What a video file gives away about you

An MP4 or MOV is a tree of boxes, and the picture is only some of them. The rest names the camera, the operating system version, the editing software, the moment it was shot to the second with its timezone offset, and where you were standing. On an iPhone that last one is frequently not a single reading but a track — a position recorded every second or so for the length of the clip, which is a route rather than a place. AVSCRUB opens the file in your browser, shows every field and where in the file it sits, and writes a copy without them. Nothing is uploaded, which for a video is not a preference.

What is actually in there

These are the fields a phone and an editor leave behind, with the box each lives in. The values are from a file opened in AVSCRUB; yours will have the same shape.

The location string is ISO 6709: latitude, longitude, altitude. Six decimal places is a few metres. The timezone offset on the date tells a reader which part of the world you were in even if the coordinates are gone, and the OS version narrows down which phone among the millions.

The part almost nothing mentions

Removing those fields is a solved problem and several tools do it. There is a second copy of the location that most of them leave behind, and it is the more revealing one.

iOS records position into a timed metadata track: a whole track sitting alongside the video and the audio, whose samples are coordinates rather than pictures, written every second or so for as long as the camera is running. A single fix says where you started. A track says where you went, how fast, and whether you stopped.

That matters practically because the two live in different parts of the file. The fields above are boxes in front of the media, and removing one means deleting a range of bytes and repairing the tables that point past it. A track's samples are inside the media data, interleaved with the picture, and taking them out means rebuilding the media itself. So a metadata stripper can remove the location field, show it struck through, report the job done, and leave the whole route in the file.

AVSCRUB reports the track when a file has one, marks it as something it does not remove, and says so again in the status line after stripping — because the moment somebody reads "the entries are gone" is the moment they decide the file is safe to send. If that is the location you were trying to remove, re-encode the video: a re-encode writes a new file containing only the tracks you asked for.

What survives a strip, even a good one

• The movie and track timestamps. mvhd and tkhd carry creation and modification times, and a player reads those boxes, so they are structure rather than decoration. They still date the file. AVSCRUB lists them and marks them as staying rather than quietly keeping them.
• A timed metadata track, as above.
• The encoder's own habits. Quantisation choices, frame structure, how the encoder handles a scene change: these are properties of the compressed picture, not fields anywhere, and no amount of editing metadata touches them. They do not name you, but they do say which family of encoder produced the file, which is sometimes more than you meant to say.

Who gets the original, and who does not

This is the part that decides whether any of it matters.

Anything that re-encodes your video to its own settings — which is what a social platform does, to make one file it can serve at several sizes — writes a new file, and the container metadata does not survive that. Anything that hands the file across as it is passes on everything: email attachments, AirDrop, a shared cloud folder, a messaging app sending it as a document rather than as a video, a memory card given to somebody.

So the risk sits exactly where people feel safest. A clip posted publicly to a platform has usually been rebuilt. The same clip sent to one person you trust, as the original, has not.

The metadata is the easy half

Every field above can be removed in a few seconds. What cannot is what is in the frame.

A reflection in a window or a pair of glasses. A street sign, a bus number, a shop across the road. A school uniform. The view from your own front room, which is recognisable to anybody who has been in it. A notification sliding across the top of a screen recording, with a name and half a message in it. A clock on the wall. In the audio: a doorbell, a train announcement, a church bell, an accent, somebody saying a name.

None of that is metadata, no tool removes it, and in practice it identifies people far more often than a coordinate does. Strip the fields by all means — it costs nothing and closes the easiest route. Then watch the clip through once, with the sound on, as somebody who does not know you.

Alongside: X-RAY does the same reading for documents and photographs, the same question for a still photograph has a simpler answer, and was this recording edited is the other direction — reading a file somebody sent you rather than one you are about to send.