IPTC and XMP metadata, explained

Photo metadata is a subject that manages to be both simple and genuinely confusing, because three standards overlap, everyone uses the words loosely, and the tools disagree about where to put things.

Here is the version that is useful in practice.

The three things people mean by “metadata”

EXIF is what the camera wrote. Shutter speed, aperture, ISO, focal length, timestamp, GPS if the camera knew. It is machine-generated, factual, and you rarely edit it — although you might well want to strip it before publishing, since it can carry the location of the photographer.

IPTC is what a human wrote. Keywords, description, headline, creator, credit, copyright, location. It exists because news agencies needed a standard way to attach who, what, where and rights to a photo travelling between organisations. This is the layer that makes an image findable and legally safe — and the layer that is empty in most people’s libraries.

XMP is not a set of fields at all. It is a container — Adobe’s format for storing metadata inside a file (or beside it, in a .xmp sidecar). IPTC fields are typically written in XMP form.

So “IPTC vs XMP” is not a choice you have to make. The sentence that clears up most of the confusion: IPTC says what the fields are, XMP is how they are stored. In a modern workflow you write IPTC fields, in XMP, inside the file.

The fields that actually matter

There are dozens. In practice, eight do the work:

Field	What it is for
Keywords	The searchable terms. The single most valuable field, and the one most often empty.
Description (Caption-Abstract)	One or two sentences describing the image. Often reused as a caption or alt text.
Headline	A short title. Not a description — think newspaper headline.
Creator	Who made it. Survives the file being emailed around; your name in a filename does not.
Copyright Notice	The rights statement. Legally the most important line in the file.
Credit / Source	Who to credit on publication, which is not always the creator.
Location (City, Country)	Where it was taken. Essential for editorial, useless for studio work.
Rights Usage Terms	What the licensee may actually do with it. Prevents expensive misunderstandings.

If you only ever fill in two, make them Keywords and Copyright Notice. The first makes the image findable; the second makes it defensible.

Why your keywords keep disappearing

This is the complaint that brings most people to this topic, and there are four common culprits.

Someone exported a JPEG and stripped the metadata. Most “export for web” settings, most CMS uploaders, and almost every social platform strip metadata on upload — sometimes for privacy reasons, sometimes to shave kilobytes. Your keywords were there. The pipeline threw them away.

The metadata is in a sidecar and the sidecar got left behind. Raw files are often not written to directly; the metadata lives in a .xmp file next to the raw. Move the raw without the sidecar and you have moved a photo with no metadata. This is why raw workflows lose keywords when files are copied by hand.

Two tools wrote to two different places. The same conceptual field can exist in legacy IPTC-IIM, in XMP, and in Exif — and older tools do not always sync them. One program writes the description to XMP, another reads IPTC-IIM, finds nothing, and reports the field as empty. It was never empty; it was in a different drawer.

The catalogue has it and the file does not. Lightroom stores your edits and metadata in its own catalogue database. Unless you explicitly save metadata to file (⌘S / Ctrl+S in the Library module, or enable automatic writing), the keywords exist only in Lightroom. Send the file to a colleague and it arrives naked.

That last one catches out more professionals than any other single mistake in this article.

Embedded vs. sidecar vs. external

Three places metadata can live, each with a real trade-off:

• Embedded in the file. Travels with the image everywhere. Requires rewriting the file, which is fine for JPEG/TIFF and generally avoided for raw.
• In a sidecar (.xmp next to the raw). Non-destructive, standard for raw workflows, and fragile: it is one careless copy away from being separated from the image it describes.
• In an external database (your DAM, your catalogue). Fast to search, easy to bulk-edit, and invisible to anyone outside the system. Send the file out and the metadata stays behind.

Serious workflows use both: the DAM is the working copy, and the metadata is also written into the file so it survives leaving the building.

Writing metadata in bulk

Whatever generates your keywords, the write step usually goes through one of three routes:

• ExifTool. The command-line standard, and the only one that does everything. It reads a CSV and writes IPTC/XMP across a whole folder in one pass. Unglamorous, entirely reliable.
• Lightroom / Bridge. Fine for keywords you type yourself; bulk-importing metadata from a spreadsheet needs a plugin.
• Your DAM. Most support a bulk metadata import keyed on asset ID, and then push the metadata back into the files on export.

This is where AI keywording fits. A model can propose the keywords and the description for a whole shoot in minutes — the part that used to be an evening. PicsTag does that in your browser and exports a CSV keyed on your asset IDs, ready for the write step above. It deliberately stops short of rewriting your files: a web page has no business modifying six hundred raws on your disk, and you already have a tool you trust for that.

The one habit worth adopting

Write the copyright and the creator at import, automatically, before you have looked at a single frame. Most tools can apply a metadata preset on ingest. It takes one afternoon to set up and it means that no image you ever shoot again leaves the building anonymous.

Keywords can be added later — with help. Provenance cannot be reconstructed once the file has been out in the world for two years.