Word to Plain Text: Strip a Document Back to Its Words
Copy-pasting out of Word is how formatting gets smuggled into places it has no business being. Paste into a CMS and you drag along font declarations and stray spans. Paste into a code comment and the quotes turn curly and the string stops matching. Paste into a terminal and the em dashes arrive as characters your script wasn't expecting. The document looked fine. The bytes were never plain.
Converting Word to plain text cuts that off at the source. What comes out is the prose in reading order — no styles, no colours, no revision furniture, no invisible layout scaffolding. It's what you want when you need to extract text from a Word document for analysis, search, or anything that treats formatting characters as data to be parsed. Free, no sign-up, 50 MB limit, nothing kept on our side.
What Gets Stripped, Specifically
A .docx carries far more than the words. Here's what's on the cutting room floor:
- All visual formatting — bold, italics, fonts, sizes, highlight colours, that one paragraph someone left in Comic Sans. Emphasis is gone, and there's no marker where it was.
- Heading hierarchy — a Heading 1 and the sentence beneath it become two ordinary adjacent lines. Nothing distinguishes them.
- Tables collapse to lines of text. The cell values survive; which column each belonged to does not. This is the biggest reason to reach for Word to Markdown instead when your document is table-heavy.
- Embedded images vanish entirely. They're pixels in
word/media/inside the archive, and this converter doesn't run OCR — so text inside a pasted screenshot never appears in the output. If that content matters, extract the image and use image to text on it separately. - Headers, footers, page numbers, watermarks — dropped. Text has no pages.
- Comments — stored in their own part of the file, not body text, so they don't come through.
The Invisible Characters Word Leaves Behind
Plain text means no markup. It does not mean ASCII, and Word's autocorrect has spent years quietly replacing the characters you typed with typographically nicer ones. Those survive conversion, because they're genuinely part of the text:
- Smart quotes — straight
"and'become curly U+201C / U+201D and U+2018 / U+2019. This is the classic reason a regex, a CSV import, or a copied code snippet silently fails. - Em and en dashes — a typed double hyphen becomes an em dash. Fine in prose, surprising in an identifier or a command.
- Non-breaking spaces (U+00A0) — they look exactly like spaces and don't match
\sin some older regex flavours, or' 'in a naive split. - Soft hyphens and optional breaks — zero-width characters sitting inside words, invisible until a string comparison fails for no apparent reason.
- Ellipsis — three typed dots collapse into the single character U+2026.
None of this is a conversion bug; it's faithful output. But if your pipeline is fussy, run a Unicode normalisation pass after converting. Knowing to look for it is most of the battle.
Tracked Changes: Accept Them Before You Convert
Word stores revisions inline — inserted text wrapped in w:ins, deleted text kept in
w:del so it can be restored. Conversion resolves the document to its accepted state:
insertions kept, deletions gone. That's almost always what you want, but "almost always" is doing
work in that sentence, particularly with a heavily marked-up legal draft where you may not know what
state the document was in.
Do it explicitly. Review → Accept All Changes on a copy, save, convert. Now the text you get is unambiguously the text you looked at. If you actually want to know what changed between drafts, the better move is to convert each version to text and run a diff — plain text diffs beautifully, which is precisely why this format exists.
When Text Is the Right Answer and Markdown Isn't
Markdown wins when structure carries meaning. Text wins whenever markup would be noise in whatever consumes it next:
- Classifiers and text analytics. TF-IDF, topic modelling, sentiment scoring,
keyword extraction — all bag-of-words at heart, and all of them will tokenise
**as though it were vocabulary unless you strip it first. - Embeddings. Chunk, embed, retrieve. Syntax characters add positions to the embedding without adding meaning. Prose chunks retrieve more accurately than marked-up ones.
- Search indexes. Postgres full-text, SQLite FTS, Elasticsearch — every one of them wants a raw text column. Markdown just means a sanitising step before insert.
- Diffing and version control. Line-level diffs read cleanly on prose. Markdown tables are the opposite: change one cell and the whole row rewrites, so the diff tells you nothing useful.
- Piping into scripts.
wc,grep,awk, a quick Python one-liner — text is the universal interchange format for all of it. - Minimum tokens. On a tight context budget, every pipe and hash is spend without return. The token counter under the output shows you the difference immediately.
The format toggle at the top of this page switches between text and Markdown and carries your selected file across, so producing both and comparing takes one click rather than a second upload. Same logic applies elsewhere in the suite — PDF to plain text and HTML to text exist for the same reasons.
.doc, .docx, and Getting Clean Results
The old .doc format is a binary compound file from before 2007 — a small internal
filesystem of streams whose layout shifted between Word versions. .docx is Open XML: a zip
archive of well-specified XML. Extraction from .docx is reliable; extraction from .doc is best-effort.
If an old file gives you odd output, open it in Word or LibreOffice, Save As .docx, and convert that.
The same advice applies to RTF documents.
Two more habits worth having. Skim the first hundred lines of the output before you use it — spotting a table that flattened badly takes ten seconds and saves a confusing conversation with a model later. And if the source is enormous, split it: a 400-page manual converts fine but won't fit a context window, and you'll get sharper answers from the chapter you actually care about.
Frequently Asked Questions
How do I convert a Word document to a plain text file?
Drop the .docx or .doc above and the text appears below, downloadable as .txt. Free, no sign-up, 50 MB per file. Both formats work — the modern zipped XML one and the older binary one Word used before 2007.
Can I batch convert multiple DOCX files to TXT?
Not here — this page takes one file at a time. For a folder of them, put the documents in a directory and use the local folder converter, which walks a whole tree and lets you tick the files you want in one pass. On the command line, Pandoc in a shell loop is the standard answer.
What happens to tracked changes and comments?
You get the document as it currently reads, with revision marks resolved rather than reproduced. Comments in the margin are not part of the body text and do not come through. If the point of the exercise is reviewing what changed between two drafts, convert both versions to text and run a word-level diff — that gives you the edits far more legibly than the sidebar does.
Do headers, footers and footnotes get included?
Footnote bodies generally come through, which is what you want for a cited document. Repeating page headers and footers are page furniture rather than content, so they should not interleave the way they do with PDF extraction — one of the genuine advantages of converting from DOCX rather than from an exported PDF of the same document.
Should I use Word to text or Word to Markdown?
Text when you want the words and nothing else — word counts, keyword sweeps, embedding chunks, feeding a text-to-speech engine. Word to Markdown when the document was written with real heading styles and you want that outline to survive into the output.
The Rest of the Toolkit
Word is one of thirteen formats here. File2Txt takes any of them from a single upload, or go directly to PowerPoint to text for decks, Excel to text for spreadsheets, or MSG to text for Outlook emails. Got a whole folder of drafts? Zip it and run ZIP to text to convert everything inside at once.
For code there's the GitHub to text converter, a GitLab version, and a local directory converter. For web pages, Web2Txt pulls a URL into text. There's also a longer guide to preparing documents for LLMs if you want the general argument rather than the Word-specific one.
Repo2Txt is built and maintained by v12hero, an independent developer building privacy-first native and web apps.