Skip to content
Convertube

PDF to Word converter

Convert PDF to an editable Word document in your browser, on a Mac or a PC. Scanned pages are read by AI text recognition (OCR) on your own device — nothing is uploaded.

Runs on your deviceFree, unlimitedAI text recognition, on your device

Drop files here

or

Accepts .pdf · Up to 20 files at a time · up to 80 MB each "Upload" and "download" here mean your browser reading a file from this device and writing one back to it. Your file is never sent to a server — open your browser's network panel and watch: no request carries your file.

If the PDF is a scan, the characters are read on your device. That needs a one-time 21 MB model download, then a few seconds per page. Accuracy is good but not guaranteed — stamps, handwriting and badly skewed pages come out worse.

What comes across into Word

In the PDFIn the Word document
Body text, including paragraphs that run across a page breakWhole paragraphs you can edit, not a text box per line
HeadingsWord headings, ranked by type size
Bold, italic, font, size and colourKept
Tables drawn with linesWord tables, cell for cell
PhotosPlaced where they sat on the page
Charts, diagrams and hard-to-typeset equationsPictures, with their captions kept as text
Pages set in two or more columnsWord columns
Running headers and footersWord headers and footers
Scanned pagesText read by OCR on your device

Which PDFs convert best

  • Saved from Word, Google Docs or a browser: the best case. Our test documents came back word for word; the numbers are below.
  • Reports, papers, manuals and standards: the text and structure come across; check tables drawn without lines, whose columns have to be worked out from how the text lines up.
  • Scans and photos of pages: read by on-device OCR, about 98% of characters right on clean print. Stamps, handwriting and crooked pages lower that, so read the result against the original.

Other ways to do it

  • Microsoft Word on Windows: File → Open, choose the PDF, and Word converts it itself.
  • Google Docs: upload the PDF to Drive, Open with → Google Docs, then File → Download → Microsoft Word. The text comes across; tables and layout are often simplified.
  • Adobe Acrobat: Export PDF → Microsoft Word, in Acrobat Pro (paid) or Adobe's online tool, which uploads the file.

This page works on a Mac too, needs nothing installed, and does the OCR for scans on your own computer, so contracts and IDs never leave it.

What it does not do

  • No password-protected PDFs. Remove the password first.
  • No files over 80 MB.
  • No handwriting. OCR is for printed text.
  • No pixel-perfect copy. Word reflows text, so line and page breaks can move a little from the PDF; for an exact copy of the look, keep the PDF.

How faithfully it rebuilds a document, measured

Short answer: we wrote eight Word documents with a known structure, saved each as a PDF from Microsoft Word, converted the PDF back on this page and compared the result with the original. Every word came back (5,456 of 5,456), and 129 of 130 paragraphs, all 23 headings as headings, all 6 tables cell for cell, and every bold and italic phrase. None took more than 5 seconds; a 213-page standard took 19.
DocumentPagesWhat it testsResult
Letter1Date, recipient, numbered paragraphs, sign-off9 of 10 paragraphs — see below
Report2A title and three levels of heading, bullet and numbered lists, a table, a pictureAll
Invoice1An 8-row table of items and amountsAll
Long article966 paragraphs, several running across a page breakAll
Chinese document1Chinese headings and paragraphs, a tableAll
Tables2A 46-row table running across a page break, and a small oneBoth tables cell for cell
Formatting1Bold and italic words inside sentencesAll 7 bold, all 7 italic
CV1Headings, a two-column tableAll

The one miss is the letter's recipient. Name, job title and company were written as one paragraph with line breaks inside it, and come back as three paragraphs. The PDF records three lines either way, so the two cannot be told apart; the words and their order are unchanged.

What this does not show

  • PDFs that did not come from Word. Word writes a clean text layer, which is the easy case. Before every change to the converter we also run 44 real PDFs — national standards, manuals, papers, slides, bank statements, forms — to make sure nothing gets worse, but they have no answer key and so give no score. The longest, a 213-page standard and a 307-page manual, took 19 seconds each.
  • Appearance. The comparison checks content and structure — words, paragraphs, heading styles, table cells, emphasis, pictures — not fonts, spacing or where each page ends. A list comes back as paragraphs that start with their bullet or number rather than as a Word list, though the text keeps its indent and a wrapped item lines up under its first word. Heading levels are ranked by type size within the document, so they can land one level away from the original — the report's title came back as Heading 1 and its Heading 1 as Heading 2 — while their order is kept.
  • Scans. A scanned page has no text to compare; it goes through text recognition instead (see below).

How it was measured

The eight documents were generated with python-docx, saved as PDF by Microsoft Word, and converted on this page with the default settings in Chrome on a laptop with an Intel Core i7-1165G7 and 16 GB of memory, timed from choosing the file to the download being ready, on 21 October 2026. The original and the result go through the same extraction. A paragraph counts only if its text matches exactly, and a table only if every cell does. The numbers and bullets of Word lists are removed from both sides first, because Word writes them into the PDF as ordinary text.

PDF to Word FAQ

How do I convert a PDF to Word on a Mac?

Open this page in Safari or Chrome, add the PDF and download the .docx — it opens in Microsoft Word, and in Pages too. macOS has no built-in way: Preview can open and mark up a PDF but cannot save it as a Word document. A scanned PDF works the same way, because the text recognition (OCR) runs on your Mac inside the browser; the first scan downloads a model of about 21 MB, once. Nothing is uploaded, and the limit is 80 MB per file.

Why does the Word document not look like the PDF?

It rebuilds more than the text: a table the PDF drew with rules comes back as a Word table, pictures are placed where they sat on the page, and headings, bold, centring, text colour and the coloured bands behind it all survive. A page set in columns comes back as Word columns, read down one column and then the next. What it has to guess at is what the PDF never wrote down. A table drawn without rules leaves no grid to find, so its columns are worked out from how the text lines up — figures packed closely together can end up merged into one cell or in the wrong column, so check a table like that against the original. And a scanned page goes through text recognition, which returns words and nothing else.

Does it work on scanned documents?

Yes. If the pages carry no text layer, an AI recognition model runs locally on your device. The first scan triggers a one-time model download of about 21 MB, after which it is cached; each page then takes a few seconds.

How accurate is the recognition?

Around 98% of characters on clean printed documents in our own testing, Chinese and English alike. Real scans are harder than test material: stamps, handwriting, low resolution and a page fed in crooked all make it worse. Always read the result against the original before relying on it.

Is my document uploaded for the AI OCR?

No. The AI models run inside your browser, which is the entire point of doing it this way. Every other converter that offers OCR sends the file to a server; this one downloads the model to you instead.

Are headings and page breaks preserved?

Headings are guessed: lines set noticeably larger than the body text become Word headings, ranked by size. Page breaks from the original are inserted so the document keeps its pagination. Both are inferences, and both are occasionally wrong.

Why is the file size limit only 80 MB?

Because pages are rendered into memory as full-resolution bitmaps, where cost is measured in pixels rather than in how well the PDF compressed. That is a much lower ceiling than video conversions get. A long scanned document can exceed it.

Can I get the plain text instead?

Yes — convert to TXT. It takes the same route, including the recognition step for scans, and returns the text without the Word formatting, which is often what you actually wanted.

Related guides

Related converters

PDF tools

About PDFAbout WordPDF converter