Can ChatGPT translate a PDF? Short ones yes, long ones not in full

Short answer: yes for a short, digital PDF. For a long document, not reliably in full: a chatbot tends to translate the beginning, then summarize, skip sections or stop. Scanned PDFs are a second weak spot, and you usually get text in a chat, not a file.

What a chatbot does well

Translation is one of the things ChatGPT, Claude and Gemini are good at. Give one of them a two-page letter as a digital PDF and ask for English, and the result is usually accurate and reads naturally. You can also ask follow-up questions: what a clause means, or whether a phrase is formal. For a short document you only need to understand, a chatbot is a good choice.

Where it goes wrong

The quality of the translation is rarely the issue. Completeness is.

• Summary instead of translation. Faced with a 50-page document, an assistant often translates the first pages and then moves to a condensed version. The switch is not always announced.
• Stopping partway. A reply has a practical length limit. The translation ends mid-document and you have to ask it to continue, again and again, and keep track of where each part ended.
• Skipped parts. Tables, footnotes, annexes and small print are the parts most likely to be left out.
• Scanned PDFs. A scan contains pictures of pages, not text. Depending on the assistant, it is read with errors or not read at all.
• Changed details. Names, amounts and reference numbers are sometimes translated or reformatted when they should be left alone.
• No file. The translation usually arrives in the chat window, and turning it into a document is up to you.

For a contract or an official letter, the missing paragraph is the one that will matter.

Why this happens

A chat assistant answers in one reply at a time, and a long document does not fit in one reply. When the full translation is too long, the model has to choose between stopping and compressing, and both look like a finished answer. It is not being careless. It is doing what a conversation normally calls for, which is a response of reasonable length.

A scanned PDF is a separate problem. If the file has no text layer, there is nothing to translate until someone reads the page images, and not every assistant does that for every page.

How to test it yourself

Pick a PDF of at least ten pages in a language you can partly follow.

1. Ask a chatbot to translate the whole document, in full, without summarizing.
2. Check the last page of the original. Is its last paragraph in the translation?
3. Count the headings or numbered sections in both versions.
4. Find a table or a footnote in the middle and look for it in the translation.
5. Compare three names and three amounts.

If everything is there, the chatbot handled your document. If the ending is missing or the section count is lower, you have a partial translation.

What to use instead

Use a tool that works through the document one page at a time. Our PDF translator turns every page into an image, reads it and translates it separately, so page 40 is treated the same as page 1 and scanned PDFs work like digital ones. It is instructed to translate everything and never to summarize or skip, to keep headings, paragraphs, lists and table rows as structure, and to leave names, numbers and dates unchanged. There are 28 target languages. The translation appears beside the original page, you can edit it there, and you download the whole document as Word or TXT.

Two things it does not do. It does not rebuild the original layout: you get the translated text in a Word file, not a look-alike PDF with the fonts and images in place. And it is machine translation: fine for understanding a document or preparing a draft, but for legal or official use a certified human translation may be required, so ask the office that will receive it.

The first page of every document is free and needs no account, so you can compare it with a chatbot on your own file. A whole document is $5, paid once.