Remove metadata from a Word document

A Word file records who created it, who saved it last, the company name, total editing time, every comment and its author, and each tracked change with the reviewer's name and timestamp. Microsoft warns that switching to No Markup only hides tracked changes until the next person opens the file. This tool removes those records from .docx files, accepting the changes or keeping them under the neutral name "Author", then reads the cleaned copy again and lists what it still finds there.

What a .docx file remembers

A .docx file is a zip package of XML files. Open one with an archive tool and you find more than the text you typed:

• Document properties: author, last saved by, title, subject, keywords, created and modified dates, revision count, last printed date.
• Application properties: company, manager, the template name, total editing time in minutes, and which program and version saved the file.
• Custom properties that templates or document systems add, such as client numbers or case references.
• Comments, each with the author's name and initials. Newer versions of Word also keep a list of the people who commented and the tasks assigned from comments.
• Tracked changes: every insertion, deletion, move and formatting change, each tagged with a name and a date.
• Editing-session ids (rsids): numbers Word stamps on text each time the file is edited and saved. They can show which passages were changed in the same sitting, and matching ids in two documents can hint that they share a history.
• Hidden text: text formatted with the hidden font effect, invisible on screen unless you turn on formatting marks.
• Folder paths: of the attached template, of linked pictures, of a mail-merge data source, and inside field codes or picture descriptions. A path through a user folder contains the user name.
• Document variables, custom XML data and a preview thumbnail.
• Printer settings: the configuration of the printer the file was last set up for, which carries that printer's name and sometimes its network path.
• Photos you pasted in, each with its own EXIF data.

Microsoft's own guidance puts it plainly: to remove tracked changes, you must accept or reject them. Choosing the No Markup view only hides them, and they are visible again the next time someone opens the document.

How to clean a Word document

1. Under Tracked changes in Word files, choose Accept all changes (the default) or Keep changes, hide who made them.
2. Press Choose files and pick one or more .docx files.
3. If a file's result opens with a box headed Still in the cleaned copy, read that first: it names what was still found in the copy, and where.
4. Read the Before and after table for each file: comments, comment authors, tracked changes, change authors, hidden text, custom properties, saved file location, template, preview image, custom XML, photos with metadata, editing-session ids and printer settings.
5. Check the notes under the table, for example that tracked changes were accepted or hidden text was deleted.
6. Download the cleaned document and open it in Word to read it once before you send it.

The two choices for tracked changes

Accept all changes turns the document into its final version. Inserted text becomes ordinary text. Deleted text is taken out of the file, not just out of view. Reviewer names and dates go with the change history. Your document's wording changes as a result: anything a reviewer struck out is no longer there to restore. This is the default because it is the only way to remove both the history and the names.

Keep changes, hide who made them leaves every tracked change in place, so the recipient can still accept or reject each one. The reviewer shown for each change becomes "Author" and the real dates are taken out. In one place the Word format insists on a date, the marks around moved text; there the tool writes the same fixed date, 1 January 2000, and the result carries a note about it. Use it when you are sending a marked-up draft to the other side of a negotiation and want them to see the proposals but not who on your team made them. Keep in mind that the deleted text is still in the file in this mode, because it is part of a pending change.

Both choices also reach change records outside the running text: in tables, in style and list definitions, and in the conflict marks Word writes when two people edit one passage at once.

When the changes are kept, the recipient can accept or reject each one separately. The mixes that come out of those separate choices are not checked one by one, while the text each person typed is checked as it was typed.

If you would rather choose change by change, do that in Word first with Review > Accept or Reject, then clean the result here.

What else is removed

In both modes the tool also removes: the document and application properties listed above, together with the lists of headings and link targets Word stores beside them; custom properties; comments with their authors, the people list and tasks assigned from comments; editing-session ids; hidden text; custom XML; the preview thumbnail; the attached template path; document variables; mail-merge settings with their data source and recipient list; editing exceptions that name a single user; sensitivity-label data; settings that point to schema or transform files on a computer; saved printer settings; the original file names that older-style drawings keep with their pictures; and metadata inside embedded photos (the photos themselves look the same). It also sets the document's option that asks Word to remove personal information from file properties on save, so an author name is less likely to creep back in later.

What is shortened to a file name

Some places hold a whole folder path where the document only needs the file. C:\Users\dana\Reports\chart.png gives away a Windows user name; chart.png does not. The tool keeps the last name and drops the folders in front of it in three kinds of place: links to files on a computer or a network share; a picture description or name that consists of such a path and nothing else; and, inside a field code, the file that the field is told to fetch or link to (INCLUDEPICTURE, INCLUDETEXT, IMPORT, HYPERLINK, LINK, DDE, RD, and the data file of DATABASE). The Removed line then mentions folder paths. A link cut down this way looks for its file beside the document, so re-link it in Word if it still has to work.

A path anywhere else in a field code stays as written: on either side of an IF comparison, for example, or in text that a field prints. Changing it there could change what the document shows. A picture description that mentions a path among other words stays as well. Each of these is named as a folder path in the Still in the cleaned copy box, so you can edit it in Word yourself. When a field code itself carries tracked changes and you keep them, its current wording and its earlier wording are each shortened the same way; where the two cannot be lined up with certainty, the path is left and named in that box.

What is kept and listed in the result

A few things stay because taking them out would change what the document says or does. When one is present, the Still in the cleaned copy box names it, with a count and the part of the file it sits in:

• Digital signatures. The signer's certificate, with their name, and the signing time are kept. Cleaning alters the file, so Word may report the signature as invalid. If the name should not go out, remove the signature in Word and clean again.
• Fields that print the author, the file name or a date into the text, such as AUTHOR, FILENAME with its path or SAVEDATE, and content controls tied to the author or company properties. Their output is visible text, so it is left alone and counted.
• Hidden field data from add-ins such as reference managers, and field types the tool has no rule for.
• Text that is in the file but out of sight for another reason than the hidden font effect. Text with the hidden effect is removed. Text in a piece of an equation that is set not to show, WordArt whose text is turned off, and text in a shape, text box or chart that was switched off in the Selection Pane, stays where it is: taking it out would mean rebuilding the equation or the drawing. The box lists it as hidden text, with the part it sits in, so that you can open that place in Word and decide.
• Paths and intranet addresses in your own text. If the text you wrote contains something shaped like C:\Users\…, a network share, a home folder or an address inside a private network, the tool leaves it, since it is your content, and points it out, with the part of the file, in a Good to know note below the result. It does not single out every string with slashes in it: only those forms.
• Links to private network addresses, such as an intranet page, and links to outside files that are not ordinary web links, such as a linked picture.
• E-mail addresses behind links. A mailto link keeps the address it opens even when the text on the page shows something else or was edited later.
• Data kept by add-ins, macros and controls: what an Office add-in saved in the document, custom ribbon buttons, the settings of controls placed on the page, macro names and their shortcut keys.
• Handwriting and pen drawings (ink). Ink is part of what the page shows, so it stays. The data behind it can also record when each stroke was made and with which pen or device, and the box lists it for that reason.
• Parts and markup the tool does not recognize. It cannot vouch for them, so it lists them.

The box also names anyone the file's properties record (its author, whoever saved it last, the writer of a comment and the like) when that name turns up in the document's own text, visible or hidden. Check each one and decide whether it stays.

What it does not touch

The visible content of the document stays as it is, and so do a few things that can still identify someone: names typed in the text, headers, footers or a watermark; objects formatted as invisible, white text on a white background, or text covered by a shape; and embedded objects such as a spreadsheet pasted in as an object.

Microsoft's Document Inspector (File > Info > Check for Issues > Inspect Document) looks for some of these, such as headers, footers and invisible content. Running it after cleaning is a reasonable final check.

Only .docx files are accepted. Save older .doc files as .docx first. Files larger than 300 MB, and password-protected documents, cannot be read. A document whose text parts unpack to more than 64 MB each or 256 MB in all, whose main text holds over a million pieces of XML (every element, every comment and every run of text between two tags counts as one; far beyond a book-length manuscript), or whose XML is nested more than 512 levels deep (2,048 in a part that is read in one pass), is refused as too large so that the browser tab stays alive. So is a package of more than 10,000 entries, and a part with an element that carries over 2,000 attributes, namespace declarations included; neither comes near what Word writes. If a single piece of text inside a part runs past about 4 million characters, the document is still cleaned, but that part is listed as not checked. Pictures are outside the 256 MB count; each one can be up to 64 MB. Everything in the package together, pictures and parts the tool never opens included, may unpack to 512 MB at most; past that the file is turned away as well. For PDFs made from your document, use a tool that can clear PDF properties such as author and title; exporting a PDF from a cleaned .docx also helps. Photos inside the document are cleaned when they are JPEG, PNG, WebP or HEIC. Other embedded pictures (TIFF, GIF, EMF), embedded videos and embedded files, such as a spreadsheet pasted in as an object, are not opened: they stay as they are, and the result lists them so you can check them yourself. Templates (.dotx) and macro-enabled documents (.docm) are turned away; save a copy as .docx and clean that.

What this tool cannot do

• Look inside binary values. Document protection keeps a password hash and its salt as encoded binary data, and ink is stored as strokes. A hash or salt that is well-formed and as long as its algorithm makes it is not interpreted; one that is not is treated as plain text and checked. Ink is listed part by part.
• Search for a default name. When the author or company on record is exactly "administrator", "admin", "user", "owner", "guest", "author", "test", "unknown" or "microsoft", in any letter case, the property is removed, but the tool does not go looking for that word in the rest of the document: ordinary files are full of it for other reasons. A recorded name of one or two Latin letters or digits is not looked for either, nor is one without a single letter or digit, or an id made up only of zeros, and the result tells you so; a name in another script, Chinese for example, is looked for whatever its length. The exception is a comment author's initials: at one or two letters they are not looked for and no note appears, since the author's full name is looked for anyway.
• Find a name that is written in code. The search for the recorded author, initials and accounts reads each text as it is and once more with a short, fixed set of spellings undone: character references such as J, percent-escapes in a link address, + read as a space in the part of a link that follows its ?, \" and \\ inside a field code, the \u004A escapes of the JSON that reference managers put into field codes, and differences of letter case or invisible characters. Other codings are not undone. A name that an add-in stored as base64, for instance, is caught only where it also appears in plain letters.
• Recognize every way of writing a folder path in the middle of a sentence. A value that is a path and nothing else is recognized in almost every spelling, and the same holds for a field-code argument that stands whole between quotes. The two that are not: an old Mac path on a disk somebody named, unless it ends in a file name with its extension (Work Disk:Profiles:Alice:plan is passed over, Work Disk:Profiles:Alice:secret.docx is caught), and File: with a capital F when neither a slash nor a drive letter comes after the colon (File:Users). In running text the usual forms are recognized as well: C:\… or C:/…, \\fileserver\team\…, an address that starts with file:, /home/…, /Users/… and similar system folders, ~/…, %USERPROFILE%\… and the old Mac Macintosh HD:…. A rarer spelling inside a sentence may go unnoticed, for instance "see C:x\Alice\notes" (a drive with no slash after it), or a home folder or a server named with nothing after it, as in "see ~jane/ now".
• Read the pictures. Photos in the document lose their metadata, but nobody looks at what they show: a name on a badge or a screenshot of a chat stays as it is. Cover it with the redaction tool before you insert the picture.
• Remove the part of a picture you cropped away. A crop in Word only hides the edges: the full picture is still in the document, and the recipient can undo the crop, which is how cropped screenshots give away more than intended. The tool does not change pixels, so this stays after cleaning. In Word, select the picture, go to the Picture Format tab, click Compress Pictures and, in the box that opens, tick Delete cropped areas of pictures. Saving the cropped picture as an image and inserting that instead works too.
• See through every document built to dodge it. A .docx put together to smuggle data past the checks may hold a few forms the tool misses or leaves in: spellings and splits that Word never writes, parts of the file that contradict each other, and formatting settings combined in ways Word's own menus never produce.

If the result lists nothing, that means the tool found nothing it checks for. It does not mean the document says nothing about you. The limits of every tool on this site are set out on the methodology page.

Where your document goes

The document is opened and rewritten inside your browser tab. Its text, comments and names are never sent to a server, and the page keeps nothing after you close it. Confidential drafts can be cleaned here with no account and no upload.

How the Word cleaning works

The tool unzips the package in a background worker, using the fflate compression library, which the page loads from this site the first time you clean an Office file. It follows the package's own relationships to find every part, then parses the main document, headers, footers, footnotes, endnotes, styles, numbering, settings and comment parts as XML, matching elements by namespace rather than by the prefix a file happens to use. Text hidden through a style, not only through direct formatting, counts as hidden. Tracked-change elements are either resolved (insertions unwrapped, deletions dropped) or rewritten with the neutral author name and without their real dates (the fixed date stands in on moved text, where a date is compulsory). Comment ranges and references are removed along with the comment parts and their relationships. Names, dates, titles and link lists are cleared from the property parts (page and word counts stay), and the package is zipped again with fixed entry dates.

Then comes a second pass that shares no lists with the first. It opens the finished .docx from scratch and holds every part, relationship, element and attribute against the published definitions of the Word format. Where those say a spot can hold a person's name, an account id, an editing date, a path or add-in data and a value is still there, the result lists it; what the definitions do not cover at all is listed too. A part counts as checked only when it is known and found clean, so the result names what is in the copy instead of calling the file clean. The methodology page explains the approach and how to confirm the absence of uploads.

FAQ

Does Accept all changes really delete the struck-out text?

Yes. Deleted text is removed from the file, so it cannot be recovered from the cleaned copy. Your original keeps it.

Can the recipient see who made the changes in Keep changes mode?

No. Each change shows "Author" as the reviewer, and the day and time of every edit are gone. Moved text cannot go without a date in this format, so it shows 1 January 2000. The changes themselves remain visible and can be accepted or rejected.

Why does my result have a "Still in the cleaned copy" box?

Because the cleaned document still holds something the tool keeps on purpose or cannot vouch for, such as a digital signature, a field that prints the author's name, or a part it does not recognize. Each item gets a line with the part of the file it is in. Decide whether it matters, fix it in Word if it does, and clean the file again.

Is this the same as Word's Document Inspector?

It overlaps. Both remove properties, comments and tracked changes. This tool also removes editing-session ids and the metadata inside embedded photos, shortens the folder paths of linked files, shows a before-and-after table, names what it left in the file, and works in a browser on any system. Document Inspector additionally checks headers, footers and invisible content.

Will the cleaned file open in Google Docs or LibreOffice?

It is a standard .docx file, so it opens in any program that reads Word documents.

Can I clean a whole folder of Word files at once?

Yes. Select them together. Each file gets its own before-and-after table, and Download all (zip) saves them in one archive. The tracked-changes choice applies to every file in that run.

Does this remove names from the document text?

No. It removes metadata, not content. Search the text, headers and footers for names before you send the file. A name printed by a field, such as AUTHOR in a footer, stays as well; the result counts those fields so you know to look. And when a name the file recorded turns up in the text itself, the result lists it among what is still in the file; whether it stays is yours to check and decide.

Sources

Checked 2026-10-11.

• Microsoft Support: Accept or reject tracked changes in Word
• Microsoft Support: Remove hidden data and personal information by inspecting documents

Last updated: 2026-10-17