Citing Less Critically: LLMs Reshape the Rhetoric and Reach of Scientific Citation
Summary¶
Uses a masked-citation task over 1,746 NLP conference papers and 63,000+ citation contexts to compare how LLMs versus human authors cite. LLMs cite significantly less critically, over-reference popular and established work (especially in contrasting citations), and reach socially distant authors rather than close collaborators. The double-edged framing - wider citation reach but reduced critical engagement and amplified visibility bias - is a concrete mechanism for how AI-assisted writing reshapes the scholarly record.
Contribution¶
Queued — not yet read.
Method¶
Queued — not yet read.
Relevance to RISE¶
Surfaced by the 2026-09-07 literature sweep as a medium-priority candidate for the RISE knowledge base (angle: research productivity of AI). Verified against: Opened https://arxiv.org/abs/2609.01432 on 2026-09-07; confirmed title, five authors, submitted 1 Sep 2026, EMNLP 2026 main-conference acceptance comment, abstract findings. Not in dedupe_titles.txt.
Critique / open questions¶
Queued — not yet assessed.
Key quotes¶
Scientific citations carry rhetorical intent. (abstract)
Scholars may cite prior work positively (supporting), negatively (contrasting), or neutrally (mentioning). (abstract)