Image metadata and privacy — what your photos reveal

Every photo you take carries data you cannot see. Some of it is useful and some of it is a quiet privacy leak, and most people publish both without knowing the difference. This guide is about knowing the difference.

The three kinds of hidden data

A photo file can carry three separate layers of metadata, and they are not the same thing:

EXIF — written automatically by the camera or phone. Shutter speed, aperture, ISO, lens, the exact timestamp, and — the one that matters for privacy — GPS coordinates, if location was on. This is machine data you did not choose to add.

IPTC — written by a human or a tool. Keywords, description, creator, copyright, credit. This is the useful layer, the one that makes an image findable and attributable.

XMP — a container format (Adobe’s) that modern tools use to store IPTC-style fields inside the file.

For the full breakdown of how these interact, see IPTC and XMP metadata, explained. Here we care about one question: what is safe to publish?

The privacy risk is mostly EXIF GPS

The headline risk is simple and specific: a photo taken with location enabled can carry the exact coordinates of where it was taken. Publish that photo unmodified and anyone can read the coordinates out of the file.

The obvious cases are the dangerous ones:

• A photo of something for sale, taken at home, revealing the seller’s home address.
• A picture of a child, geotagged to their school or house.
• “Working from a café” shots that map a person’s daily routine.
• Journalists, activists or sources whose location is exactly the thing that must stay private.

The timestamp and camera serial number are lesser concerns, but they are there too: a series of photos can be correlated by camera serial even across accounts.

What most platforms do (and don’t) strip

Do not assume you are covered. Behaviour varies and it is not in your control:

• Most large social platforms strip EXIF on upload — partly for privacy, partly to save bytes. This also strips your copyright and keywords, which is a different problem.
• Many CMSes and “export for web” pipelines strip everything, useful and sensitive alike.
• Direct file sharing, email attachments, cloud drive links, and plenty of smaller sites strip nothing. The original file, GPS and all, goes straight through.

The only safe assumption is that you do not know, so you handle it yourself before publishing.

What to strip and what to keep

The goal is not to nuke all metadata — that throws away the useful layer with the risky one. The goal is selective:

Strip before publishing:

• GPS coordinates (unless the location is deliberately part of the content — a travel or real estate photo where the location is the point).
• Precise timestamps, if a routine could be inferred.
• Camera serial numbers, for anyone who wants images uncorrelatable.

Keep:

• Copyright and creator — this is how you stay attributable and defensible. Stripping it is how your work ends up uncredited.
• Keywords and description — this is what makes the image findable wherever it lands.

ExifTool is the standard for doing this precisely (exiftool -gps:all= *.jpg removes location while leaving everything else). Most photo editors have a “remove location” or “export without metadata” option, though the blunt ones remove your copyright too — check what you are actually stripping.

Where AI tagging fits — and the upload trap

Here is the part that connects to this whole site. The moment you decide to add useful metadata with an AI tool — keywords, descriptions, alt text — you usually hit a new privacy problem, bigger than the EXIF one: almost every AI tagging service works by uploading your images to its servers.

So to add keywords, you send the full-resolution original — GPS, faces, unpublished content and all — to a third party, to be processed and, often, retained. You have solved the metadata-quality problem by creating a much larger data-exposure one.

This is the entire reason PicsTag runs the AI in your browser instead. The model is downloaded to your device and runs there; your images are read from disk into memory and never uploaded, because there is no server to upload them to. You can verify it — open your browser’s network tab while you tag a batch and watch nothing leave. You get the useful metadata layer without handing your originals to anyone.

The difference between local and cloud AI processing, and when each is the right call, is covered in local vs cloud AI image tagging.

The two-minute habit

Two settings save most of the risk:

1. Decide, deliberately, whether your camera or phone should record location at all. For most people, most of the time, off is the right default — you can turn it on for the trip where you actually want it.
2. Before publishing anything sensitive, strip GPS. Keep copyright. It is one command or one export setting, and it is the difference between sharing a photo and sharing your address.