Converting PDF to Word, Explained
Why converting PDF to Word is hard, what survives (text, headings, lists) and what breaks, why scanned PDFs need OCR first, and what a DOCX file really is.
By Udit Agarwal · Updated October 17, 2026
Converting a PDF to Word is hard because a PDF does not contain paragraphs, headings or tables: it contains instructions that draw runs of characters at exact positions on a page. A converter has to guess the structure back from those positions, sizes and fonts. Running text, headings and simple lists usually come across well, while multi-column layouts, tables, form fields and the exact look of the fonts often do not. A scanned PDF holds only pictures of pages, so it needs OCR before there is any text to convert.
PDF, standardized as ISO 32000, is a page description format. It records what to paint where, so a page looks the same on every screen and printer. Text is painted by short sequences of drawing commands inside the file, and a heading might be stored like this:
BT
/F1 18 Tf
72 720 Td
(Quarterly results) Tj
ET
That reads: begin text, use font F1 at 18 points, move to 72 points from the left and 720 points up from the bottom of the page (a point is 1/72 inch), paint the string, end text. Nothing says "this is a heading" or "this paragraph continues on the next line". Each line is usually a separate command, many programs place words or even single letters separately, and the space between two words may exist only as a gap in the coordinates.
Some PDFs also carry tags, an optional structure layer for accessibility that the PDF/UA standard (ISO 14289) requires, which can mark headings, lists and tables. Many PDFs have none, and converters that work from positions, including ArrayKit's, do not read them.
A converter reverses the process with a chain of guesses:
Each step fails in predictable ways. A pull quote set in large type becomes a heading. A sentence that happens to end exactly at the margin splits its paragraph. On a page with two columns, lines sit side by side at the same height, so a converter that groups by height reads straight across both columns.
| Element | Usual result | Why |
|---|---|---|
| Running text in one column | Comes across well | Its lines sort cleanly from top to bottom |
| Headings | Mostly right | Inferred from size and weight |
| Bulleted and numbered lists | Mostly right, but flat | The markers are visible; nesting levels are not |
| Two or more columns | Often scrambled | Lines side by side share a height |
| Tables | Often flattened into text | The grid is drawn as lines, not stored as cells |
| Fonts, spacing and positions | Replaced | Word reflows the text in its own fonts |
| Headers, footers, page numbers | Ordinary paragraphs | They are text like any other |
| Fillable form fields | Not converted to Word fields | Their values sit in a separate form layer |
| Images and charts | Depends on the converter | They are separate objects from the text |
Fonts cause a quieter problem. A PDF can map its glyphs to characters in a private way, and when the table that translates them back to normal text (the ToUnicode map) is missing, the page looks fine but the text copies out as gibberish. No converter can recover that text except by OCR. PDF Font Finder shows which fonts in a file lack the map.
A scanned PDF is a stack of page images. You can read it, but there is no text layer, so a converter finds nothing. A quick test: try to select a sentence or search for a word. If nothing highlights, the PDF is a scan.
The fix is optical character recognition. PDF OCR recognises the text on each page and returns a searchable PDF with the recognised words added as an invisible text layer, and that file converts like any other. OCR makes mistakes on faint, skewed or low-resolution scans, and every mistake carries into the Word file, so proofread the result. The guide to how OCR works explains what affects accuracy.
DOCX is the Office Open XML word-processing format, standardized as ECMA-376 and ISO/IEC 29500, and Word's default since 2007. A .docx file is a ZIP archive of XML files plus any images. The one ArrayKit's converter writes holds six parts:
| Part | What it holds |
|---|---|
[Content_Types].xml |
The type of every part in the package |
_rels/.rels |
Where the main document and its properties are |
word/document.xml |
The text, paragraph by paragraph, each with a style name |
word/styles.xml |
Fonts, sizes and the Heading 1 to 4 definitions |
word/_rels/document.xml.rels |
The link from the document to its styles |
docProps/core.xml |
The title and author |
Because it is a ZIP, you can drop any .docx into Unzip Online and read the XML yourself; the guide to ZIP files explains the container. The older .doc format, which Word used until 2007, is a binary format instead. Word, LibreOffice, Pages and Google Docs all open DOCX.
PDF to Word reads a PDF's text layer in your browser and rebuilds it as an editable .docx, following the steps above. For each page it works out the body text size, then:
In an 11-point document, that makes a 20-point title Heading 1, a 16-point section title Heading 2 and a 14-point subheading Heading 3, while a 12-point line counts as a heading only if it is bold, short and set off by a gap.
Before you download, the page lists the first 80 blocks it found with their type, plus counts of headings, paragraphs and list items. Keep page breaks, on by default, starts each PDF page on a new Word page. An optional author field writes a name into the document properties, and the title is taken from the file name.
The Word file is set in Calibri 11 point on A4 pages with 2 cm margins, whatever the original page size. Headings use Word's Heading 1 to Heading 4 styles, bold at 18, 15, 13 and 12 points, so they appear in the Navigation pane. List items use the List Paragraph style with the bullet or number kept as typed text, so Word will not renumber them.
It does not carry over images, fonts, colours, bold or italic inside a paragraph, links, or tables, whose cells arrive as plain text separated by spaces, sometimes with several rows run together into one paragraph. Two-column pages interleave. A PDF with no text layer gets a warning that it needs OCR, and one that needs a password to open cannot be read until you remove the password with Decrypt PDF. The PDF is read and the .docx is written in your browser tab, and the file is not uploaded.
Why does my converted document look different from the PDF? A PDF fixes every character to a position, while a Word document reflows text to fit the page and the fonts available. A converter keeps the words and as much structure as it can infer, but line breaks, spacing and fonts change by design.
Can I convert a scanned PDF to Word? Not directly, because a scan contains pictures of text rather than text. Run it through OCR first to add a text layer, then convert the searchable PDF and proofread it for recognition errors.
Why do tables break when I convert a PDF to Word? A PDF draws a table as text placed in rows, often with separate ruled lines, and nothing records which text belongs to which cell. Converters have to infer the grid from positions, and for data tables a spreadsheet conversion such as PDF to Excel usually works better.
Why is the converted text gibberish? The PDF's fonts use a private mapping from glyphs to characters and lack the table that translates them back. The page looks right, but any extraction produces nonsense, and OCR on the page images is the usual way out.
Can Word open a PDF directly? Yes. Word 2013 and later can open a PDF and convert it to an editable document, with the same kinds of limits: simple text converts well, while complex layouts and tables may shift.
Does converting to DOCX keep the images? It depends on the converter. ArrayKit's PDF to Word rebuilds the text only, so save the pictures separately with Extract Images from PDF and insert them in Word.
What is the difference between DOC and DOCX? DOC is the binary format Word used until 2007. DOCX is the Office Open XML format, a ZIP package of XML files that every current word processor opens.
Is it safe to convert a confidential PDF online? Many converters upload the file to a server. ArrayKit's PDF to Word reads the PDF and writes the .docx in your browser, so contracts and reports stay on your device.