AI Photo Captions and People Tags: What Survives Export

A visible caption is not necessarily part of the photograph

An AI photo caption is a description generated from the image or its surrounding information. It might say that a photograph shows three people standing beside a red car, or that a scene appears to be a family gathering. The caption may appear directly beneath the image in an application, giving the impression that it travels with the photograph. In practice, it can be stored in a separate database controlled by the service.

This distinction matters when creating an archive. Downloading an image file does not guarantee that its caption accompanies it. A service might include the description in an export, write it into an image metadata field, place it in a separate JSON file, or omit it altogether. The export method determines which information leaves the application. Treat the original image and its associated descriptions as separate archival objects until their relationship is documented.

• Original image: the photograph as supplied or captured.
• Caption: a textual description associated with the photograph.
• People tag: an identification or label associated with a person.
• Export record: information describing how the application packaged the material.

Five places where descriptive information may be stored

Image metadata can be embedded within a photograph file. Common standards include EXIF, IPTC and XMP. EXIF commonly carries camera information, capture dates and technical settings. IPTC supports descriptive fields such as captions and creator information. XMP can carry structured descriptions and other metadata. These standards provide possible locations for information, but an application does not necessarily write every field or preserve every field during export.

Other applications maintain descriptions in a database, a cloud account or a companion sidecar file. A sidecar is a separate file associated with an image, often using a matching filename. A JSON export may contain captions, identifiers and people labels in a structured form. Such a file can be useful, but it must remain connected to the correct photograph. A folder containing images and an unrelated collection of descriptions is difficult to interpret later.

• Embedded EXIF, IPTC or XMP metadata.
• Application-specific cloud or local database.
• Sidecar files stored beside photographs.
• Structured JSON or CSV export.
• A manually maintained catalogue.

People tags need a more careful record

A people tag can mean several different things. It might be a name manually attached to a photograph, an automatically suggested identity, a label assigned to a face group, or an internal identifier connected to an account contact. These are not equivalent records. A face group called Person 12 is not a reliable identification, and an automatically suggested name is not proof that the person has been correctly identified.

For a family archive, preserve the name and the basis of the identification where known. For example, a catalogue entry might read: 'Margaret Ellis, identified by family album inscription, approximately 1984.' This provides more context than a bare label. Where an identity is uncertain, record that uncertainty explicitly. Avoid converting a machine-generated suggestion into a confirmed name simply because it appears in an application.

• Record the displayed name separately from an internal application identifier.
• Distinguish confirmed identities from suggestions.
• Preserve uncertainty when a face cannot be identified.
• Include the photograph's filename or stable catalogue reference.

Prepare an export that keeps the relationships intact

Start by copying the original photographs into a preservation folder without changing their contents. Keep a second working copy for any metadata editing or format conversion. Before exporting from a tagging service, look for separate options for original images, edited images, metadata, albums and account information. The wording varies between services, and an album download may not contain the same information as a full account export.

For example, an image named 19840615_001.jpg might have a separate description record containing its caption, people labels and original album name. The catalogue should refer to that exact filename and identify the source application. If the export uses numeric identifiers, preserve the identifier-to-filename mapping. Do not assume that photographs listed in the same order in two folders correspond to one another.

• Preserve an unchanged original image.
• Export the highest-quality available original.
• Save captions and people labels in a separate structured file.
• Retain original filenames and application identifiers.
• Record the export date and source account or service.

A field-by-field survival record

A useful handover catalogue records the information required to interpret each photograph independently of the application. The following table describes what should be retained and what to look for in an exported archive. These are preservation targets, not promises about a particular service. The actual outcome depends on the export format and the provider's current implementation.

Where a field is missing, retain that fact in the catalogue instead of silently leaving the reader to assume that the information never existed. This is particularly important when a description or identity was visible in the original account. A missing exported field may indicate an export limitation, not an absence of information in the source collection.

• Caption: kept if embedded or included in a linked metadata or JSON file; otherwise lost from the export.
• Capture date: kept when embedded date metadata is preserved; changed if a conversion or export substitutes a different date.
• People tag: kept if exported as a readable name or linked identifier; lost if it remains only in the service database.
• Place: kept if coordinates or a descriptive location are exported; changed if coordinates are replaced by a less precise place label.
• Album membership: kept if the export includes album names and image references; lost if only individual image files are downloaded.

Make the descriptions usable outside the account

An exported archive should make sense to someone who has never used the original application. Keep a short README explaining the folder structure, filename conventions, metadata files and any identifiers that need interpretation. Include the name of the service from which the material originated, the approximate export date and the relationship between the original photographs and their descriptive records.

For a digital estate inventory template, add the photograph collection as an entry with its storage location, account dependency, export location and known access arrangements. A digital legacy checklist should also identify which descriptions exist only online. The aim is to distinguish the photographs themselves from the additional information that makes them searchable and meaningful. Begin by choosing one representative photograph and creating its catalogue entry before applying the same structure to the rest of the collection.

Questions readers ask