Answer
Do all photos have metadata?
Almost every photo straight from a camera or phone does. Images that reach you by other routes often carry very little, and some carry nothing at all.
Updated
A photo taken on a phone or a dedicated camera is written with a substantial metadata block as a matter of course: camera make and model, lens, exposure settings, orientation, a timestamp, a colour profile, an embedded thumbnail, and on a phone with location services enabled, GPS coordinates. None of that is optional in normal use, and most people never see any of it.
Images that arrive by other routes are a different story. A screenshot has only what the screenshot pipeline wrote. An image exported from design software has whatever that software chose to include, which is often just a colour profile. A photo downloaded from a social network has usually been re-encoded on upload, which discards the original block — the public copy is genuinely close to empty.
AI-generated images sit somewhere unexpected on this scale, because they frequently carry more than a camera photo rather than less. Stable Diffusion tools write the full prompt, negative prompt, seed, sampler and model hash into a text chunk. Several hosted generators attach content credentials. That is a rich record of exactly how the image was made.
The practical consequence is that you cannot tell by looking. Two visually identical files can differ by several kilobytes of hidden text, and the only way to know which one you are holding is to open it in a viewer and look.
It is also why "this format does not support metadata" is a claim worth distrusting. All nine formats this site handles have somewhere to put it, and where each one puts it differs enough that a tool built for JPEG will quietly miss it everywhere else.
More on this: Metadata remover, What is EXIF metadata?, Supported formats.
Terms used here: EXIF, C2PA, Embedded thumbnail, Chunk.