Metadata is data that describes other data. That is the full answer to "what is metadata", and it sounds vaguer than it is. In a photo, the data is the grid of pixels. The metadata is what sits next to those pixels and tells you about them: which camera took the picture, at what shutter speed, on what date, sometimes where, and who owns it.
The old example is the library catalog card. Title, author, call number, year. None of it is the book, yet without it you would never find the book on the shelf, or know whether you were allowed to copy it.
The classic types of metadata
Librarians and archivists split metadata into three kinds. The split predates digital cameras by decades and still maps onto image files reasonably well.
| Type | Question it answers | Photo examples |
|---|---|---|
| Descriptive | What is this, what is in it? | Caption, keywords, headline, location name |
| Structural | How is the file built? | Width, height, compression, color components |
| Administrative | How was it made, who owns it? | Camera, exposure, dates, copyright, license URL |
Administrative metadata is often subdivided again into technical metadata (camera and exposure settings) and rights metadata (copyright, licensing), with provenance sometimes counted on its own.
In practice the boxes leak. EXIF, the standard your camera writes, is mostly technical, yet it has a Copyright field. IPTC is mostly descriptive and rights-oriented, yet it carries a creation date. Don't lose sleep over the categories. The tag names are what you will actually search for.
What is metadata in a real photo file?
Definitions only go so far, so here is an actual JPEG. I generated a 2000 × 1500 image with Python's PIL, which writes nothing beyond a small JFIF header, then added the tags a camera and a photographer would typically leave behind:
exiftool -Make=Canon "-Model=Canon EOS R6" -FNumber=8 -ExposureTime=1/250 -ISO=100 "-DateTimeOriginal=2026:05:14 18:42:07" -GPSLatitude=36.998 -GPSLatitudeRef=N -GPSLongitude=110.098 -GPSLongitudeRef=W "-Artist=Jane Doe" "-Copyright=© 2026 Jane Doe" photo.jpgReading it back with group names (-G1), short tag names (-s), and -a so ExifTool doesn't hide the GPS tags behind Composite tags of the same name:
exiftool -G1 -s -a -Make -Model -FNumber -ExposureTime -ISO -DateTimeOriginal -GPS:all -Artist -Copyright -FileModifyDate photo.jpg[IFD0] Make : Canon
[IFD0] Model : Canon EOS R6
[ExifIFD] FNumber : 8.0
[ExifIFD] ExposureTime : 1/250
[ExifIFD] ISO : 100
[ExifIFD] DateTimeOriginal : 2026:05:14 18:42:07
[GPS] GPSVersionID : 2.3.0.0
[GPS] GPSLatitudeRef : North
[GPS] GPSLatitude : 36 deg 59' 52.80"
[GPS] GPSLongitudeRef : West
[GPS] GPSLongitude : 110 deg 5' 52.80"
[IFD0] Artist : Jane Doe
[IFD0] Copyright : © 2026 Jane Doe
[System] FileModifyDate : 2026:05:15 09:12:40+02:00What that output tells you:
- The bracketed group is where each tag physically lives. IFD0 is the main EXIF directory, ExifIFD holds exposure data and the capture date, GPS is a separate sub-directory.
- GPS coordinates are stored as degrees, minutes and seconds, with the hemisphere in a separate Ref tag. Lose the Ref and 110° W becomes 110° E, which moves this photo from Arizona to central China.
exiftool -n -GPSPosition photo.jpgprints the decimal form,36.998 -110.098, which is what you paste into a map.- FileModifyDate sits in the System group because it is not stored in the file at all. More on that below.
ExifTool also added ExifVersion (0232), ComponentsConfiguration and FlashpixVersion without being asked, because the EXIF spec marks them as mandatory. A full exiftool -G1 -a -s photo.jpg shows them, plus the file-level information ExifTool computes on the fly.
Everything in that file was typed by me, and no Canon EOS R6 was involved. Plain EXIF is trivial to forge, which matters later.
Embedded metadata vs file-system metadata
Two different things get called metadata, and mixing them up explains most "my photo dates are wrong" complaints.
Embedded metadata lives inside the file's bytes: EXIF, IPTC, XMP. It travels with the file wherever it goes, unless some software rewrites or strips it.
File-system metadata belongs to the disk: file name, size, permissions, modification time. ExifTool shows it under System (FileModifyDate, FileAccessDate). Copy the file, save it from an email, or extract it from a zip, and the operating system usually gives the copy a fresh modification time.
I copied photo.jpg with a plain cp to stand in for saving an email attachment:
======== photo.jpg
[System] FileModifyDate : 2026:05:15 09:12:40+02:00
[ExifIFD] DateTimeOriginal : 2026:05:14 18:42:07
======== photo-from-email.jpg
[System] FileModifyDate : 2026:10:07 09:07:23+02:00
[ExifIFD] DateTimeOriginal : 2026:05:14 18:42:07The capture date inside the file did not move. The file-system date jumped to the moment of the copy. Software that sorts by "date modified" will file this photo under October; software that reads DateTimeOriginal will file it under May. ExifTool itself bumps FileModifyDate every time it writes a file, unless you pass -P. If the embedded date is the wrong one, see how to change the date taken on photos.
Where metadata sits inside a JPEG
A JPEG is a chain of segments, each introduced by a two-byte marker. Metadata goes in the APPn segments ahead of the compressed image. exiftool -v photo.jpg lists them:
JPEG APP0 (14 bytes):
JPEG APP1 (456 bytes):
JPEG DQT (65 bytes):
JPEG DQT (65 bytes):
JPEG SOF0 (15 bytes):
JPEG DHT (29 bytes):
JPEG DHT (179 bytes):
JPEG DHT (29 bytes):
JPEG DHT (179 bytes):
JPEG SOSAPP0 is the JFIF header and APP1 holds the EXIF block. In files that have them, a second APP1 holds XMP, APP13 holds IPTC, APP2 the ICC color profile, and APP11 any C2PA manifest. After SOS comes the pixel data. A single segment cannot exceed 64 KB, which is why EXIF has a hard size ceiling in JPEG and long XMP gets split into extended XMP. The JPEG format page goes through each segment.
The practical upshot: removing metadata from a JPEG means dropping segments, not touching pixels. A decent stripper doesn't re-encode the image.
Why photo metadata matters
Search and asset management come first for anyone with a large archive. Lightroom, Photo Mechanic and DAM systems index captions, keywords and dates. Without them, 40,000 images are a folder of names like IMG_4821.jpg.
Rights information travels with the file. Google Images reads the IPTC Creator, Credit Line and Copyright Notice fields and shows them under "Image credits", and it uses Web Statement of Rights for its Licensable badge. The EXIF vs IPTC vs XMP guide explains which standard holds which field.
Privacy is the one most people hit by accident. A phone camera with location permission writes GPS into every shot, so a photo of your front door carries your address. You can check what a JPEG, PNG or WebP carries with the EXIF reader, and drop it all with the metadata remover.
Verification is the newer use. Fact-checkers compare a photo's claimed time and place with its EXIF, but since EXIF can be forged in one command (see above), signed provenance such as C2PA Content Credentials carries more weight.