Translating a PDF used to mean either retyping every word into Google Translate or paying a translation agency. In 2026, machine translation handles most everyday cases in seconds — but knowing the trade-offs between different approaches saves you from a translated document that looks strange or, worse, says something different from the original.
Decide what kind of translation you need
Three rough cases, and they need different approaches:
- "I just want to understand the gist." Speed matters more than polish. Quick automatic translation is fine.
- "I need a usable translated document I can share." Quality and layout matter. Manual review is essential.
- "I need a legally or technically certified translation." Machine translation is a starting point, not the final answer. A human translator must review.
Picking the wrong approach for the situation produces either wasted effort or risky output.
Method 1: Browser-based translation that preserves layout
The cleanest first option for digital PDFs is a browser tool that translates while keeping the layout. Docento.app and several similar tools translate text in place — fonts, page positions, and images stay where they were.
This works best when:
- The PDF has a real text layer (born-digital, not scanned).
- The source language is well-supported.
- The translated text is roughly the same length as the original. German into English, for example, often expands; that can break tight layouts.
Method 2: Google Translate's PDF upload
Google Translate accepts PDFs directly. Upload, choose target language, get a translated PDF back. It's free, fast, and handles layout reasonably well.
Caveats:
- Documents up to 10 MB.
- The translated file goes through Google's servers — fine for public material, not for confidential documents.
- Translation quality is good but not human-grade. Don't use for legal or medical text.
Method 3: DeepL document translation
DeepL has the strongest machine translation for European languages and several others. Their document translation accepts PDFs and preserves formatting well.
Free tier has a small file limit; paid tiers handle larger documents and have better privacy commitments. Like Google, the file is uploaded — pay attention to your data sensitivity.
Method 4: Extract, translate, rebuild
For maximum control:
- Extract the text from the PDF. See how to convert a PDF to text.
- Translate the text in any tool.
- Rebuild the PDF in Word, LaTeX, or wherever the source lived.
This is the most labour but gives the cleanest result for important documents. The output respects the typography and layout conventions of the target language (different quote marks, different list styles, different paragraph spacing).
Method 5: Side-by-side bilingual PDF
A specialised case: instead of replacing the source language with the target language, present them side by side. Useful for:
- Language learners.
- Legal documents where both versions need to coexist.
- Translations that will be reviewed by a bilingual editor.
Most layout tools (LaTeX, InDesign, even Word) support two-column layouts with original and translation in parallel. Several specialised translation tools produce side-by-side output natively.
Translating scanned PDFs
If the PDF is a scan, you need OCR first:
- Run OCR to extract text.
- Translate the extracted text.
- Optionally, recreate the document in the target language.
Translation quality on OCR'd text inherits all the OCR errors. Reconcile any numerical data, names, and key terms manually before trusting the translation.
Handling specialised vocabulary
Machine translation falters on:
- Technical jargon that doesn't translate one-to-one.
- Legal terminology which is jurisdiction-specific.
- Proper nouns (people's names, place names) that can be incorrectly localised.
- Idioms that translate literally, losing meaning.
- Code, formulas, and identifiers that should not be translated at all.
For documents with much specialised vocabulary, build a glossary first. DeepL Pro, Google Translate, and most translation memory tools support glossaries that ensure specific terms are translated consistently.
Right-to-left and bidirectional text
Translating into Arabic, Hebrew, Persian, or Urdu introduces right-to-left layout issues:
- Margins flip.
- Page numbers move from one corner to the opposite.
- Mixed RTL/LTR content (English code blocks inside Arabic prose) needs explicit direction markers.
Most modern translation tools handle this correctly, but always check the result. Re-typesetting in the target language is sometimes the only way to get a clean RTL document.
Quality control
After any machine translation:
- Spot-check key paragraphs by translating back to the original language. If the round-trip isn't roughly equivalent, the translation may be off.
- Check numbers and dates — these can be silently mangled. 1,000 in English vs 1.000 in German vs 1 000 in French.
- Verify proper nouns — names of people, places, companies should usually not change.
- Read like a native speaker would, ideally with a native speaker's help, before sending the document anywhere consequential.
Privacy considerations
Most cloud translation services log the documents they translate, and some use them for training. For confidential documents:
- Use a tool with a clear no-training policy (DeepL Pro, paid Google Cloud Translation).
- Use an on-device translation tool. Docento.app processes locally and doesn't transmit the file.
- For maximum privacy, run translation locally with an open-source model.
For documents bound by GDPR, HIPAA, or similar regulations, machine translation services often aren't compliant. Get explicit guidance from your data team before uploading.
When to use a human translator
Machine translation is fine for:
- Internal communication.
- Drafts that will be reviewed.
- Personal use.
- Casual reading.
It's not enough for:
- Legal documents (contracts, court filings, immigration papers).
- Medical records.
- Marketing copy meant for native speakers.
- Anything where ambiguity has consequences.
Use a professional translator for these, with machine translation as a starting point if you want.
Conclusion
Machine translation handles most everyday PDFs well. Pick the tool based on language pair, document sensitivity, and how polished the output needs to be. Docento.app handles browser-based translation without uploading. For documents that matter, pair machine translation with a human review pass, and budget time for layout adjustment in the target language.