PDF to Word

Start here

1. Drop a PDF on the file box. The first page appears in the well, the rail lists every page, and the note above says whether any page has to be read by recognition.
2. Pick an output layout. Editable flow for a document you will keep writing in, Exact layout when a line shifting even slightly would be wrong.
3. Press Convert to Word. The bar names the step and the page it is on, and Cancel stops it without writing anything.
4. Read the result before you save it: the strip counts pages and words, the ledger counts headings, paragraphs, tables and pictures, and a page whose recognition dropped faint lines is marked on the rail. Then press Download.

About this tool

Retyping a PDF because you need one paragraph out of it is a miserable way to spend an afternoon. Converting a PDF to Word here rebuilds the whole thing as a real .docx, the kind Microsoft Word, Google Docs, LibreOffice and Pages will all open and let you rework like anything you wrote yourself.

The trick is reading where every word physically sits on the page rather than the order the file happens to store it in. A two-column page is read down one column and then down the other. Paragraph breaks come out of the spacing, headings out of type size and weight, lists out of their bullets and numbers. Each run keeps its size along with its bold or italic, a link arrives as a working link, and the paper size, orientation and margins are lifted off the PDF. A table that draws its own ruling lines comes back as a Word table you can edit cell by cell.

Two output layouts sit beside the page, and the choice is remembered. Editable flow, the default, lets text reflow the way a Word document should, so page breaks land where Word wants them. Exact layout pins each line to the coordinates it already had, which is what a filled form or a certificate needs. A scanned page has no text layer, so it is recognised on your device during the same run; the page well says which pages that applies to before you press anything, and the first scan downloads the recognition model once, with its size shown.

Good to know

• A table is recovered from the lines it draws. A fully ruled grid becomes a Word table, while one with merged cells, or one held together by alignment alone, arrives as aligned paragraphs.
• Photographs and logos come across at the size and position the page gave them. Drawn shapes get left out, though: charts, diagrams, shading, anything the PDF paints with lines rather than pixels. The result counts them and names the pages they were on.
• Fonts are matched to metric-compatible stand-ins rather than embedded, and there are seven to choose from: Calibri, Cambria, Georgia, Arial, Verdana, Times New Roman and Courier New. A document set in an unusual typeface will look close without looking identical. Text colour is dropped.
• Rotated and vertical text is skipped, since there is no honest way to lay it out as an ordinary Word line; the result says how many pieces stayed behind.
• Columns, paragraphs and headings are inferred from where the words sit, so an unusual layout can still split or merge a block. A minute of tidying in Word sorts it out.