IPTC digital source type for AI images

01What is the IPTC digital source type field?

The IPTC digital source type is a field in the IPTC Photo Metadata Standard that records the source from which a digital image was created. Its value is not free text. It must be a term from a controlled vocabulary that IPTC publishes as NewsCodes for Digital Source Type, and each term has a URI such as http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia.

The field is older than generative AI. The IPTC Photo Metadata User Guide explains that it was added in 2008 to tell camera captures from scans of film and prints. In 2022 the vocabulary was extended for generative AI and synthetic media, and in 2024 IPTC added terms, retired ambiguous ones and improved labels. The guide also states the vocabulary can describe video, audio and text, not only images.

One technical detail matters later. The user guide explains that fields added in the IPTC Extension schema after 2005 exist only in XMP, and Digital Source Type is one of them. Tools that read only the older IIM block of a JPEG will not see it.

The user guide's own worked example for a generated image pairs the field with a Credit Line that names the person and the tool used. That is a sensible habit: the controlled value tells machines how the image was made, and the free-text credit tells people who made it and with what.

02The field and its values

The vocabulary has about twenty terms, some of them retired. Most sites only need the handful that separate camera images from generated or AI-edited ones. The table quotes IPTC's own labels from the Digital Source Type NewsCodes page as read in November 2026.

Two older terms are retired on the same NewsCodes page: softwareImage, retired in June 2022, and minorHumanEdits, replaced by humanEdits. If an old workflow still writes them, update it.

03trainedAlgorithmicMedia vs compositeWithTrainedAlgorithmicMedia

These two values cause most labelling mistakes. trainedAlgorithmicMedia means the image was created by an AI model trained on captured content: there was no original photograph, only the model's output. compositeWithTrainedAlgorithmicMedia means a real image was augmented, corrected or enhanced using a generative model, as IPTC's definition puts it, such as with inpainting or outpainting operations.

A simple test helps. Ask whether a camera captured the main subject. If not, use trainedAlgorithmicMedia. If a camera did and a generative tool then removed, added or extended part of the scene, use compositeWithTrainedAlgorithmicMedia. If you combined several separate images and one of them was generated, compositeSynthetic is the closer fit.

• Product photo with a generated background replacing the studio wall: compositeWithTrainedAlgorithmicMedia.
• Illustration produced entirely from a text prompt: trainedAlgorithmicMedia.
• Collage of three photos plus one generated element: compositeSynthetic.
• Photo cropped and colour-corrected by hand, no generative tools: not an AI value; IPTC's humanEdits term covers it.

04How does Google use the digital source type?

Google's image metadata documentation lists Digital Source Type among the IPTC photo metadata properties it extracts, and names four values it supports: trainedAlgorithmicMedia, compositeSynthetic, algorithmicMedia and compositeWithTrainedAlgorithmicMedia. Its definitions match IPTC's in substance.

The same page explains that IPTC metadata is embedded in the image itself, so it moves with the file between pages, while structured data has to be added on every page where the image appears. It lists Digital Source Type only among the IPTC fields, not among the structured data properties for images, and it warns that Google does not guarantee image metadata will show in results. If you use both IPTC metadata and structured data for an image, the same page says Google uses the structured data when the two conflict, so keep any values identical. Signed provenance is a separate track: Google can also read C2PA metadata for its About this image feature, covered in C2PA Content Credentials on WordPress.

05Does WordPress keep the digital source type?

Not reliably in the files it serves. WordPress creates resized copies of every upload, and the image_strip_meta filter reference says metadata is stripped on resize by default, adding that the GD editor always strips profiles. The field you embedded may survive in the original upload while the medium and large copies your theme displays arrive without it.

Google's image metadata page also recommends keeping at least the creator, credit line and copyright notice fields, and warns that removing image metadata may be illegal in some jurisdictions. Image optimization services can strip metadata too, so check their settings for an option to keep IPTC or XMP data.

Practical options, roughly from least to most effort: link the original file from the caption; keep the label visible in the caption so readers get it even without metadata; configure your optimization layer to keep XMP; or, on servers using Imagick, return false from image_strip_meta so resized copies keep their metadata. That last change applies only to Imagick, and it keeps all embedded metadata on every resized copy, which can include camera details you may prefer not to publish.

06Checking a file

Check the image a visitor actually downloads, not the one on your computer. The free command-line tool ExifTool reads the field as XMP-iptcExt:DigitalSourceType; running exiftool -G1 -json -XMP-iptcExt:DigitalSourceType on a downloaded file prints the value if present.

1. Embed the field before upload in your image editor or metadata tool, using the full IPTC URI.
2. Check the local file to confirm the value is there.
3. Upload to WordPress, publish, and download the image from the live page.
4. Check the downloaded copy; if the field is gone, the resize or an optimizer removed it.
5. Check the full-size original's URL as well, and repeat after any change to image settings.

1[2  {3    "SourceFile": "example-image.jpg",4    "XMP-iptcExt:DigitalSourceType": "http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia"15  }6]

If you also describe the image in JSON-LD, use the matching schema.org member and keep both in step. How to disclose AI help in markup shows that pattern, all authenticity guides collects the rest of this section, and for how image labels fit into structured data as a whole, see the handbook home.