Document Metadata Inspector
Inspect the hidden metadata inside a PDF, Word document, or image: creation and editing history, revision signals, and embedded AI-generation tags. Free, browser-only file forensics.
The Document Metadata Inspector reads the metadata already embedded inside a file rather than analyzing the visible content: PDF properties, Word document editing history, and image provenance tags. Upload a PDF, DOCX, PNG, or JPG file up to 10MB and the tool extracts and displays whatever metadata the file actually contains - nothing is guessed or inferred from the visible content.
For PDFs, the tool reads the Producer and Creator application fields, creation and modification dates, and any XMP metadata. For Word documents (.docx), it reads editing-session forensics stored inside the file format itself: RSID (revision save ID) values that indicate how many distinct editing sessions produced the document, the presence of tracked changes or comments, and the core.xml properties recording author, last-modified-by, and revision count. For images, it checks three separate signal sources: EXIF metadata (which can include a Software or Make tag left by tools like Midjourney), PNG text chunks (Stable Diffusion and tools such as Automatic1111 commonly embed a full generation prompt and parameters in a tEXt or iTXt chunk), and an optional C2PA Content Credentials check, an open standard several AI image generators and camera makers use to attach a tamper-evident record of how an image was created or edited.
Metadata can be a strong signal or no signal at all, and the tool shows both honestly. An embedded Stable Diffusion prompt or a Midjourney software tag is a reasonably strong indicator because those tools embed it automatically and it rarely appears by accident. But the complete absence of any metadata proves nothing: screenshots, re-saves, exports through Google Docs or a school's learning management system, messaging apps, and many editing tools routinely strip metadata as a side effect. Treat a clean, metadata-free result as inconclusive, never as confirmation of human authorship.
This is a metadata reader, not a full forensic document examiner, and results should never be the sole basis for an academic, employment, or legal decision. Everything happens locally in your browser - your file is never uploaded to a server, and nothing is stored once you close the tab.
Frequently Asked Questions
- If a file has no metadata at all, does that prove it was not AI-generated or was not edited?
No. Missing metadata is common and usually meaningless. Screenshots, PDF re-saves, exports from Google Docs or a learning management system, and many messaging apps strip metadata automatically as a normal part of how they handle files, regardless of how the content was created. Treat a metadata-free result as inconclusive, not as evidence of human authorship or an unedited document.
- How reliable is the RSID and tracked-changes information from a Word document?
RSID values and revision counts reliably reflect the editing history Word actually recorded, but they describe editing behaviour, not authorship or intent. A high RSID count suggests many distinct editing sessions, typical of a document drafted and revised over time, but a single-session document is not automatically suspicious either. These are supporting details for context, not a verdict on how a document was written.
- What is C2PA and how strong a signal is it?
C2PA (Coalition for Content Provenance and Authenticity) is an open industry standard some AI image generators and camera makers use to attach a tamper-evident Content Credentials manifest recording how an image was created or edited. A valid manifest naming an AI generation tool is one of the stronger signals this tool can surface. However, most images circulating online have never had a C2PA manifest attached in the first place, and a manifest can be stripped by re-saving or converting the file, so its absence tells you nothing about how the image was made.