Metadata by format: a comparison of all 10 file types
Published 19 August 2026
Ten formats, two families, and a wide range of what each one can actually hide. This is the full reference — the image hub and document hub each cover their own family in more depth; this page is the side-by-side view across all ten.
| Format | Can carry | Cleanup |
|---|---|---|
| JPG | EXIF (GPS, serials, timestamps), IPTC, XMP, a hidden thumbnail | Byte-level, hash-verified |
| PNG | Text chunks (author, XMP), a full EXIF block, timestamps, crop geometry | Byte-level, hash-verified |
| WebP | EXIF and XMP chunks, usually inherited from a source JPEG or PNG | Byte-level, hash-verified |
| GIF | Comment blocks, occasionally an XMP packet | Byte-level, hash-verified |
| TIFF | The same EXIF and GPS data as a JPEG, at absolute byte offsets | Rebuilt (random-access format), hash-verified |
| BMP | Nothing standard — only data appended past the declared file size | Byte-level, hash-verified |
| The Info dictionary and a separate XMP metadata stream | Re-serialized, field list reported | |
| Word (.docx) | Properties, comments, tracked-change authorship, an embedded thumbnail | Archive rebuilt, field list reported |
| Excel (.xlsx) | Properties, cell comment threads, an embedded thumbnail | Archive rebuilt, field list reported |
| PowerPoint (.pptx) | Properties, comment threads, template origin, an embedded thumbnail | Archive rebuilt, field list reported |
Why "cleanup" isn't the same for every row
The six image formats share one property regardless of how different their metadata systems are: none of them require decoding pixels to remove metadata, so all six get byte-level surgery — copy every range except the metadata, verified by hashing the compressed image data before and after. TIFF is the exception inside that group, not because its guarantee is weaker, but because its random-access structure means achieving the same byte-identical pixel result takes a full rebuild of the tag directory rather than a simple copy.
PDF and the three Office formats can't offer that same hash comparison at all, and that's a structural fact rather than a shortcoming: a PDF is parsed and re-serialized, and a .docx/.xlsx/.pptx is a ZIP archive that has to be rebuilt from its parts rather than copied by byte range. What you get instead is the same transparency in a different shape — an exact list of which parts changed and why, rather than a hash that a legitimate rebuild would never match regardless of whether the cleaning was correct.
Which format leaks the most
For photos, it's JPG and TIFF — both can carry full EXIF including GPS to a few metres and hardware serial numbers, the most identifying combination any format here supports. For documents, it's Word: the properties risk is shared across all three Office formats equally, but only Word's tracked-changes mechanism can leave literal deleted sentences, attributed and timestamped, sitting in a file that looks finished. BMP and GIF sit at the opposite end — both predate, or never adopted, any real metadata mechanism, and carry close to nothing by design rather than by cleaning.
One engine, both families
All ten formats run through the same sandboxed frame and the same no-upload guarantee — pick a format above, or start from the homepage if you're not sure which one you're dealing with yet.