The principle behind image metadata applies unchanged to documents: the metadata is a separate part of the file, not the content. A contract and its author field are stored in different places. That is why you can send a document that looks anonymous and still hand over the name of the person who drafted it, the company they work for, and every earlier version they deleted.
The containers differ, so the hiding places differ. Three families cover most of what people actually exchange.
Metadata is a separate part
An image is a stream of segments with an Exif block bolted on. A PDF is a graph of objects with a document catalogue, and some of those objects are metadata. An OOXML or OpenDocument file is a ZIP archive containing XML parts, and some of those parts are metadata.
In every case the content survives if the metadata part is removed or rewritten. That is the useful part of the analogy, and also the limit of it: unlike an image's compressed pixel data, document content is usually re-encoded when you save from an editor, so "remove metadata without touching the content" is a container-level operation, not a save-as.
PDF: info, XMP, attachments
Every PDF can carry a document information dictionary, reachable from the trailer and usually
visible in a reader's properties dialogue. It holds /Author, /Creator,
/Producer, /CreationDate, /ModDate, /Title,
/Subject and /Keywords. The producer field names the software that
generated the file, which can reveal a scanning service, a typesetting system or an internal
publishing tool.
Modern PDFs also carry an XMP packet — an XML document inside the file describing the document ID, the creator tool, and a history of transformations. The two sources can disagree. A file whose info dictionary was cleaned may still have an XMP packet naming the original author, and vice versa, which is why checking one and not the other gives false confidence.
Two more places are easy to forget. PDFs can embed page thumbnails, and those images can show content that was later removed from the visible page if the file was assembled carelessly. PDFs can also carry arbitrary file attachments in an embedded-files name tree — a spreadsheet, a source image, an email export — that no reader shows you by default.
Office files and tracked changes
A .docx, .xlsx or .pptx is a ZIP. Inside it,
docProps/core.xml holds dc:creator, cp:lastModifiedBy,
cp:revision, dcterms:created and dcterms:modified.
docProps/app.xml holds the application name and version, and often
Company, Manager, Template and total editing time.
docProps/custom.xml can hold properties added by templates and add-ins.
The content parts carry more. Comments live in word/comments.xml with an author
attribute per comment. Tracked changes are stored as insertions and deletions marked with an
author and a timestamp, which means text you deleted is still in the file until the changes are
accepted and the document is saved again — and sometimes after that, in leftover revision data.
OpenDocument and thumbnails
An .odt, .ods or .odp is also a ZIP. Its
meta.xml part holds meta:generator, dc:creator,
meta:initial-creator, meta:creation-date, dc:date,
meta:editing-cycles, meta:editing-duration and document statistics.
The initial creator is frequently the original author even when the current creator has been
changed, so a file that looks reassigned can still name the first person who wrote it.
OpenDocument packages normally include a Thumbnails/ folder with a rendered preview
of the first page. That preview is content, not just metadata: it can show a title, a client
name, or a sentence that was cut from the final version.
| Format | Where the metadata lives | Common fields |
|---|---|---|
| Info dictionary, XMP packet, thumbnails, attached files | Author, Producer, CreatorTool, creation and modification dates, document IDs |
|
| DOCX / XLSX / PPTX | docProps/core.xml, app.xml, custom.xml, comments, revision marks |
Creator, last modified by, revision count, company, manager, comment authors |
| ODT / ODS / ODP | meta.xml, Thumbnails/, settings.xml |
Initial creator, generator, edit cycles and duration, page preview |
What documents give away
A CV sent to a recruiter can carry the name of the previous employer's template author. A contract draft can carry the law firm's company field and the exact hour it was last edited. A spreadsheet of figures can carry comments in which a colleague explains what a row really means.
For a journalist, the pattern is worse: a document received from a source can carry the source's
name in dc:creator, a company in app.xml, and a revision count that
shows how many times it was edited before it reached you. Publishing the file as received
publishes that trail.
None of it is visible on the page, and most of it survives being opened and re-saved, because editors preserve properties by default. The broader pattern is covered in photo metadata privacy risks; the same logic of separate parts and inherited fields applies here.
What MetaPurge handles today
Being straight about scope: MetaPurge currently processes images only — JPEG, PNG, WebP, GIF, TIFF, ICO, SVG and AVIF input. It does not read PDF, Office or OpenDocument files, and it does not read HEIC yet. Dropping a document into the tool will not produce a cleaned document.
Document support is on the roadmap, listed there as planned rather than shipped. The roadmap section on the home page is the place to check what has actually changed; treat any date you see elsewhere as an intention, not a promise.
Until then, the honest workflow for documents is manual: open the file's properties, inspect the archive if it is a ZIP-based format, remove comments and revision marks before export, and verify the result with a parser you trust rather than with the application that produced it. The method is the same one described in verifying metadata removal, which also explains why a clean-looking properties dialogue proves very little. For the image side of the same problem, start with what EXIF metadata is.
Start with images you can clean
Drop a photo in. MetaPurge lists every tag, then rewrites the file without touching a pixel. Documents are on the roadmap.
Open the metadata remover