Higher reasoning effort improves accuracy and detail of GPT 5.5 references
Higher Reasoning Effort Improves Reference Verifiability and Changes Reasoning Summaries in GPT-5.5
Computation and Language
Summary
Large language models like GPT-5.5 can suggest scientific references when writing papers. This study looked at whether making the model think harder would produce more accurate and verifiable references. The authors found that when GPT-5.5 used more effort, it provided references that matched real records more often and gave longer, more detailed explanations of its reasoning. However, having more detailed reasoning summaries didn’t always guarantee better reference accuracy.
What this means in practice
- •For scientific publishing teams: Improve automatic generation of reference lists by making GPT models use higher reasoning effort to increase accuracy and transparency.
- •For academic software developers: Integrate higher-effort GPT reasoning to provide users with more verifiable references and transparent summaries in academic writing tools.
Authors
Melika Mobini, Andres Algaba, Vincent Ginis
Abstract
Large language models are increasingly being used in different stages of scientific knowledge production, including the suggestion of references. These models can operate at different levels of reasoning effort and provide summaries of their reasoning processes. This raises two questions: Does increasing reasoning effort improve the verifiability of generated references, and do the resulting reasoning summaries reveal how the model's behavior changes with greater effort? We examined summaries from GPT-5.5 as it generated scholarly references for 1,000 papers under paired low- and high-effort settings. Generated references were matched to external bibliographic records, and reasoning summaries were scored for four predefined features and explored for additional features discovered by GPT-5.5. High effort increased the mean paper-level resolver-verified reference rate from 78.3% to 84.8%, and produced longer reasoning summaries. All four predefined features received higher scores under high effort. The discovered features changed less uniformly: some appeared equally often in both conditions, whereas others were more common under high effort. Within each effort condition, neither feature set showed a consistent increase in verification rates with higher scores.