How can embedded image metadata identify an AI-created image?

Definition: The IPTC Digital Source Type field is an embedded controlled-vocabulary value for describing the source used to create a digital image. Google Search Central lists four IPTC NewsCodes for algorithmic or synthetic source descriptions, so the file can carry a documented source category.

Key facts

• Google's image metadata documentation, last updated 10 December 2025, says a site can add structured data or IPTC photo metadata to each image.
• The IPTC Photo Metadata Standard defines Digital Source Type as the type of the source of a digital image. It is a CV-code with cardinality 0..1 and uses the Digital Source Type NewsCodes vocabulary.
• Google supports four values: trainedAlgorithmicMedia, compositeSynthetic, algorithmicMedia, and compositeWithTrainedAlgorithmicMedia.
• IPTC photo metadata is embedded in the image file. Google says it can move with the image while staying intact, and it only needs to be embedded once per image.
• When IPTC metadata and structured data conflict, Google uses the structured data information.

What does the Digital Source Type field record?

The IPTC Photo Metadata Standard defines Digital Source Type as “the type of the source of this digital image.” The field asks an implementer to select a value from a controlled vocabulary. Its cardinality is 0..1, so one image metadata record can contain zero or one value for this property. The standard identifies the XMP property as Iptc4xmpExt:DigitalSourceType with a URI value and the JSON property as digitalSourceType with a string or URI value.

Iptc4xmpExt:DigitalSourceType [URI <External>]
digitalSourceType [string/uri/]

The field is therefore a source classification carried by the image. It can distinguish algorithmic creation from a composite process when the metadata uses one of the controlled values that Google documents.

Which source codes does Google support?

Google's IPTC table lists these four NewsCodes:

• trainedAlgorithmicMedia: the image was created algorithmically using a model derived from sampled content.
• compositeSynthetic: the image is a mix or composite that includes at least one synthetic element.
• algorithmicMedia: the image was created purely by an algorithm without sampled training data, such as software using a mathematical formula.
• compositeWithTrainedAlgorithmicMedia: the image combines trained algorithmic media with other media, including through inpainting or outpainting operations.

The values describe different source paths. A fully algorithmic image made from sampled content maps to the trained-model value in Google's list, while an image that combines trained algorithmic media with another medium maps to the composite value that names that combination.

How does a website carry the value?

Google documents IPTC photo metadata as information embedded directly inside the image. The image and its metadata can move from page to page while staying intact, so the site owner embeds the IPTC value once in the image file. Google also says to add image metadata to each image on a site, including each page instance when the same image appears on multiple pages.

Structured data is a second route for image metadata. If a page supplies both forms and they contain conflicting information, Google's documented precedence is structured data.

What can a reader conclude from the value?

The value tells a reader which source category the image metadata assigns to the file. It gives a machine-readable distinction between a model-derived image, a composite containing a synthetic element, a purely algorithmic image, and a composite involving trained algorithmic media. Google's documentation says it extracts the Digital Source Type property, while also warning that IPTC photo metadata may not show up in search results.

The opened documentation describes extraction and supported meanings. It does not describe a separate verification process for checking whether the declared value is accurate.

Related terms

• Does Google need image metadata for every occurrence of an image?
• Which image metadata wins when IPTC and structured data conflict?
• Image alternative text

FAQ

Which value describes an image made with a trained model?

Use trainedAlgorithmicMedia when the image was created algorithmically using a model derived from sampled content, as described in Google's supported NewsCodes.

Which value describes an image edited with inpainting or outpainting?

Use compositeWithTrainedAlgorithmicMedia when trained algorithmic media is combined with other media through an operation such as inpainting or outpainting.

Will the metadata always appear in Google Search?

No. Google says that structured data or IPTC photo metadata may not show up in search results, even when the property is supported.

Sources

1. Google Search Central, "Image metadata in Google Images", 10 December 2025
2. IPTC, "IPTC Photo Metadata Standard 2025.1", 26 November 2025

---

Check how a page reads to answer engines. Open the GEO score analyzer