NISO to Develop Metadata Model for Provenance and Attribution in AI Systems
NISO has announced a 12-month pilot for a minimum viable metadata model to facilitate the tracking of provenance and attribution of content in AI systems, addressing a critical challenge AI presents for scholarly publishing.
With the development of AI-assisted search, machines have effectively become readers, with AI agents analyzing and retrieving vast amounts of content and data to answer queries. This transformation has led to a “zero-click” environment in which users can access information drawn from published content without ever visiting a publisher site. This scenario poses a problem for scholarly communications, as both publishers and librarians have traditionally relied on usage statistics to assess the value of content. At the same time, large language models are notoriously unreliable in disclosing where content came from as well as in citing sources, which can weaken credibility and trust in scholarly research. To address this challenge, the NISO pilot will develop and test a minimum metadata model that accurately tracks provenance and attribution when AI systems retrieve information or data from scholarly content.
The idea for a minimum metadata model emerged from stakeholder conversations about provenance, usage, and attribution in AI systems that began in 2025. In February 2026, the NISO Plus Baltimore meeting hosted two pre-conference sessions addressing these topics. The following spring, NISO, Cambridge, and COUNTER co-hosted workshops in which stakeholders explored potential solutions, and the resulting report was released in June. Since then, NISO has recruited a group of content creators, publishers, repositories, AI tool developers, and end-users to develop the minimum metadata payload for delivering provenance and attribution information and setting the stage for future work in this area.
The NISO project has received funding from participants in the TRACE (Trusted Retrieval and Attribution for Content Ecosystems) initiative, a cross-industry collaboration to develop the infrastructure needed to ensure AI systems work optimally in the research publishing environment. The metadata project is one of several workstreams tied to TRACE, with others addressing tracking AI usage and studying how AI systems interface with scholarly content.. Todd Carpenter, Todd Toler (Ithaka S+R), Tasha Mellins-Cohen (COUNTER), and Priya Madina (Springer Nature) introduced TRACE at the October 5 STM meeting in Frankfurt, just prior to the start of the Frankfurt Book Fair.
For additional details on TRACE and its associated projects as well as NISO’s pilot, check out Todd Carpenter's October 6 post for The Scholarly Kitchen..