How-to

How to translate a scanned PDF file

Scanned PDFs are images of text, not actual text — which is why most translators including Google Translate either reject them, return an empty result, or show a "can't translate this file" error. To translate a scanned PDF you need OCR (text extraction) before translation. DocTranslating runs OCR automatically as part of the translation pipeline, supports 100+ languages, and rebuilds the translated text into a copy of the original PDF. For accuracy on important documents, verify the OCR output first on PDFEquips so extraction errors don't compound with translation errors.

Updated June 5, 2026 · 8 min read

If you've ever uploaded a scanned PDF to a free translator and gotten back an empty file, a "can't translate this file" error, or a translated copy with all the text missing — you're not doing anything wrong. Most online translators, including the free document upload in Google Translate, don't run OCR on scanned content. This guide explains why that happens, what you actually need to translate a scanned PDF, and how to do it without losing the original layout.

Why scanned PDFs won't translate normally

A normal PDF — one exported from Word, an editor, or a browser — has a hidden text layer that translators read directly. A scanned PDF doesn't. When you scan a document, your scanner or phone camera captures a picture of each page. The result looks like text, but to a computer it's just an image — there's nothing extractable underneath. That's why selecting text in a scanned PDF usually doesn't work either: there are no characters to select, only pixels.

Most translation tools assume the text layer is already there. When they don't find it, they fail in confusing ways. Common symptoms include:

• The translator returns an empty file, or a copy that looks identical to the original.
• You get a "can't translate this file" or "unable to translate this document" message.
• Only digitally-embedded elements (page numbers, watermarks, form fields) get translated.
• The download button stays greyed out, or the job appears to finish but produces nothing usable.
• The same file works in one tool but fails in another, with no clear explanation.

What you actually need: OCR + translation

Translating a scanned PDF is a two-step process under the hood, even when a single tool handles both:

1. OCR reads each page image and extracts the recognisable text — words, numbers, and basic layout.
2. Translation takes that extracted text, translates it, and writes it back into a copy of the document.

DocTranslating runs both steps automatically when you upload a scanned PDF — you don't need to OCR it yourself first. The catch worth understanding upfront: translation quality can only ever be as good as the OCR that feeds it. A blurry scan produces blurry OCR, and blurry OCR plus translation compounds the errors. The result can look fluent and still be subtly wrong, so important documents are worth verifying before relying on them.

Step-by-step: translate a scanned PDF

1. 1

   Open DocTranslating and upload your scanned PDF

   Drag the file onto the upload area, or click to browse. The tool detects the file is a PDF; you don't need to do anything special to flag it as scanned — OCR runs automatically when needed.

2. 2

   Set your source and target languages

   Pick the language the document is written in and the language you want to translate it into. For scanned PDFs, set the source language explicitly rather than relying on auto-detect — auto-detection is less reliable on OCR'd text than on clean text.

3. 3

   Choose the Gemini engine

   For scanned PDFs, Gemini is the strongest choice. It's LLM-based, so it uses surrounding context to infer meaning when OCR produces partially garbled words, while sentence-level engines like DeepL pass garbled words through unchanged. You can also write custom instructions to keep terminology consistent across the document.

4. 4

   Translate, then review the result carefully

   Start the translation, download the file when it's ready, and compare it page-by-page with the original. Pay special attention to numbers, dates, proper nouns, addresses and anything legally important — these are where OCR errors typically hide because they don't have surrounding context the translator can use to self-correct.

Which translation engine is best for scanned PDFs?

All DocTranslating engines that accept PDFs run OCR on scanned content, but they handle imperfect OCR output very differently. No OCR is 100% accurate — the real question is how the translator copes when it sees a partially garbled word.

Improving OCR before you translate

OCR quality depends almost entirely on the input. A clean, properly-rotated scan at decent resolution produces near-perfect OCR; a faded, skewed, low-resolution scan produces unreliable OCR no matter which tool runs it. A few things worth doing before you upload:

• Rescan at 300 DPI or higher if you have access to the physical document. Lower resolutions blur characters and OCR misreads them.
• Straighten skewed pages — OCR engines expect text on horizontal lines.
• Increase contrast on faded or grey scans so characters stand out from the background.
• Confirm the file isn't password-protected — encrypted PDFs can't be read until decrypted.
• Set the source language explicitly, especially for non-Latin scripts (Arabic, Chinese, Cyrillic, Devanagari). Auto-detection on OCR'd text is much less reliable than on clean text.

Edge cases and current limitations

Handwritten documents

OCR for printed text is mature and reliable. OCR for handwritten text is much harder, and results are inconsistent across the whole industry — not just one tool. If your scanned PDF is handwritten, expect significant manual cleanup, and for anything legally sensitive prefer manual transcription over machine OCR.

Large or long scans

The Gemini engine caps each file at 25 pages and 100 MB. Longer or larger scans need a workaround:

Scanned PDFs in right-to-left languages

If you're translating a scanned PDF written in Arabic, Hebrew or Persian, there is a current limitation worth knowing: the PDF text-extraction layer can return RTL content in visual draw order rather than logical reading order, which means even OCR'd words can come out scrambled before translation starts. RTL Word and PowerPoint files work fine, and translating into an RTL language works fine — it's RTL PDF sources that are affected. If you have access to the original editable file, translate that instead. Otherwise this is being worked on but isn't solved yet.