Use AI for · Translation · 26 Sep 2026
Use AI for
AI translation

AI translation for business:faster in other languages,no errors going out

With AI, the foreign-language version of an email, a product description or a manual takes minutes. This guide covers which tools fit which job, how to instruct the AI, where it tends to go wrong and what data protection law and the EU AI Act expect from you.

Get in touch
On this page
Enquiry
Nikolai Schöbel und Jeremias Burger, Co-Founder Scalableloops

Let's talk about your project.

First we check whether the project fits your business model. Then you get a proposal with phases and effort.

Discuss AI translation or call: +49 151 1576 5566
Use AI for · Translation

AI translation for business means using a language model to turn a text into another language, producing a readable draft in seconds. It works well for emails, quotes, product copy and internal documents, provided two things are in place: the data goes to a provider under a business contract, and someone fluent in the target language checks everything before it leaves the company. For official documents, AI does not replace a certified translation.

In brief
  • Dedicated translators such as DeepL and general AI chats such as ChatGPT, Claude or Gemini suit different kinds of work.
  • According to DeepL, texts sent to DeepL Pro are deleted after translation and never used for training; the free version works differently.
  • A glossary of your terminology and a clear prompt naming audience and tone make a noticeable difference to the result.
  • Anything that reaches customers or authorities is checked by a person who knows the language. Official documents still need a sworn or authorised translator.
Published 26 Sep 2026Nikolai Schöbel and Jeremias Burger9 min read
Nikolai SchöbelJeremias Burger

Nikolai Schöbel and Jeremias Burger

Co-founders of Scalableloops GmbH. Nikolai Schöbel leads online marketing and AI strategy, Jeremias Burger the AI architecture. Both build AI systems and train teams on them in their own agency work.

On this page
  1. What can AI translation do for a business?
  2. Which tools work for AI translation at work?
  3. How do you translate with AI step by step?
  4. What does a good prompt for AI translation look like?
  5. Where does AI translation typically go wrong?
  6. How do you make sure an AI translation is good enough?
  7. What data can you put into an AI translator?
  8. What does the EU AI Act require when you translate with AI?
  9. Frequently asked questions
  10. How to get started with AI translation
  11. Where the information on this page comes from
Overview

What can AI translation do for a business?

An AI translator is built on a language model that grasps the meaning and context of a sentence and rewrites it in the target language. Unlike the word-by-word tools of the past, it recognises idioms, adjusts sentence structure and can change the tone on request, for example making a text more formal or simpler.

Typical tasks include replying to an overseas customer, producing an English version of a quote, writing product descriptions for an online shop that sells in several countries, translating operating instructions, preparing training material for staff whose first language is not yours, and quickly understanding a letter in a foreign language.

What you get is a draft, not a checked translation. The AI can pick the wrong technical term, mix up figures and units, or produce a sentence that reads smoothly but changes the meaning. So for every translation that speaks for your company, one rule applies: a person who knows the target language reads it against the original.

Tools

Which tools work for AI translation at work?

You will come across two kinds. Dedicated translators such as DeepL are built for translation. According to DeepL, DeepL Pro translates whole files, including Word, PowerPoint, Excel and PDF files, and offers glossaries that fix how specific terms are translated.

General AI chats such as ChatGPT, Claude or Gemini also translate when you ask them to. Their strength is that you can add further requests to the same prompt: adjust the tone, rewrite for a particular audience, shorten the text at the same time. In its documentation for Claude, Anthropic notes that performance varies between languages and recommends stating the source and target language explicitly.

Which kind you choose depends less on translation quality in a single case than on how you work. If you translate many files with fixed terminology, document translation and a glossary help most. If you want to translate and adapt at once, a chat is often more convenient. If your team already uses an AI assistant, our guide to the AI assistant for business explains which business plans exist.

FeatureDedicated translatorGeneral AI chat
ExamplesDeepL, DeepL ProChatGPT, Claude, Gemini
StrengthTranslates whole files and largely keeps the layoutTranslates while rewriting, shortening or adjusting tone
Fixing terminologyVia a glossaryVia a term list in the prompt
Typical useManuals, data sheets, draft contracts, presentationsEmails, marketing copy, social posts, summaries of foreign texts
Data protectionFor company data, the paid version with a data processing agreementFor company data, only a business account with a data processing agreement
Process

How do you translate with AI step by step?

Good translations start before the first click. The more clearly the AI knows who the text is for and which terms are fixed, the less you have to correct afterwards. This process works well for business texts.

  1. 01

    01

    Clarify the purpose

    Does the text only need to be understood, or is it going to customers, partners or authorities? A raw draft is enough for understanding. Anything external needs the steps below.

  2. 02

    02

    Prepare the source text

    Short, unambiguous sentences translate better. Remove internal notes and any personal data the translation does not need.

  3. 03

    03

    Fix your terminology

    Keep a list of technical terms, product names and expressions that must not be translated. In a dedicated translator it becomes a glossary, in a chat part of the prompt.

  4. 04

    04

    Write the prompt

    State source and target language, audience, tone, form of address and the term list. Ask the AI to flag unclear passages instead of guessing.

  5. 05

    05

    Have it reviewed

    Someone confident in the target language compares the translation with the original: figures, units, names, commitments and technical terms.

  6. 06

    06

    Keep improving

    Add corrections to your term list. Each translation then gets better, and your terminology stays consistent across all texts.

Prompts

What does a good prompt for AI translation look like?

A good prompt names the language, audience, tone and fixed terms. For a customer email going to Germany it might read: “Translate the following email from British English into German. The recipient is a purchasing manager we address formally, so use ‘Sie’. Tone: friendly and factual. Do not translate product names. Always use ‘Angebot’ for ‘quote’. Put square brackets around anything you are unsure about.”

For product copy in an online shop, mention the market: “Translate this product description into French for customers in France. Keep all dimensions and weights unchanged. Write naturally rather than literally and keep the bullet list.” Our guide to using AI for writing covers how to build product copy with AI, and our guide to using AI for ecommerce covers multilingual shops.

To understand incoming post quickly, this is often enough: “Summarise this letter in English in five sentences and list every deadline, amount and demand.” For the reply, the same rule applies again: draft by AI, check by a person.

After translating, it pays to ask the AI a follow-up question: “List every place where the translation departs from the original or where a term could be read in more than one way.” That does not replace a human review, but it shows you where to look more closely.

Limits

Where does AI translation typically go wrong?

The most common errors are hard to spot because the text reads fluently. A mistranslated technical term, a swapped unit or a lost negation does not stand out on a quick read. Pay particular attention to figures, date formats, names, commitments and legal wording.

Culture and forms of address are another source of error. Whether a customer expects a formal or informal tone, which courtesies are usual and which examples make sense in the target market: the AI only gets these right if you tell it.

For some texts, AI is simply the wrong route. In Germany, for instance, courts, notaries and authorities often require a certified translation of official documents, and only a translator who has been generally sworn in or authorised by a court or authority may produce it. Contracts, liability wording and safety instructions should also be checked by specialists before the translated version takes effect.

Quality

How do you make sure an AI translation is good enough?

The translation industry has its own standards for this. ISO 17100 sets out requirements for translation services and requires a second qualified person to compare the translation with the source text, often called the four-eyes principle. ISO 18587 covers full post-editing, meaning a human revises machine translation output with the aim of a result comparable to a human translation.

You do not have to meet these standards in-house, but the idea is a useful benchmark: the machine provides the draft, and a person who knows the language checks it against the original. If you commission a translation agency, ask whether it uses machine pre-translation and how it post-edits.

A short checklist helps with your own review: Are all figures, units and names correct? Were the glossary terms used? Do form of address and tone suit the recipient? Is a sentence missing, or has one been added? Would a native speaker write it this way?

Data protection

What data can you put into an AI translator?

Personal data and trade secrets only belong in tools covered by a contract. If a provider processes data on your behalf, the GDPR requires a data processing agreement under Article 28. The German data protection authorities (Datenschutzkonferenz) also recommend that employers provide work accounts rather than letting staff use personal ones.

DeepL shows how big the gap between free and paid versions can be. Its privacy policy states that content from the free version is processed for a limited period to train and improve its neural networks. For DeepL Pro, the provider says texts are not stored permanently, are deleted once the service has been performed and are never used for training. For transferring personal data, the Pro terms provide for a data processing agreement.

The same pattern applies to general AI chats: business account rather than personal account, training on your inputs switched off, contract in place. If you translate particularly sensitive documents, you can also run a language model on your own hardware. Our guide to local AI for business explains what that involves.

Obligations

What does the EU AI Act require when you translate with AI?

Under Article 4 of the EU AI Act, organisations that use AI tools have had to take measures to support AI literacy among their staff since 2 February 2025. Since 27 July 2026, Article 4 has applied as amended by Regulation (EU) 2026/1744: you must take measures that support the development of AI literacy, but you do not have to guarantee a specific level of competence for any individual. No certificate is required. More on this in our article on the AI literacy obligation under Article 4.

Article 50 and its transparency obligations have applied since 2 August 2026. Paragraph 4 is the relevant part for translation: anyone who publishes AI-generated text to inform the public on matters of public interest must disclose that it was generated by AI. The obligation does not apply if the text has undergone human review or editorial control and a person or company holds editorial responsibility. A careful review therefore pays off twice.

If you use translation in customer contact, for example in a multilingual chat, separate disclosure rules apply. Our guide to using AI for customer service covers how to handle multilingual enquiries with AI.

Frequently asked questions

Frequently asked questions

Which AI translator is suitable for a business?

One that covers your tasks and can be used under a business contract. For many files with fixed terminology, a dedicated translator such as DeepL Pro with a glossary fits well. If you want to translate and rewrite at the same time, use an AI chat such as ChatGPT, Claude or Gemini on a business account.

Can DeepL be used in line with the GDPR?

For company data, the paid version is the one to consider. According to DeepL, texts sent to DeepL Pro are deleted after translation and not used for training, and a data processing agreement is provided for personal data. In the free version, DeepL processes content for a period to train its models. Whether your use is compliant overall is a question for your data protection officer.

Can I translate contracts with AI?

As a working draft and to understand the content, yes. Before a translated contract takes effect, a specialist who knows both languages and the area of law should check it. If an authority or court requires a certified translation, a sworn or authorised translator has to produce it.

How do I teach the AI our terminology?

With a term list. In dedicated translators such as DeepL you create a glossary that fixes how a term is translated. In an AI chat you include the list in your prompt. Add corrections from each review to the list so they apply next time.

Do I have to label AI translations?

Not in general. Article 50(4) of the EU AI Act requires disclosure for AI-generated text published to inform the public on matters of public interest. The obligation does not apply if a person has reviewed the text and someone holds editorial responsibility.

What is post-editing?

Post-editing is when a person revises a machine translation. The standard ISO 18587 describes full post-editing, which aims for a result comparable to a human translation.

What now

How to get started with AI translation

  1. 01

    List your texts

    Write down which texts your team regularly translates or just needs to understand, and who handles them today.

  2. 02

    Sort out tool and contract

    Choose a dedicated translator or an AI chat on a business account, sign the data processing agreement and switch off training on your data.

  3. 03

    Build a term list

    Collect technical terms, product names and fixed phrases, and decide who checks translations before they go out.

  4. 04

    Prepare your team

    Practise prompts and reviews on real texts from your daily work. We can help with AI training for your team.

AI translation takes the first draft off your team's plate and makes foreign-language post understandable in seconds. It becomes reliable once you have a business contract, a maintained term list and a person who reads everything before it goes out.

or call: +49 151 1576 5566

Further reading
Sources

Where the information on this page comes from

Projekt-Detail

Problem

Our solution

Our role

Highlights

    Got a project in mind?

    We reply personally. First a use-case check, then an architecture proposal.

    Start your inquiry