Quick answer
Translating a PDF takes five steps: work out which kind of PDF you have, set the language pair, pick a model, translate, then check the output in four specific places. Step one decides the rest — a text PDF and a scanned PDF need different handling, and treating a scan like text is where most bad translations start.
Step 1 — Work out which kind of PDF you have
| Kind of file | How to tell | What it means |
|---|---|---|
| Text PDF | You can select a sentence with the cursor | Text can be extracted; structure is the only risk |
| Scanned PDF | Selecting does nothing, the page is one image | OCR has to run before anything can be translated |
| Scan with an OCR layer | Selection works, but the highlight sits slightly off the letters | A hidden text layer already exists — this is what produces two layers of text after translation |
Try to select a line of body text before you do anything else. Thirty seconds here prevents most of the problems listed further down.
Step 2 — Set the language pair and pick a model
- Set the source language explicitly rather than relying on detection — mixed-language documents are detected wrong more often than not
- Choose the model by document type, not by benchmark score: see which AI model to use
- If the file is a scan, confirm OCR runs as part of the flow before you start: see how to translate a scanned PDF
Step 3 — Lock down what must not change
- Product names, part numbers and legal terms belong in a glossary before the first run, not in corrections afterwards
- Formulas, units and figures inside tables should come through untouched: see keeping formulas intact
- If the file is very large or stalls on upload, split it first instead of retrying: see translating a large PDF
Step 4 — Run the translation
Upload the file, confirm the language pair and the model, and start. The free allowance is counted in pages rather than in megabytes — see what you get for free. You can start from the free PDF translator.
Step 5 — Check the output in four places
Do not re-read the whole document. Check these four:
- The first page after a table — reading order breaks there before anywhere else
- Any multi-column page — columns are the most common cause of two sentences merging into one
- One table containing numbers — check the numbers, not the wording
- The last page — truncation shows up at the end
When something comes out wrong
| What you see | What actually happened | Where to fix it |
|---|---|---|
| Garbled characters, scrambled paragraphs | Structure was lost during extraction | Why PDF translation breaks |
| Text overlapping itself | A hidden OCR layer was left in the file | Two layers of text |
| Nothing was translated at all | The page is an image and OCR never ran | Translating a scanned PDF |
| Upload fails or stalls | Too large to process in one pass | Translating a large PDF |
| You need to edit the result | PDF is the wrong container for editing | PDF to Word |
Related tools
Bottom line
Most of a PDF translation is decided before you press translate: whether the file is text or a scan, and whether its structure survives extraction. Get those two right, then check four places in the output. That is the whole method.