Skip to Main Content

JSTOR Seeklight: Overview [Tier 3]

  • Updated

On this page:

Developed in collaboration with our community, JSTOR Seeklight enables you to quickly generate descriptive metadata for your collections at scale. Once generated, your metadata can be reviewed and edited, so that you retain expert oversight of your content.

JSTOR Seeklight is in active development with new features and improvements being frequently added. For more information on how the tool is powered and our approach to AI-assisted digital stewardship, see our JSTOR Digital Stewardship Services FAQ.

Note: The ability to create projects with JSTOR Seeklight is limited to JSTOR Digital Stewardship Services Institutional Admin(s) only. See Managing user access for more information.

This article provides an overview of JSTOR Seeklight's functionality. For detailed instructions on generating metadata with the tool see JSTOR Seeklight: Generating, reviewing, and editing metadata.

Generating, reviewing, and editing metadata

JSTOR Seeklight generates descriptive metadata using an AI-compatible Dublin Core-based metadata schema. Generated metadata is stored within a project and can be sorted, filtered, reviewed, and edited using the JSTOR Stewardship interface.

Projects containing AI-generated metadata have a Tag field. There are three tags, designed to help you track and maintain your metadata quality:

  • AI-Generated - By default, all items with AI-generated metadata have this tag. It cannot be removed.
  • Reviewed - You can mark and unmark metadata as being "Reviewed" from the project page and on the item page. This signals that a person has read and assessed the generated metadata.
  • Edited - When you edit an AI-generated metadata field, the "Edited" tag is automatically applied to the item, and the confidence score from the edited field is removed.
Note: Tags are applied at the item level. If you edit one or more AI-generated metadata fields on an item, the item will display the "Edited" tag even if you did not edit any other fields.

Tags don't display on JSTOR and are only visible in the JSTOR Stewardship interface. They are designed to help track if your AI-generated metadata has been evaluated and/or edited by a person. However, tagged items do display a disclaimer when published to JSTOR. See Publishing to JSTOR.

For more information on reviewing and editing your metadata, see JSTOR Seeklight: Generating, reviewing, and editing metadata.

Exporting project metadata

You can export a CSV file of all of your project's records, including system data and tags, from the Overview tab of your Project Settings.

You can also download an Excel spreadsheet including one or more of your project metadata records by selecting the item(s) you'd like to download on the project page, clicking the three-dot overflow menu, and selecting Download Metadata. Note that this spreadsheet doesn't include system data or tag information.

Confidence scores

Each JSTOR Seeklight generated field has a confidence score. Confidence scores indicate the system's assessment of its generated metadata output. As AI-generated metadata can never be guaranteed to be fully error-free, these scores are designed to help guide you in reviewing and editing your metadata.

There are three confidence score levels:

  • Low - The metadata may be incorrect or incomplete. A thorough human review is suggested before acceptance or publication.
  • Medium - The metadata may be reasonably accurate but may need review to catch missing context, minor errors, or subtle inaccuracies.
  • High - The generated metadata may be highly reliable, but not infallible. A review is still recommended.

In some cases, JSTOR Seeklight may be unable to generate a confidence score due to unique considerations for a field. In that case, you will see "Not Available" for the score. Fields that are not generated by JSTOR Seeklight will not display a confidence score.

Once you edit an AI-generated field, the confidence score for that field will display as "Edited."

You can also filter fields by confidence score in your project.

Note: While the confidence scores provide guidance on when to review metadata, no AI-generated content is guaranteed to be 100% accurate. Even high-confidence metadata can contain errors, omissions, or subtle inaccuracies. We recommend always reviewing your metadata, especially for critical metadata fields.

Project summaries

The project summary is an AI-generated summary presented in the framework of a finding aid, based on the current metadata in your project. To generate a project summary, open your AI-compatible project and select the Project Summary tab.

Project summaries cannot be edited within the project interface, and at this time cannot be published to JSTOR. You can download a TXT (.txt) file of the summary to edit and use in your own system. Project summaries can be re-generated when you edit your item metadata or add additional items to your project.

Note: If you have edited your item metadata and do not see the option to regenerate your project summary, try refreshing the page.

Project summaries contain the following categories:

  • Title - A high-level title for the collection
  • Date(s) - Dates of the items in the collection, based on the Date field
  • Extent of the Collection - A brief description of the number and content type of items
  • Creator(s) - The creator(s) of the items, based on the Creator field
  • Language(s) - The identified language(s) of the items, based on the Language field
  • Abstract - A brief, high-level description of the collection
  • Biographical & Historical Note - A brief description of the biographical and/or historical information available in the items' metadata
  • Scope & Content Note - A description of the scope of the collection including content types, themes, notable items, and any identified limitations in the collections' scope

Publishing to JSTOR

Before publishing, you'll be prompted to confirm you have reviewed your metadata. We recommend all metadata be reviewed prior to publishing to JSTOR, as high quality metadata enables better discoverability and usability.

Items with AI-generated metadata will display the following text at the end of the Item Details panel on JSTOR: The collection contributor used AI to facilitate the creation of some metadata for this item.

See Publishing items and collections for more information.

Articles in this section