Metadata Gaps Limit Scholarly AI Attribution in Nexus-Score Study

A recent study published on arXiv (2607.22684) indicates that the absence of metadata hampers AI systems in accurately attributing scientific contributions. Researchers evaluated an AI system that cited works without access to pertinent paper lists, resulting in the generation of identifiers that were sometimes non-existent. By utilizing OpenAlex records, they either concealed or reinstated author, institution, funder, reference, and text-access links while keeping the works and tasks constant. When the appropriate link was restored, attribution was possible; however, incorrect link restoration yielded zero correct responses out of 469 mismatched tests. Missing links resulted in fabricated responses, refusals, or depletion of tool budgets, and web searches did not recover concealed author links. The study concludes that AI systems can only credit works when the relevant record connections are visible.