How to Preserve Formatting When Translating Word, PDF, and Excel Documents

You get back a translated 30-page contract and it’s a mess: the payment terms table has collapsed across the page, section headings are gone, some columns are overflowing into each other. The translation itself is accurate - but the document looks like it went through a blender. That’s 3-4 hours of manual cleanup ahead of you. This is the classic result when a translation tool doesn’t understand formatting. Let’s break down why it happens and how to stop it for good.

Why formatting breaks: the root cause¶

Most translation tools - including Google Translate - are architecturally built the same way: extract text → translate → put it back. Formatting either gets lost in step one, or gets reconstructed “approximately” with no understanding of the original layout.

There’s a second problem that’s often overlooked: text expansion during translation. According to Eriksen Translations, translating from English into German produces 30-35% text expansion. Text that filled two lines in English will need 2.5-3 lines in German. In a table with fixed column widths, that’s an overflow - text spills outside the cell or gets cut off.

Three core reasons formatting breaks:

• The tool extracts flat text - with no information about styles, fonts, or element positions
• Translation changes text length - by 20-35% in either direction depending on the language pair
• Layout reconstruction happens automatically - and the algorithm doesn’t know the original designer’s intent

Let’s go through each format.

Word (.docx): where things go wrong and how to fix it¶

Word is the most translation-friendly format - if you handle it right.

Why Google Translate breaks Word documents¶

When you upload a .docx to Google Translate, it pulls out all the text, translates it, and tries to reassemble the document structure. For simple, single-column documents without tables - this occasionally works acceptably. For real-world documents - it doesn’t.

What specifically breaks:

• Styles (Heading 1, Heading 2, Normal, Caption) - often reset to defaults or get mixed up
• Tables with merged cells - the cell structure breaks, merges disappear
• Text boxes and frames - text overflows boundaries or vanishes entirely
• Numbered lists and bullet points - numbering can go out of sequence or duplicate
• Headers and footers - sometimes lost or duplicated into the body

DeepL: the best mass-market tool for Word¶

DeepL is one of the few mainstream tools that treats Word as a structured document rather than a flat text string. Their algorithms preserve fonts, tables, images with captions, hyperlinks, numbered lists, and footnotes.

As DeepL describes in their blog:

Rather than merely preserving the existing document structure, this new method observes what the layout is, gathers detailed information about it, stores this data, and then uses it alongside the extracted text to completely reconstruct the document.

DeepL doesn’t just “paste translated text back” - it rebuilds the document from scratch using the original structure as a blueprint. That’s a meaningful architectural difference.

Free tier: up to 5 MB and 3 documents per month. DeepL Pro starts at $9.99/month, removes limits, and adds glossary support.

CAT tools: the gold standard for formatting preservation¶

CAT tools (Computer-Assisted Translation) - memoQ, SDL Trados Studio, Smartcat, Wordfast - are a separate class of software used by professional translators. They open a .docx, split it into segments (sentences or paragraphs), and let the translator work on text only. Formatting stays completely untouched in a separate layer.

As Smartcat explains:

CAT tools extract translatable content from complex file types (Word, InDesign, XML, HTML, and many others) while preserving the original formatting, so translators work with text, not code.

The translator sees only words - the layout stays frozen and automatically reconstructs when the translation is finalized. This works reliably for documents of any complexity.

The downside: CAT tools have a learning curve. If you translate regularly or in high volume, they’re worth it. For a one-off document, DeepL is simpler.

Preparing your Word file before translation: a checklist¶

Most people skip this stage - and then wonder why formatting breaks even with a good tool.

Do this before sending for translation:

• Accept all tracked changes - open “Review” → “Accept All Changes”. If you leave Track Changes on, the tool sees both the original and the revised text at once
• Use named styles, not manual formatting - instead of “bold 16pt font” use the “Heading 2” style. Tools understand styles far better than direct formatting
• Don’t embed translatable text in images - text inside a picture won’t be translated and can’t preserve formatting. Captions on diagrams should be Word text elements, not part of a PNG
• Fix table column widths - if your table has “auto width” columns, they’ll distort when text length changes after translation. Set fixed widths where precision matters
• Check font embedding - if the document uses non-standard fonts, you may have rendering issues when the file is opened by the translation tool or the recipient

PDF: the hardest format to translate with formatting intact¶

PDF is a frozen document. Unlike Word, it doesn’t have “layers” - text, images, and formatting are baked into a single file. That’s why translating PDF while keeping the layout is genuinely hard.

Two types of PDF - and why this matters¶

Not all PDFs are the same.

Text-based PDF (digital-native) - created from Word, InDesign, or another editor. The text is real - you can highlight and copy it. This is much easier to translate.

Scanned PDF - a set of images. If you scanned a paper document and saved it as PDF, there’s no actual “text” inside, just pictures. Before translating, this file needs to go through OCR (optical character recognition).

How to tell the difference: open the PDF and try to select text with your mouse. If it highlights correctly - it’s a text PDF. If nothing highlights, or everything highlights as one undifferentiated block - it’s a scan.

For more on handling scanned documents, see the dedicated article on translating scanned PDFs and photos.

How to translate a text PDF with layout preservation¶

The ideal path: find the original .docx or .indd file and translate that instead of the PDF. If the PDF was created from Word - ask for the original. Translating Word is easier and the result is better in every way.

If you don’t have the original: you need a tool that can disassemble the PDF into layers.

Here’s why Google Translate for PDF is a bad idea. As Pairaphrase explains in their analysis:

Google Translate extracts only the text and discards layout information, so when it reassembles the document, formatting is lost or approximated. Multi-column documents collapse into single columns. Tables lose their cell boundaries. Images and page headers disappear without a trace.

You get a translation, but without tables, columns, or proper element positioning. For personal understanding - fine. For a document you need to submit somewhere or show a client - not fine.

PDF translation tool comparison:

Edge case: PDFs with complex layouts¶

If a PDF has two-column layouts, charts with labels, or complex tables with merged cells - even good tools won’t get it 100% right. The PDF format doesn’t understand document semantics - it just stores element positions as coordinates on a page.

Practical tip for these documents: convert the PDF to Word (using Adobe Acrobat, ilovepdf, or Smallpdf), review the conversion result, fix obvious artifacts - and then translate the Word file. It adds a step, but the result is far more reliable.

Excel (.xlsx): formulas and formatting are a separate problem¶

Excel isn’t just tables with text. It’s formulas, named ranges, conditional formatting, pivot tables. And translation breaks here differently than in Word or PDF.

Two independent tasks in Excel translation¶

Translating an Excel file involves two completely separate jobs:

Job 1: Translate text content - column headers, row labels, text in cells. Standard translation task.

Job 2: Convert function names - if an Excel file is used in a different language locale, function names may differ. In English Excel it’s =VLOOKUP(...), in German it’s =SVERWEIS(...). Same function, different name.

Important: don’t translate formulas like regular text. That will break the spreadsheet.

Microsoft Functions Translator: a dedicated add-in for formulas¶

Microsoft Functions Translator is a free Excel add-in that converts function names between 80 languages. =SUM becomes =SUMME (German), =SOMME (French), and so on, while the function arguments stay unchanged.

Install it through Insert → Add-ins → search “Functions Translator” in Excel. Free.

What you can and can’t translate in Excel¶

Classic mistake: someone “translates” an Excel file by copying everything into Google Translate and pasting back. The result: =IF(A1>10, "Yes", "No") becomes something like a localized version where the function name is translated as plain text - and Excel throws an error because it doesn’t recognize the translated word as a built-in function.

Excel formatting: what survives and what doesn’t¶

Most CAT tools and DeepL preserve cell formatting when working with .xlsx - colors, fonts, column widths, merged cells. But there are edge cases:

• Conditional formatting - if rules are tied to specific rows/columns, they can break after rows are added or shifted
• Pivot tables - the underlying data should survive, but check field names and headers
• Charts with labels - axis labels and legends may require manual translation

Tool comparison: Word, PDF, Excel¶

When you need an online service with a human translator¶

If the document is official - a contract, medical records, a translation for an embassy or university - good formatting preservation alone isn’t enough. You need human verification and often a certified translator’s signature.

ChatsControl is an online certified translation service where AI produces a draft and a sworn translator reviews and signs it. You upload the document and get back a finished PDF with preserved structure and the translator’s signature in 2-4 hours. It works well for official documents where both legal certification and a clean final file matter.

Common mistakes and how to avoid them¶

Mistake 1: translating a screenshot or photo instead of the source file¶

If someone sends you a “Word document” as a screenshot or photo - no tool can preserve formatting, because there’s nothing to preserve. Always ask for the source .docx, .xlsx, or .pptx.

Mistake 2: saving to PDF before translating¶

“I’ll convert it to PDF and translate that” - this makes everything harder. PDF locks the layout, and reconstructing it after translation is significantly more effort. If you have a .docx, translate the .docx.

Mistake 3: ignoring text length changes¶

You’ve built a document with fixed text boxes for translation from English into German or Ukrainian? The translated text might not fit. German, for example, runs 30-35% longer than its English equivalent. Build 20-30% of empty space into text boxes and table columns when designing multilingual templates.

Mistake 4: translating Excel via Copy-Paste into Google Translate¶

Pasting an entire worksheet into Google Translate and back is a classic mistake. It breaks formulas, corrupts formatting, and can destroy named ranges. Always translate only text cells, individually and carefully.

Mistake 5: not checking the result in the original application¶

Always open the translated file and compare it side-by-side with the original. Automated formatting checks are never 100% reliable - your eyes will catch things the algorithm missed.

FAQ¶

How do I keep formatting when translating a Word document online?¶

Use DeepL - it does the best job of preserving styles, tables, hyperlinks, and images in .docx files. Upload your file at deepl.com, choose the target language, and download the result. Free up to 5 MB. Before uploading, accept all Track Changes and remove comments - that significantly improves the output.

Why does Google Translate break PDF formatting?¶

Google Translate extracts flat text from the PDF with no layout information, translates it, and tries to reassemble the document. The result: the text is there, but tables are scattered, columns have merged, and complex elements are gone. For PDFs with complex layouts, DeepL Pro or Adobe Acrobat AI give substantially better results.

How do I translate Excel while keeping formulas intact?¶

Never translate Excel by pasting into Google Translate - formulas will break. Translate text cells and headers separately and carefully. If you need to convert function names between language versions (VLOOKUP → SVERWEIS for German Excel) - use the free Microsoft Functions Translator add-in.

How much does text expand during translation?¶

It depends on the language pair. English → German: +30-35%. English → Spanish/French: +20-25%. English → Japanese/Chinese: -30-40% (characters are more compact). Factor this in when designing document templates that will be used in multiple languages.

Can you translate a PDF without any formatting loss at all?¶

For complex PDFs with multi-column layouts, charts, and embedded tables - practically no tool gets it 100%. Good tools (DeepL Pro, Adobe Acrobat AI) handle 80-95% without manual correction. The most reliable approach is finding the original .docx and translating that instead of the PDF.

What should I do if the layout is still broken after translation?¶

The simplest fix is to translate from the source format (.docx instead of PDF). If there’s no source file, convert the PDF to Word, fix the obvious conversion artifacts, and then translate. For Excel - keep formulas untouched, translate only text cells, and verify the output in the same application.

Which free CAT tools preserve formatting well?¶

Smartcat has a free plan with no volume limits. OmegaT is fully free and open-source. MateCat is free online. All three support .docx and .xlsx and preserve formatting at the same level as their paid counterparts. They do require some setup time compared to just uploading to DeepL.