What Your Documents Reveal: Hidden Metadata in Word, PDF and Audio Files
Updated September 2026
Photos are not the only files that talk. A Word document records who created it, who last saved it, the company name on the license, how many minutes it was open and an ID for every editing session. A PDF names the person and the software. An MP3 names the encoder and whatever the exporting app felt like adding. Here is what each format carries, what AI tools write into it, and how to remove it cleanly.
Quick answer
The fields that most often give a file away:
- 1DOCX: creator and lastModifiedBy in docProps/core.xml, Application, Company and TotalTime in docProps/app.xml, rsid session IDs on every paragraph, reviewer names in comments.
- 2PDF: Author, Creator, Producer and dates in the Info dictionary, the same again in the XMP stream, and the trailer ID pair.
- 3MP3: ID3v2 frames including encoder settings (TSSE), encoded-by (TENC), custom TXXX fields, comments and embedded art, plus the ID3v1 trailer.
- 4All three: hidden Unicode characters inside the text, where there is text.
Drop a file below to see its own list.
Word, Excel and PowerPoint
An Office file is a zip archive of XML parts. Most parts are content. A few are about you. docProps/core.xml carries the creator, the last person to save, the title, subject and keywords, the revision count and the created, modified and last-printed timestamps. docProps/app.xml carries the application and version, the company name from the Office license, the template the document started from and the total editing time in minutes. If a document management system touched the file, docProps/custom.xml and a customXml folder carry its tags. Comments come with the names of everyone who left one.
The quiet one is rsid. Word generates a random revision save ID for every editing session and stamps it on the text written in that session, then lists them all in settings.xml. Two files that share rsid values share history. Document Inspector does not remove them.
| Field | Typical value | Why it matters |
|---|---|---|
| dc:creator | A person's name, or a tool name such as ChatGPT or PptxGenJS when a generator made the file | Names the origin |
| cp:lastModifiedBy | The last Windows or Mac account name to save | Names the last editor, often a different person |
| Company | The organization on the Office license | Reveals the employer or client |
| TotalTime | Minutes the document was open for editing | Reveals how long the work took |
| w:rsid attributes | Hex IDs on paragraphs and runs | Fingerprints editing sessions across files |
| Comments and people.xml | Reviewer names and comment text | Reveals who reviewed |
A PDF carries its properties twice. The Info dictionary is the classic block: Title, Author, Subject, Keywords, Creator, Producer, CreationDate and ModDate. The XMP metadata stream repeats those in XML and adds the creator tool, an edit history and a DocumentID and InstanceID pair that ties versions together. The trailer holds its own /ID pair. Producer is the field people forget: Word, Google Docs, Canva, Acrobat, macOS, LibreOffice and every PDF library write their own name and version there, and a PDF built by a script announces the library that built it.
Removing the metadata cleanly means deleting the Info keys and the object, unlinking and deleting the XMP streams on the catalog and every page, writing fresh IDs and re-serializing the file so that no dead objects from earlier saves remain readable. Appending an edit, which is what some tools do, leaves the old values recoverable.
MP3, WAV and FLAC
Audio tags are less personal and more revealing about tooling. An ID3v2 block at the start of an MP3 carries title, artist and album, but also TSSE, the encoder settings frame that records the software and version that wrote the file; TENC, encoded by; TXXX, custom key and value pairs where exporters store their own identifiers; COMM comments; PRIV private frames; USLT lyrics; and APIC, the embedded cover art. A 128-byte ID3v1 tag often duplicates the basics at the end of the file. WAV files carry a LIST INFO chunk with software and artist fields, sometimes Adobe XMP and broadcast metadata. FLAC carries a Vorbis comment block.
Music generation tools tend to write their name into one of these frames. Removing the tag blocks removes that record. It does not touch the audio itself, so an inaudible watermark in the waveform, if the generator embedded one, remains, and no tag remover can change that.
Text inside documents carries text problems
The body of a DOCX, the shared strings of a spreadsheet and the slides of a deck are text, and text pasted in from a chat interface or a web page brings its invisible characters with it. StripShot runs the same glue-aware Unicode cleaner over document text that it runs over pasted text, so a cleaned Word file is clean in both senses.
What a clean strip keeps
- Everything you can see. Text, formatting, styles, headers and footers, images, slides, worksheets, PDF pages, fonts and links, audio frames.
- File compatibility. The output is the same format and opens in the same applications.
- A record of what changed. The re-scan lists anything left, and the JSON report lists what was found and what was removed.
Two honest gaps: tracked changes in a Word file keep the author on each change until you accept or reject them, and images embedded inside a document keep their own EXIF until you strip them separately. PDF annotations keep the commenter's name. Each format page on this site says so.
Scan a document
See what a DOCX, PDF or MP3 says about you
Runs in your browser. Nothing is uploaded.
Drop a file here, or click to choose
DOCX, XLSX, PPTX, PDF, MP3, WAV, FLAC
Options
Need unlimited strips?
Scans are free. Pro is unlimited.
Images, video, audio, documents, PDF and text in one tool. Pro adds unlimited strips, whole-folder batches and zip download with a JSON report.
Frequently asked questions
Does a Word document show it was written with ChatGPT?
It can. If the file was generated by a tool or template, docProps/core.xml may list the tool as the creator, and the application field in docProps/app.xml names the program that saved it. If you pasted ChatGPT text into Word, the properties name you and Word, but the body text may carry hidden characters and typographic tells from the paste. Scan the file to see both.
What does a PDF reveal about me?
The Info dictionary and XMP stream typically hold your name as Author, the application that made the content as Creator, the library that wrote the PDF as Producer, and creation and modification dates. The trailer ID pair links every saved version of the same file. Annotations carry the commenter's name.
What metadata is in an MP3?
ID3 tags at the start hold title, artist, album, comments, lyrics, embedded artwork, the encoder settings and any custom fields the exporting app added. A legacy 128-byte ID3v1 tag often sits at the end, and some files carry APEv2 or Lyrics3 blocks too. AI music tools can write their name into these.
Does Word's Document Inspector remove everything?
No. It removes the properties and comments it knows about but leaves the rsid session IDs stamped on every paragraph, which fingerprint the document's editing history, and it does not look for invisible characters in the text.
Does stripping metadata change my document?
No. StripShot rewrites the package or container without the metadata parts. Text, formatting, images, pages and audio frames are carried over as they are. The output is re-scanned so you can see exactly what is left.