Guide
What Does “Keep Formatting” Mean in PDF Translation? A 7-Point Test
Try to translate a PDF without losing formatting and you'll notice every tool promises the same thing, in slightly different words — and none of them say exactly what they mean. That's not necessarily dishonest. It's that "formatting" isn't one property you can promise in a tagline; it's at least seven, and a tool can nail most of them and still fail the one that matters for your document. Here's how to check for yourself, on any tool, in about ten minutes.
Every PDF translator promises to keep your formatting
Look at how PDF translation tools actually describe this feature, in their own words:
| Tool | What they say |
|---|---|
| DocTranslator | "Keeps original formatting" |
| Online Doc Translator | "re-inserts translated text into its original position" |
| DeepL | "preserves formatting" |
| Smallpdf | "as much as possible" |
| iLovePDF | "as close as possible to the original" |
Quotes are each tool's own public description, linked to the source. We're not scoring these tools here — we haven't run controlled tests against them, and a one-line tagline isn't evidence either way. What we can say is that none of these phrases tell you which of the seven things below they actually mean.
Two tools are more specific in their own documentation, worth citing directly: Smallpdf's own blog notes that "images and special layouts don't currently carry over" into the translated document, and iLovePDF's help content notes that a visually similar substitute font may be used if the original isn't available. Those are exactly the kind of specific, checkable claims this guide is asking for — more tools should publish them.
"Formatting" isn't one thing — it's at least 7
Here's the breakdown we use ourselves. Each one is independently checkable in a couple of minutes, on any tool, on your own document.
Page size
Check: Does an A4 page stay A4? Does Letter stay Letter?
A translated PDF that silently resizes the page is already a different document — this alone can get a submission rejected before anyone reads a word of it.
Pagination
Check: Does the page count change after translation?
Chinese-to-English text often expands 30–50%. If a tool doesn't account for that, content overflows onto extra pages, breaking page references, headers/footers, and anything that assumed a fixed page count.
Tables
Check: Are column widths, borders, and — critically — numeric alignment preserved?
This is where translate-PDF-tables-keep-layout searches usually come from a real problem: a table that looks fine at a glance but has a number shifted one column over is worse than an obviously broken one, because it's easy to miss.
See this tested on a real document →Fonts
Check: Is size, bold/italic, color, and line spacing close to the original — or substituted?
Some tools substitute a visually similar font when the original isn't available. That can be invisible on screen and still change how a document prints or how text wraps.
Images, stamps, signatures
Check: Is the original graphic layer preserved untouched — or regenerated?
A redrawn signature or an altered stamp isn't just a cosmetic miss. If a tool touches these at all, treat that as a warning sign, not a feature.
See this tested on a real document →Selectable, copyable text
Check: Can you actually select the translated text with your cursor, search it, or copy it out?
Some tools return what is effectively a picture of translated text laid over the original page. It can look identical to a real translation and still fail this test completely.
Scanned documents
Check: For a scan or photo, does the tool rebuild a real text layer, or only regenerate a translated-looking image?
This is the hardest case and where most tools quietly hand-wave: there's no text to edit in place, so the output is a best-effort reconstruction, not an edit. That changes the risk profile completely.
See this tested on a real document →How to test any PDF translator yourself
You don't need to take our word — or anyone's — for this. Pick a real document (ideally one with a table) and run it through whichever tool you're evaluating, then work down this list:
- Open the source and the translated file side by side.
- Check the page dimensions and page count match (or note exactly how they differ).
- Pick one table row with a number in it and confirm the number, its column, and its alignment are identical.
- Try to select a sentence of the translated text with your cursor.
- Look for anything graphical — a logo, stamp, signature, or photo — and confirm it's untouched, pixel for pixel.
- If your source was a scan or photo, repeat the text-selection test specifically on that file.
Ten minutes, and you'll know more about that tool's real behavior on your document than any marketing page will tell you.
So, what's the best PDF translator for preserving layout?
Honestly: it depends on which of the seven properties above matters most for your document, and we'd rather tell you how to check that than give you an answer we can't back up with tests we haven't run. Here's what we can say about our own approach, specifically:
- For a native PDF (one with a real text layer), KeepFormat edits the existing text in place — same page size, same page count, same table grid, same fonts, fully selectable output — because nothing is regenerated, only replaced.
- For a scan or photo, there's no text layer to edit, so any tool (ours included) has to reconstruct the page. We publish exactly what that looks like, successes and failures, in our scanned bank statement translation guide — including a case where a dense table was mostly left untranslated.
That's the model we'd suggest applying to any tool you're considering: ask what it does for a native file versus a scanned one, and ask to see it, not just hear it described.
| Property | A vague promise sounds like | What to actually verify |
|---|---|---|
| Page size | "preserves formatting" | Does an A4 page stay A4? Does Letter stay Letter? |
| Pagination | "preserves formatting" | Does the page count change after translation? |
| Tables | "preserves formatting" | Are column widths, borders, and — critically — numeric alignment preserved? |
| Fonts | "preserves formatting" | Is size, bold/italic, color, and line spacing close to the original — or substituted? |
| Images, stamps, signatures | "preserves formatting" | Is the original graphic layer preserved untouched — or regenerated? |
| Selectable, copyable text | "preserves formatting" | Can you actually select the translated text with your cursor, search it, or copy it out? |
| Scanned documents | "preserves formatting" | For a scan or photo, does the tool rebuild a real text layer, or only regenerate a translated-looking image? |
Frequently asked questions
Why does “preserves formatting” mean different things for different tools?
Because “formatting” bundles together several independent properties — page size, pagination, table structure, fonts, images, whether the text is selectable, and how scanned documents are handled. A tool can be excellent at some of these and weak at others, and a one-line marketing claim doesn't tell you which.
How can I tell if a translated PDF's text is selectable, or just an image?
Open the file and try to select a word with your cursor, or press Cmd/Ctrl+F and search for a word you can see. If it highlights and searches normally, it's real text. If nothing happens, you're looking at a picture of text, not a translation you can copy, search, or have read by a screen reader.
Does the page count change when you translate a PDF?
It can, especially translating into English from a more compact language like Chinese, where text often expands 30–50%. Whether that pushes content onto a new page depends on how much the tool is willing to resize text to compensate, versus preserving the original type size exactly.
Is a scanned PDF translation just a picture, or is it real text?
That depends entirely on the tool, and it's worth checking directly — try to select the translated text the same way described above. Many tools that handle scans well visually still return a flattened image underneath, with no real text layer at all.