How to Convert a PDF to Word Without Losing Formatting

Last updated: 2026-09-16

Why "PDF to Word" so often comes out formatted wrong

Every symptom below has one root. A PDF describes a printed page: this glyph, in this face, at this position. It does not record that six of those glyphs are a heading, or that four rules and twelve short strings are a table. Microsoft's note on opening PDFs names the format in one phrase — a PDF is a fixed file format. The layout is coordinates; the structure was never written down.

Every converter has to infer it back, and that is where they diverge. Either the inference is abandoned — the page arrives as one flat picture, or a scatter of positioned boxes, one per line — or it is approximated: a table has become rows of tab stops, two columns have collapsed into one read across instead of down, the typeface has been swapped for one that sets to a different width. Nothing is wrong enough to notice on open; everything is wrong enough to fight for an hour.

Still choosing a route? How to convert a PDF to an editable Word document compares four.

Text boxes instead of real paragraphs

The most-reported symptom: you open the .docx, it looks like the PDF, you click into a sentence to fix a typo — and the sentence is sitting inside a rectangle with handles on it.

Microsoft describes a text box as an object you add to your document that lets you put and type text anywhere in your file, and anywhere is the whole story: a text box is an absolutely positioned container, holding a coordinate and a size and keeping them. A Word paragraph is flow content in the body, so it knows where the margins are and moves when its neighbours move.

Map every block of PDF text onto its own box at the printed coordinates and you reproduce the original exactly — at the instant of opening. After that the boxes have no idea the others exist: add a line to one and it grows over its neighbour instead of pushing it down.

Why it keeps happening no matter which tool you try

This is not bad luck, and trying again somewhere else often returns the same shape of file — not a statement about anyone's product, but a property of the strategy. A converter can aim at two goals: reproduce the appearance of the page, or reconstruct the document that was printed to make it. For appearance, an absolutely positioned box is the most direct instrument there is — ink where ink was, no decision about what that ink meant. Reconstruction demands a commitment, block by block, to what the thing is.

So a page of text boxes is the signature of an appearance-first strategy, and that changes the fix: not a cleaner file through the same kind of tool, but a converter answering the structural question at all. If you are weighing up an AI assistant instead, that is another mechanism — a language model writes text, which is a different job from emitting a document file.

What "no text boxes" actually requires — and how to check it

Getting a document rather than an arrangement takes one extra stage: the converter classifies the page — this is a heading, this is body text, these twelve strings are a table — and writes each conclusion as the Word construct for it, a heading style or a table object or a section with a column count, not a rectangle with words in it.

Four checks tell you which you were handed. They work on any converted file, ours included:

1. Press Enter mid-paragraph. Body text pushes the rest of the page down. If nothing moves, or the layout shatters, or the cursor will not land in the text, you are inside a box or on top of an image.
2. Open the navigation pane. It is populated from heading styles — the same mechanism Word uses to build an automatic table of contents from the headings in your document. An empty pane means your headings arrived as bold text, not structure.
3. Click a line and read the font box. You want one real typeface name; a blank box means the selection spans several faces — what a page of per-line boxes looks like from the ribbon.
4. Click inside a table. Microsoft documents that the Table Design and Table Layout tabs appear only once you have clicked inside a table. No tabs, no table — just text with lines drawn near it.

On our side: text-based PDFs are rebuilt straight from the page geometry into paragraphs, tables and sections, with no model in the loop, and every conversion then runs an automatic check before you get it: page count against the original, the .docx text compared back against what was read off the page, an audit that the styles really are editable.

Tables messed up: a real table object vs. drawn lines

Tables are the most expensive thing on a page to get wrong, because a PDF table is very often not a table: sometimes text with rectangles drawn around it, sometimes only text aligned into columns by tab positions. Either way the grid is something a reader's eye assembles, not something the file declares. So the converter either detects it — column boundaries, row boundaries, merged cells — and writes a genuine Word table, or it copies the appearance: tab-separated text plus a few drawn lines. Both print much the same. Only one lets you add a row, or retype a figure without the columns wandering.

Here, table detection is its own stage of the pipeline rather than a side effect of reading text, and the result is emitted as a table object. Tables split across a page break get a second pass: nothing in the source says "this continues overleaf", so two halves are joined only when the column structure agrees and a repeated header row or a continuation caption corroborates it — leaning towards leaving two tables alone rather than fusing unrelated ones. If the table is the whole reason you are converting, PDF table to Excel takes the spreadsheet route.

Layout broken: columns, section breaks and font substitution

Columns and sections. Word does not mark a two-column stretch by moving text around. It divides the document into sections, and a section carries column count, margins, headers and footers. Microsoft describes the continuous section break as the one that starts a new section on the same page, used precisely to change the number of columns without starting a new page. A converter that never reads the column boundary as a section boundary has one option left: place the text across the full width. That is how a two-column paper collapses into one column whose sentences alternate between two trains of thought. Our builder brackets a columned stretch in a matched pair of continuous section breaks, the closing one carrying the column count.

Font substitution. Barely a conversion problem. A PDF embeds the font program itself; a .docx refers to a typeface by name. Open it where that name cannot be resolved and the renderer quietly substitutes something similar-looking, which changes character widths and therefore where every line breaks. Microsoft's own remedy is font embedding, offered so a document shared with someone who lacks the same fonts keeps its fonts, layout and styling. What we do is narrower: font names are normalised to their canonical form, so the document remembers which typeface it was set in. Whether your machine has that typeface is outside any document's control, and a legacy name can still be substituted. In Chinese or another CJK script the risk is more specific, and converting a PDF with Chinese characters to Word works through it.

When the PDF is a scan

Everything above assumes there is something to infer from. A scanned page holds pixels and no text layer: no glyphs, no coordinates, no font names. Classification therefore happens page by page — one file can be text throughout and scanned for its signed appendix — and pages with no text layer go to a vision model that reads the page and its structure together. The output enters the same builder a text page does, rather than being embedded as an image. That path has its own page: scanned PDF to Word.

What we actually do

With the limits attached. When you convert PDF to Word without losing formatting here, a file that already contains text never touches a model: it is rebuilt from the page geometry into Word paragraphs, table objects and sections. That costs nothing, which is why there is no watermark and no paywall on it. Basic mode is free and needs no account: 10 pages a day and 100 a month, rising to 100 pages a day and 1,000 a month once you sign in. In Basic mode scanned pages come back as page images, with any text the PDF already carries left editable; AI mode transcribes them into real editable text. AI mode, which transcribes scanned pages into real editable text, requires a signed-in account with a payment method on file and draws on that same account allowance; beyond it, purchased pages are charged. Converting with your own API key is not metered at all. Page packs on the pricing page are not switched on yet. Converted files are deleted six hours after conversion by default.

What we are not telling you is that a rebuilt document matches the source page down to the last line break: real paragraphs may break a line somewhere a fixed page did not. The useful claim is narrower — the things you will edit are the things Word knows how to edit.

Run the four checks on one page of your own file: convert PDF to Word without losing formatting.

Sources

Word's behaviour above is cited from Microsoft's documentation, checked 2026-09-14.

• Microsoft Support — Opening PDFs in Word
• Microsoft Support — Add, copy, or remove a text box in Word
• Microsoft Support — Insert a table of contents
• Microsoft Support — Add a cell, row, or column to a table in Word
• Microsoft Support — Insert a section break
• Microsoft Support — Embed fonts in documents or presentations