Convert PDF to Word

No. A PDF places every word at a fixed spot; a Word document flows. The tool keeps the content and its structure: the text with bold, italics and raised or lowered characters (a note mark, a superscript), the title and headings (up to three levels), paragraphs, bulleted and numbered lists, tables, and pictures at their size (scaled down to fit the page when wider or taller), in reading order, each PDF page starting a new page if you leave that option on. Text set in columns comes across as one column, each column read down in turn. It does not keep exact positions, fonts, sizes other than the headings', colours, links (their text is kept), form fields, comments or footnotes as notes.

A tagged PDF (one made accessible, for example by Word, LibreOffice or a browser with tagging on) names its headings, lists, tables and figures, and the tool uses those tags; text the tags leave out (a stamp added later) is placed by its height on the page. Any other PDF is read from its layout: lines larger than the body text are headings (a numbered heading keeps its number as text), and a table is found where two or more rows in a row have as many cells each, lined up in clear columns with white space between them; rows just above or below whose cells fall within those columns (a header, a row with an empty cell) join it. A line that starts with a bullet (•, a small drawn shape, or a dash or asterisk that lines beside it start with too) is a list item; one that starts with a number or a letter (1. or a) or 1.1) is a list item only when the list's next or previous number is beside it, so a sentence that happens to open with "E. coli" stays text.

Yes. Word numbers a list when it would print the same numbers the PDF printed: its kind (1, a, A, i or I), the marks around them, where it starts (a list that goes on after a paragraph or on the next page goes on from its number) and outline numbers such as 1.1 and 1.2. When Word cannot print them ("Step 1:", a list that skips a number, or outline numbers that start below their first level, such as a list that starts at 1.1), each item keeps the label the PDF printed, as text before it.

A cell merged across columns comes across across those columns, and one merged over rows across those rows, where the cells' places show it: a row with fewer cells than the table has its cells placed under the columns their text stands in. A cell merged over columns reaches over the columns its row leaves empty after it; one merged over rows fills the columns its row leaves empty at the start. In a table read from the layout this is done for the rows above the table's body (its header); an empty cell in the body stays empty.

A picture stored upright and whole keeps its own pixels: as PNG, or above a megapixel as JPEG at quality 90 with any transparent parts made white. A picture the page shows turned, cropped (to a rectangle or to any other shape, such as a circle, what the shape hides left white), as a mask or as layers painted over each other (as scanners often store a page), or one larger than 40 megapixels, is drawn as the page shows it, at its own resolution between 150 and 300 dpi and at most 16 megapixels. A drawing made of lines and shapes (a chart, even one of two bars, of bars of one height, standing or lying down, filled or drawn round, stacked upright or lying down, each segment in its own colour, or of bars that carry their values, a pie whose slices carry their shares, a diagram, a logo) is drawn the same way at 200 dpi, with the short labels on and around it, which are then not repeated as text, and a tagged figure keeps its alternative text. Shading behind text that is no drawing (a table's shaded cells and rows, a highlight, a panel behind a paragraph) is left out with its text kept, and the result says how many shaded boxes were left out, as it does for a frame drawn round text (a box round a paragraph), while a table whose cells are drawn as frames keeps them as the table's own grid; filled boxes of one size side by side, or one above another, that stand at least half a line of text apart, each with a short label in it (no more than one for boxes one above another), are taken for a chart's bars, and the bars of a chart are drawn together however far apart they stand, so long as nothing but white lies between them (two boxes with a sentence between them are no chart). A plain rectangle standing alone with nothing in it (a frame or a coloured panel) is left out, and the result says how many. The result says when a picture or a drawing could not be taken; a drawing's labels then stay in the text.

A page comes across as one picture when it has no text this tool can take: a scan, which is a picture of a page, or a page of vector drawings (a chart, a diagram, the boxes of a printed form), which are drawn into a picture of the page. A form's fields are not the page's own drawing and are left out: a PDF of fields alone has nothing to put in a Word document, and the page says so. Its picture is put in the document, but words in it cannot be edited, because this tool does not recognise text in pictures (OCR). The result says which pages were scans. A scan whose text was recognised before (its words lie invisibly over the picture, as in a searchable PDF) comes across as that text, and the picture of the page is left out; the result says which pages those were.

No. The PDF is read and the .docx is written inside your browser; the file never leaves your device.

Running headers and footers are left out, and the result says how many: a line smaller than a heading, in the top or bottom 8% of the page, repeated on at least three pages and on at least half of them, its numbers the same on each or counting with the page (such as "Page 3 of 10"), and the same picture repeated at one place there (a logo), which the result counts too; different pictures at one place are each kept. One that changes from page to page in other ways comes across as text on each page. Text in a layer the PDF switches off is left out, as a viewer hides it, each word by where it is drawn, even where the same words are shown elsewhere on the page, and text drawn twice at one place (a shadow, a bold made by drawing twice) comes across once. A word broken over two lines with a hyphen keeps its hyphen. Some PDFs use fonts without a proper character map, and their text comes out as odd symbols: that is a limit of the file, not something this tool can repair.

No. The page loads a PDF engine into your browser and every operation runs on your device. You can check this yourself in the browser's network panel: after the page has loaded there are no upload requests. Closing the tab discards the file.

No. These basic tools are free and never add a watermark. The only limits are the tool's own: up to 300 pages per conversion; an encrypted PDF (one that needs a password to open, or one whose encryption only restricts what may be done with it) has to have its encryption removed first; scanned pages come across as pictures, without their text. Within them, the practical limit is your device's memory.

Encrypted PDFs cannot be processed here. Remove the password first, then use this tool.

Nothing you open or type. While the tab is open, the tool pages keep a short list of the steps taken in it, in the tab's session storage: the step (page opened, started, succeeded, first success, used again or failed), the tool's name, the time, how many files were used and, for a failure, an error code; never a file name, the contents of a file or your text. It holds at most the latest 200 steps, and the browser deletes it when the tab closes. The page never sends it. Where a Report a problem button is shown, you can attach the last 20 steps to your report; nothing is attached unless you tick that box. In local storage the page keeps only choices you made: a Light or Dark theme until you choose System again, the language you picked with the EN / 中文 switch if it differs from your browser's language, and this tool in your recently used list if you switched that list on at the tools page.