Useful for Exploration, Risky for Precision: Evaluating AI Tools in Academic Research

AI tools excel at exploration but fail at precision: Q&A tools provide valuable overviews and generally accurate summaries, making them useful for initial research exploration, but they are "not always reliable for precise information extraction" and therefore unsuitable for tasks requiring exact data.

The "explainability gap" puts burden back on researchers: Explainable AI (xAI) accuracy was "particularly low," with highlighted source passages frequently failing to match generated answers, meaning researchers still must manually verify outputs rather than trusting the tool's explanations.

Literature review tools lack reproducibility and transparency: These tools showed "low reproducibility, limited transparency regarding chosen sources and databases, and inconsistent source quality," making them inappropriate for systematic reviews where documentation and consistency are critical.

Current benchmarks miss human-centered needs: Existing evaluation methods focus on technical performance but fail to capture "usability, interpretability, and integration into research workflows"—factors essential for understanding how AI tools actually function in real research settings.

Verification remains a researcher responsibility: The paper's core finding suggests that despite AI's efficiency gains, researchers cannot reduce their validation workload; they must still critically assess AI outputs before relying on them for academic work.