A library could soon reach renewal season with a puzzle. Its researchers have relied heavily on a subscribed collection, yet the usage reports show fewer article views and downloads. Part of the activity may not appear because an AI research service, operating under institutional access rights, consulted the articles on the researchers’ behalf.
That discrepancy would have consequences. In a budget review, a collection that contributed frequently to research could still appear underused. The problem would not lie with the scholarship or the AI service. It would lie with the measures connecting use to value.
AI already has a substantial role in scholarly discovery, especially at the beginning of an inquiry. Libraries are also starting to provide tools that answer questions using licensed, peer-reviewed literature. Researchers can compare findings, assemble citations, and examine a field of evidence without taking the familiar route through a journal platform.
This gives subscribed scholarship more ways to be useful. It also makes some of that usefulness harder to see.
Subscription assessment has long relied on observable actions. A researcher authenticated, searched a database, opened an article, and perhaps downloaded the full text. Libraries used the resulting reports to calculate cost per use, understand collection performance, and prepare for renewal negotiations.
An AI-assisted inquiry produces a different trail. A response may examine a dozen articles, rely substantially on four, and cite two. Counting every retrieval as meaningful use would overstate the collection’s contribution. Counting only the citations or subsequent downloads would miss evidence that helped shape the answer. The valuable event lies somewhere between retrieval and citation: the point at which an article materially informs the research.
Emerging reporting practices are beginning to recognize this distinction. They separate an AI-generated response from the investigation of an article and from access to its full text. They also seek to count material selected for synthesis rather than everything inspected during retrieval. Publisher-hosted tools can make this activity comparatively easy to record. Approved services working across several licensed collections will require reporting to move reliably between systems.
Libraries need enough detail to understand how their collections are contributing. Publishers need visibility when licensed articles support research beyond their platforms. Neither requires access to the substance of a private inquiry. A useful record should identify the sources that materially informed an answer while protecting the prompt and the researcher’s developing line of thought.
Licenses will need similar precision. Allowing subscribed content to answer an authorized researcher’s question is not the same as allowing that content to train or fine-tune a model. The first may become part of ordinary institutional access. The second requires separate permission and a different commercial discussion.
A subscription can consequently offer more than reading access. It can provide governed computational access to full text, corrections, supplementary files, tables, and data, with dependable links back to the published record.
AI can bring premium scholarship into more questions, comparisons, and research decisions. The opportunity is to make that added usefulness visible when libraries and publishers determine what a subscription is worth. Know more
Knowledgespeak Editorial Team
