, missing or added content, wrong numbers or dates, and inconsistent terms. For each issue give the original, the problem and a corrected version. Original: [paste] Translation: [paste]
You get: A reviewer's list of errors with fixes
Back-translation for risky text
Translate this [target language] text back into English as literally as possible, without smoothing it. Then tell me in plain English where the meaning differs from this original: [paste original].
You get: A literal back-translation that shows where meaning drifted
Localise a headline
This English headline is for [product] aimed at [audience]: [headline]. Give me three [target language] versions that sound like a native marketer wrote them, each under [8] words, and explain in one line what each one changes from the original idea.
You get: Natural headline options instead of a literal translation
Reply to a foreign-language email
Here is an email in [language]: [paste]. Summarise it in English in three bullet points. Then draft a polite reply in [language] that says [your points], in the same level of formality as the sender. Add an English version of the reply underneath so I can check it.
You get: An understood email and a checked reply in one go
Build a glossary from a document
Read the attached [document] and list every product name, technical term and recurring phrase that should be translated the same way every time into [target language]. Give a table with the term, a suggested translation, and whether it should stay in English.
You get: A ready glossary to load into DeepL