ChatGPT translation is good enough to see what a text says and to produce a usable first version, but not good enough to publish without checking. The quality depends mainly on the context, audience and terminology you give it.
You can translate in two ways: through the separate translation page OpenAI launched in January 2026, or in an ordinary conversation. The two routes solve different problems.
When it goes wrong, it goes wrong in four predictable places: technical terms used in the wrong sense, terminology that drifts across a long text, errors written with total confidence, and style that slips. Those patterns also explain how ChatGPT differs from a translation engine such as DeepL.
Further down you will find four instructions that make the translation measurably better, and the point at which you are better off hiring a translator.
How to translate with ChatGPT
Translating with ChatGPT works through two routes: the separate translation page and an ordinary conversation. The translation page is faster for standalone texts. The conversation is stronger as soon as you want to steer the result or are working with a longer document.
What is ChatGPT Translate?
ChatGPT’s translation page is called ChatGPT Translate. It lives at chatgpt.com/translate and works like a regular translation tool. According to the industry publication Slator, OpenAI launched the page on 15 January 2026. On the page itself, OpenAI mentions support for more than forty languages.
You paste the text, dictate it or take a photo of it. The model detects the source language and you choose the target language. The difference from a classic translation engine lies in the presets: you can make the translation sound more businesslike, make it more fluent, aim it at an academic audience or have it simplified as if you were explaining it to a child.
For a single paragraph, an email or a product description, this is the fastest route. No prompt is needed, because the instructions are already built into the buttons.
When is it better to translate in a ChatGPT conversation?
It is better to translate in a ChatGPT conversation as soon as you want to steer the outcome, enforce a fixed glossary or work with a text so long that it has to go in sections. You paste the text with an instruction above it.
That route takes more effort and gives you more control, because you decide the tone, the terminology and how borderline cases are handled.
How good is ChatGPT translation?
At sentence level, ChatGPT often translates more fluently than a classic translation engine, because it takes register and audience into account instead of just converting words. An English passive construction, for example, more often becomes active in Dutch, which is exactly where machine translations tend to stay stiff.
At document level, the picture changes. The longer the text, the greater the chance that terminology and tone start to drift, because the model works piece by piece rather than with a translation memory that holds on to earlier choices.
In my own work with both models, Claude loses less context in long documents than ChatGPT, and that difference shows most in translation work of any length.
The honest summary: for understanding, preparation and internal use, the quality is more than adequate. For text that goes out under your name, the translation is a draft, not a finished product.
Where does ChatGPT translation go wrong?
When you translate with ChatGPT, four things go wrong systematically, and all four are easy to spot once you know what to look for.
The first is a technical term in the wrong sense. The Dutch word termijn becomes instalment in a payment context and deadline in a planning context. A model that is not given the context picks the form that occurs most often statistically, instead of the form that is right for your document.
The second is drifting terminology. If you translate a fifteen-page manual in sections, the same component suddenly gets a different name halfway through, because there is no glossary holding the choice in place.
The third is nonsense written with confidence. Names of institutions, legal articles and quotations are sometimes altered into something that sounds plausible in the target language, a phenomenon known as AI hallucination. In translation work it is harder to spot than in ordinary text, because the sentence still reads smoothly.
The fourth is style drift. Over a longer session the tone slides back towards the average, so the businesslike instruction from your first message has less effect in message twelve than in message two.
ChatGPT vs DeepL: what is the difference?
ChatGPT and DeepL differ in what they are. DeepL is a translation engine that converts a text. ChatGPT is a language model that you give an assignment and can steer.
That difference determines where each is strongest. A translation engine delivers a predictable, consistent result and works with terminology lists and translation memories, which counts for a lot in repetitive work such as product catalogues. A language model delivers an adjustable result and wins as soon as tone, audience or a tricky passage call for some back and forth.
How do you get a better translation from ChatGPT?
You get a better translation from ChatGPT with four instructions, and none of them takes more than half a minute.
First, name the audience and the register. A translation for a client in Flanders reads differently from one for a buyer in the US, and a model that is not told this chooses a neutral middle ground that works well for nobody.
Next, include a glossary. A list of ten to twenty technical terms with their fixed translations, pasted above the text, solves most terminology problems in one go. A glossary like this is the manual version of what a translation agency calls a term base.
Then work in sections and repeat the instruction for each section. A paragraph or two at a time, with the glossary attached, keeps the quality more stable than five pages in one go.
For tricky passages, ask for two variants, and have the translation back-translated into the source language when the stakes are high. If the back-translation differs from your original, that is where your error is.
When is it better to hire a translator?
Translating with ChatGPT stops making sense where the consequences of an error outweigh the time saved. The line is not a matter of principle but of weighing risk, and in four cases the answer is always the same.
Sworn translations are out, because they need a certification that no model can provide. Legal and medical texts are out, because a single wrongly chosen term changes the meaning and the liability. Marketing copy with hard claims is out, because a claim that holds in the source language may carry different legal weight in the target language. And literary or humorous text needs a translator, because rhyme, rhythm and wordplay rarely survive a direct translation.
Outside those four, it is not a question of either/or but of order. A translator who post-edits what a model has prepared delivers the same result faster, and that is now a large part of the translation profession.
In my own way of working, human editing is not an optional step but part of the workflow, and in translation work that applies twice over: you are checking a text in a language that, by definition, you command less well than the source language.
Also watch what you paste. A client file or a quote in a personal account is a different matter from the same text in a business environment.
What does ChatGPT translation mean for your work?
Translating with ChatGPT changes three things about how translation work runs in an office. The first is the barrier to entry: a first version of an English quote or a German product page now takes minutes instead of an external assignment, which means more gets translated, not less.
The second is the checking step. Context, technical terms and tone are not extras but the instructions that decide whether the outcome is usable, and the four failure patterns above are exactly what the check should focus on.
The third is the agreement within the team. Leave this to run without agreements and you get translations of varying quality in customer communication, with terminology that differs per employee. A fixed way of working prevents that, and it belongs in the same framework as the rest of using ChatGPT for business.
I look at AI language output with the same question every time: can this go out under my own name? For translations the answer is yes more and more often, and still never without someone having looked at it. LearnLLM’s ChatGPT Course covers that way of working at your own pace, with a certificate on completion.
Frequently asked questions about ChatGPT translation
Questions about ChatGPT translation often also concern what happens to your text and how to translate a complete file.
Does OpenAI use your translated texts to train its models?
With personal accounts, OpenAI may use the texts you enter, including texts you have translated, to train its models unless you turn this off. You switch it off in the settings under Data Controls with the option Improve the model for everyone. OpenAI does not use a temporary chat for training and deletes it within 30 days. On the business plans, such as ChatGPT Enterprise, OpenAI does not train on your data by default.
Can ChatGPT translate a whole document?
Yes, you can have ChatGPT translate a whole document by uploading the file with the plus button in the chat bar instead of pasting the text. ChatGPT accepts common text files, PDFs, Word documents and presentations. According to OpenAI’s help centre, the upload feature does not support a PDF that consists only of scanned images, so convert a scan like that to text first.
