How to Convert PDF to Word on Mac (5 Ways, Ranked)
Five routes from a PDF to an editable Word file on a Mac, ranked by how much of the document survives. The free built-in route is measured here: what it keeps, what it quietly loses, and what happens when the PDF turns out to be a scan.
A PDF is a description of where to put marks on a page, not a document with paragraphs in it. That is why every route on this list is a reconstruction, and they differ strictly in how much they bother to reconstruct. To measure the difference, one page was built twice. The first was a text PDF generated by a browser's print engine: a two-page "Quarterly Report" containing a title, two section headings, a paragraph wrapping over three printed lines, a two-column block, a real 3x3 table, a two-item bulleted list, and a running header printed on every page. It weighed 72,651 bytes.
The second was that exact page rendered to a 150 DPI picture and wrapped back into a PDF, which is exactly what a scanner or a phone photo gives you. It came out at 124,343 bytes. It is nearly twice the size of the text file, and it carries none of its text.
What the extraction actually returned
Pulling the raw text out of the 72,651-byte file returned 904 characters. The order they arrived in is the whole problem. First came Quarterly Report, then Section 1: Revenue. Then the intro paragraph appeared, but broken at every printed line rather than at the sentence: it ended one line on "Every number is" and started the next on "stated in thousands". Next came the left column only. Then the table arrived as loose lines: Region Q1 Q2, then North 1,200 1,310, then South 980 1,024. The two bulleted lines followed, with no bullets. Then the right column appeared, arriving four blocks after the left one. Finally, the running header ACME INTERNAL - DO NOT DISTRIBUTE was dumped at the end of the page rather than the top.
Unzipping the resulting Word file showed exactly what survived. There were zero <w:pStyle> elements. No Heading 1, no Heading 2, no navigable Word heading styles at all. There were zero <w:tbl> elements. The table was not a table any more. It kept three bold runs, and the original point sizes did survive as direct formatting. The file held 20pt, 14pt, 11pt, 10pt and 8pt text.
The type sizes are in the file, which is why rebuilding is possible at all. Reading the styled runs off page 1 gives real font names and sizes. Quarterly Report is Georgia-Bold at 20.0pt. The body is Georgia at 11.0pt, the table's header row is Georgia-Bold at 10.0pt, and the running header is Georgia at 8.0pt. The information is sitting there. The question is only whether a given route uses it.
1. PDFToWord
PDFToWord rebuilds structure instead of just pouring out text. A 14pt bold title becomes a Word heading you can navigate, and wrapped lines join back into the sentences they came from. Tables are rebuilt as real Word tables you can put the cursor in. Pictures are lifted at the size they were printed. Repeated headers are left out. The line printed at the top of every page does not become forty paragraphs in your document.
Type 1-4, 9 to pull specific pages. Drop a folder, and several files convert at a time without a button. They can be bound into a single Word file, each on a new sheet. Turn off paragraph joining for a form so every line comes through on its own. The original PDF is never touched. On a scan, it refuses rather than lying. The row is marked "No text layer" instead of handing you an empty document.
2. The route your Mac already ships
This beats route 1 for one real job. When all you need is the words, it is quicker to run a command than it is to open an app. Your Mac ships with /System/Library/Automator/Extract PDF Text.action. Its description reads "Extracts plain or styled text from PDF documents", and the copyright line reads 2006-2019, so it has been sitting in that folder for the better part of twenty years. You pair it with /usr/bin/textutil in the Terminal. The -convert option accepts a list of formats, and docx is on that list. The chain ends in a real Word file with nothing installed, and works on any PDF that has a text layer.
The chain was run on the test file. The styled text came out as an RTF of 1,378 bytes, and textutil -convert docx turned it into a 4,112-byte .docx with an exit code of 0. It genuinely works. What it costs you is exactly what was measured: broken lines, interleaved columns, a table that is no longer a table, no heading styles, and the page header sitting in the body. Every loss stops mattering if you were going to rebuild the layout by hand anyway.
3. ScanToDocx
ScanToDocx reads the page with Apple's Vision framework rather than a text layer. It finds the document as a four-cornered shape and unwarps the perspective, so a page shot at an angle over a desk comes out square. It reads four ways round to turn a sideways page upright. A page photographed sideways is the single biggest cause of a document that comes back empty.
A grid on the page is rebuilt as a real Word table, rows and columns intact. It detects over thirty languages page by page. Twenty photos of one contract can become a single .docx, each page on a new sheet. The limit is strict: it takes images, not PDFs. It accepts PNG, JPEG, HEIC, HEIF, TIFF and WebP. If your scan arrived as a PDF, you have to export its pages as images first.
4. Copying out of Preview into DocCafe
Select the text in the PDF, copy, paste, and save. It takes twenty seconds and gives the worst layout result. It is right for one quote and wrong for a document, because the copy comes out of the same text layer whose failures were measured earlier.
The target is DocCafe, a native .docx editor for Mac and Windows that replaces Microsoft Word offline. Nothing is uploaded. It does not convert anything itself.
5. A browser-based converter
Converting in a free web converter means writing a complete copy of the document onto a computer you do not administer, which is a different question from whether the conversion is any good, and it is covered here.
Where every route falls down
A PDF is a format you print to, not one you read back. DocConvert converts documents between 26 input formats and 25 output formats locally, and PDF is one of the formats it can write but cannot read. Its own error message says so: "PDF can't be used as a source - DocConvert can create PDFs but can't read them."
A scan has no text in it. Reading the text layer of the scanned twin returned 0 characters, and the scanned file was bigger than the text one for carrying less. Running macOS's own Vision framework on that page produced 19 blocks of text, but OCR makes mistakes. It misread two words in a single 76-word page, turning "Renewals closed ahead of plan" into "plat" and "Headcount unchanged" into "unchange". It read a comma as a full stop. Anything recognised needs reading before it is sent.
Tables survive nothing that pours text out. OCR read the table down its columns rather than across its rows, so the row "North 1,200 1,310" existed only as three fragments scattered through the output. Two columns come out interleaved on the built-in route, and that free route produces no heading styles at all. ScanToDocx does not take PDFs, only images. Forms, equations and non-Latin text are genuinely hard for all of these.
Microsoft 365 Personal is $9.99 a month or $99.99 a year. Or you can license native software that does the work on your machine. All four tools here sit in the Latte category on the menu. One plan covers every app, licensed per device. Try it on the free trial.