American University Library Policy on the Use of AI-Generated Descriptive Metadata
Effective August 2025
American University Library is actively exploring the use of artificial intelligence (AI) to enhance the discoverability and accessibility of our digital collections. As of August 2025, we are piloting the use of AI-generated descriptive metadata for text and image-based resources, in addition to our existing use of AI-generated audiovisual transcripts. These efforts are guided by the principles outlined in American University’s Guidance on the Responsible Use of Artificial Intelligence (AI), Responsible Use of Artificial Intelligence Principles, and the AU Library Artificial Intelligence Exploratory Working Group: Report and Recommendations.
Our approach is grounded in a human-centered AI philosophy, which prioritizes:
- Human oversight and expertise in all AI-assisted processes
- Transparency in the use and origin of AI-generated content
- Ethical standards that align with the university’s data policies and values
As part of our strategic goals to scale mass-digitization and streamline metadata workflows, we are participating in the revamped JSTOR Digital Stewardship Services charter membership, which includes the AI-tool JSTOR Seeklight, to generate descriptive metadata for selected digital assets. For more information about this tool, please consult the JSTOR Seeklight FAQ.
To ensure transparency and accountability, all records containing AI-generated metadata are clearly labeled. These labels appear in the AI-Generated Metadata Control field and the AI-Generated Metadata Statement field within the metadata record.
American University Library remains committed to evaluating the effectiveness, accuracy, and ethical implications of AI tools in library practice.
For any metadata corrections, please contact archives@american.edu indicating the url, local identifier, title, and include information you wish to correct.
How are collections selected for AI-generated description?
Not all our collection material is appropriate for AI-generated metadata description. We select projects for AI-generated metadata based on several factors including:
- University guidance on the use of AI
- University Data Classification Policies
- Privacy concerns and federal privacy laws
- Donor intent and deed of gift restrictions
- Ownership and copyright
- Type of material needing description
- Amount of existing description and contextual information
- Impact of incorrect information vs access
- Volume of material
- Resources available
What Quality Control measures are in place for AI-generated metadata?
The quality of metadata is closely tied to the accuracy and relevance of its descriptions in relation to what is known about the item. In many cases, especially when limited contextual information is available, metadata creators must rely solely on visual observation. In such instances, AI tools can offer valuable efficiency. However, the more contextual information Library staff have, the less value the AI-generated metadata will bring. For collections utilizing AI-generated descriptive metadata, human review (including editing and correcting) is just as critical to quality control as it is in fully manual workflows. Key quality control measures include:
- 100% human review of core fields such as title, description, and subject
- Targeted review of other fields based on AI confidence scores and representative sampling
- Enhancement and refinement of AI-generated content where necessary
- Manual creation of many additional metadata fields by human catalogers
This hybrid approach ensures that metadata remains accurate, meaningful, and aligned with professional standards. We will adjust quality control measures as necessary.
Which metadata fields may include AI-generated information?
- Title
- Creator
- Date
- Description
- Genre
- Geographic place
- Volume
- Issue
- Language
- Name subject
- Publisher
- Topical subject
- Audiovisual transcript
American University Library Policy on the Use of AI-Generated Descriptive Metadata
Effective August 2025
American University Library is actively exploring the use of artificial intelligence (AI) to enhance the discoverability and accessibility of our digital collections. As of August 2025, we are piloting the use of AI-generated descriptive metadata for text and image-based resources, in addition to our existing use of AI-generated audiovisual transcripts. These efforts are guided by the principles outlined in American University’s Guidance on the Responsible Use of Artificial Intelligence (AI), Responsible Use of Artificial Intelligence Principles, and the AU Library Artificial Intelligence Exploratory Working Group: Report and Recommendations.
Our approach is grounded in a human-centered AI philosophy, which prioritizes:
- Human oversight and expertise in all AI-assisted processes
- Transparency in the use and origin of AI-generated content
- Ethical standards that align with the university’s data policies and values
As part of our strategic goals to scale mass-digitization and streamline metadata workflows, we are participating in the revamped JSTOR Digital Stewardship Services charter membership, which includes the AI-tool JSTOR Seeklight, to generate descriptive metadata for selected digital assets. For more information about this tool, please consult the JSTOR Seeklight FAQ.
To ensure transparency and accountability, all records containing AI-generated metadata are clearly labeled. These labels appear in the AI-Generated Metadata Control field and the AI-Generated Metadata Statement field within the metadata record.
American University Library remains committed to evaluating the effectiveness, accuracy, and ethical implications of AI tools in library practice.
For any metadata corrections, please contact archives@american.edu indicating the url, local identifier, title, and include information you wish to correct.
How are collections selected for AI-generated description?
Not all our collection material is appropriate for AI-generated metadata description. We select projects for AI-generated metadata based on several factors including:
- University guidance on the use of AI
- University Data Classification Policies
- Privacy concerns and federal privacy laws
- Donor intent and deed of gift restrictions
- Ownership and copyright
- Type of material needing description
- Amount of existing description and contextual information
- Impact of incorrect information vs access
- Volume of material
- Resources available
What Quality Control measures are in place for AI-generated metadata?
The quality of metadata is closely tied to the accuracy and relevance of its descriptions in relation to what is known about the item. In many cases, especially when limited contextual information is available, metadata creators must rely solely on visual observation. In such instances, AI tools can offer valuable efficiency. However, the more contextual information Library staff have, the less value the AI-generated metadata will bring. For collections utilizing AI-generated descriptive metadata, human review (including editing and correcting) is just as critical to quality control as it is in fully manual workflows. Key quality control measures include:
- 100% human review of core fields such as title, description, and subject
- Targeted review of other fields based on AI confidence scores and representative sampling
- Enhancement and refinement of AI-generated content where necessary
- Manual creation of many additional metadata fields by human catalogers
This hybrid approach ensures that metadata remains accurate, meaningful, and aligned with professional standards. We will adjust quality control measures as necessary.
Which metadata fields may include AI-generated information?
- Title
- Creator
- Date
- Description
- Genre
- Geographic place
- Volume
- Issue
- Language
- Name subject
- Publisher
- Topical subject
- Audiovisual transcript