Google Translate PDF: Limits, Scanned Pages, and the Right Workflow
Translate a PDF with Google's document tool, check its size and scan limits, and use a four-gate checklist to decide whether the result is usable.
Google Translate can translate a PDF through its Documents tab on a computer. As checked on September 8, 2026, Google's consumer document tool allows files up to 10 MB and PDFs of 300 pages or fewer. It does not translate text contained in images or scanned PDF pages, even when those pages appear in the downloaded result. Google also says document translation is unavailable on mobile and smaller screens. These are the current Google Translate document rules, not the separate Google Cloud API's limits.
The useful question is not just “Did my PDF upload?” It is “Was the right text extracted, translated, and returned in a form I can use?” Those are separate checks.
Translate the PDF, then keep both versions
On Google Translate, choose Documents, select the source and target languages, and select your local PDF through Browse your computer. Start Translate, then use Download translation when processing finishes. Automatic source-language detection is an option in the official workflow.
Save the original and translated files separately. Before annotating or repairing anything, compare a difficult page—not just the cover. A clean first page says little about a table, a scanned appendix, or a dense two-column section later in the file.
If the document contains confidential material, check whether you are allowed to send it to an external service before uploading. A technically supported file is not automatically an approved upload.
Use four gates instead of a single success check
The following is our inspection framework, not a score assigned by Google. Download the reusable PDF translation preflight sheet and record the actual pages and discrepancies in your own document.
| Gate | What you establish | What it does not establish |
|---|---|---|
| Upload eligibility | The file fits the service's current input rules | That every visible word can be extracted |
| Source text | Copied or recognized text matches the original | That the translated meaning is correct |
| Meaning | Important statements survive translation | That tables, notes, and reading order remain usable |
| Document usability | The result works for its intended reading or sharing task | That someone has approved it for every other purpose |
This distinction changes troubleshooting. If an identifier is already wrong in copied source text, changing the target language will not repair the extraction. If the words are accurate but attached to the wrong table row, more fluent prose will not repair the association.
Start with the earliest failed gate. Otherwise you can spend time polishing an error introduced before translation began.
Check source text before blaming translation
Try selecting individual words on several representative pages and pasting them into a plain-text editor. Include a heading, an ordinary paragraph, and the hardest layout in the document.
Look beyond whether something copies:
- Does a two-column paragraph copy in a readable sequence?
- Does a name retain its letters and diacritics?
- Does the value 0.05 retain its decimal point?
- Is a value still associated with its row label?
- Are image captions actually present in the copied text?
A selectable text layer can contain mistakes. Equally, a failure to copy does not by itself identify the cause: an image-only page, a problematic encoding, or a restriction may require different handling. Compare the visible page with the available text before deciding to run OCR again.
For documents combining Chinese, English, and numeric fields, the mixed-language OCR guide shows why each part needs a separate check. For the broader scan-to-translation sequence, use how to translate a scanned PDF.
When Google Drive and Docs are the better Google route
Google Drive's conversion to Google Docs is a different operation from uploading a PDF to Google Translate. It can extract text from images and PDFs. Google's conversion guidance recommends files no larger than 2 MB, upright pages, text at least 10 pixels high, and sharp images. It warns that lists, tables, columns, footnotes, and endnotes may not be detected reliably.
That makes it a possible route to editable source text, not a promise to reproduce a finished PDF. In Drive, open the file with Google Docs, inspect the conversion, and correct material extraction errors first.
Once the source is a Google document, Tools → Translate document lets you name a translated copy and choose its language. Google says the result opens as a separate document. If that command is missing because the file is in Microsoft Office editing mode, convert it to Google Docs first. See Google Docs' translation instructions.
Keep these decisions separate:
- Is the converted source text complete enough to translate?
- Do you want an editable reading copy or a visually similar PDF?
- How much layout reconstruction are you prepared to do?
If conversion has interleaved two columns, repair their sequence before translating. If you need an exact table structure, reconstruct and check that structure rather than assuming an export will restore it.
Two worked decisions: a small scan and a longer text PDF
These are hypothetical examples for using the checklist, not files we uploaded to Google or measured translation results.
Example A: a 1.5 MB, 12-page scanned article. Its small size does not solve the source-text problem. Suppose the visible text cannot be selected and inspection confirms the pages are images. Choose an OCR/conversion route, then compare the recognized title, body, and references with the page images. If a reference number changes during recognition, fix it before translation. A document can fit an upload limit while failing the next gate entirely.
Example B: an 8 MB, 240-page text-based report with scanned appendices. The body might provide usable text while the appendices do not. Treat it as a mixed document: inspect the appendix pages independently and note which material still needs recognition. Do not accept an untranslated appendix merely because the main chapters read well. If you split material into working copies, keep a page-to-source record so nothing disappears when the result is assembled.
Neither case tells you which vendor will produce the best translation. They tell you what must be established before comparing outputs fairly.
Decide whether the result is good enough for its purpose
For personal reading, you may accept awkward line breaks while still requiring accurate names and numbers. For a document someone else will use, layout relationships and review history can matter just as much as wording.
Compare the source and result on:
- one important statement, including its negation, condition, or qualification;
- a dense page and the ending, looking for missing content;
- tables, checking labels against values rather than counting intact cells;
- figures and their captions;
- footnotes, references, and the order in which the page should be read.
Record unresolved errors instead of marking the whole document “checked” after a quick skim. Use the PDF translation QA checklist for a more complete acceptance pass, and involve an appropriately qualified bilingual reviewer where a misunderstanding would materially affect a decision.
If Google's result meets your reading need, there is no reason to change tools just because another service has a longer feature list. If it does not, name the failure first: missing source text, meaning, layout, or review effort. The PDF tool comparison and format-preservation workflow address those broader choices.
For an eligible file that still needs a complete translated copy, BookTranslator's PDF workflow is another option. Choose its mode deliberately: OCR mode reconstructs readable content from scanned or image-heavy PDFs rather than matching the original page layout, fonts, and sizes. It is not a guarantee of error-free recognition or a substitute for checking the result. The same four gates still apply.
Súvisiace príspevky
