Can Gemini Translate a PDF? Scanned Pages, Long Files, and What You Get Back

Yes — upload a PDF to Gemini, ask for a translation, and you'll get one. It reads born-digital PDFs from their text layer and scanned PDFs as images through its vision capability, and the translation quality is on par with the other frontier models: fluent, idiomatic, good with context. Of the chat assistants, Gemini is also the most comfortable with a long file, thanks to a context window that can hold a substantial document in one go.

So the honest question isn't whether Gemini can translate your PDF. It's whether a chat reply is the thing you actually need — and that depends entirely on what the PDF is.

What Gemini does with a PDF

• Born-digital PDF (text is selectable): Gemini reads the text layer directly. Translation is fast and complete for short and medium files.
• Scanned PDF (pages are images): Gemini reads each page as an image. Clean printed scans translate well; faint, skewed or handwritten pages are hit-and-miss, as with any vision model.
• Long PDF: Gemini's larger context means fewer “continue” prompts than other assistants, but upload size limits and response length limits still apply, and very long documents may be summarized in places rather than translated line by line.

In every case the output is text in the chat — sometimes a markdown table where the model detected one, more often paragraphs. The PDF itself is never modified. If you need a translated PDF, you're building it yourself from the chat output.

Where it's the right tool

Use Gemini when the document is the input and understanding is the output: a letter you need the gist of, a contract clause you want explained, a long report you'd rather ask questions about than read in translation. Its length tolerance makes it particularly good at “what does this 80-page thing say about X?” — a question a document translator can't answer at all.

Where it isn't

Anything whose meaning lives in its layout. A bank statement is rows; a lab report is a table of values against reference ranges; a certificate is fields, seals and a registry number in a particular corner. Gemini will translate the words on those pages accurately and hand them back reflowed — and a reviewer comparing the translation to the original field by field has nothing to compare against. For those documents, a layout-preserving translator that rebuilds the page is the shape you need; the AI PDF translator guide lays out the three categories and how to pick.

We ran a scan and two digital papers through Gemini with the request to keep the original layout. The results, page by page and next to Reglyph, are in our side-by-side test.

A practical split

1. 1Read with Gemini. Upload, ask for a translation or a summary, ask follow-up questions. Best for understanding.
2. 2Translate documents with a document tool. Upload the scan to a layout-preserving translator and get a PDF back that mirrors the original. Best for filing, sharing and review — see the [document translation hub](/translate/documents/) for what each document type needs.
3. 3Certify what gets submitted. Neither route produces a certified translation. For USCIS and courts, a qualified translator reviews and signs — much faster from a layout-faithful draft than from a chat transcript.