ARISMA: Guidelines for AI- and LLM-Assisted Systematic Reviews, Scoping Reviews, and Mapping Studies

AI should augment, not replace, human judgment in research synthesis: ARISMA establishes that "every consequential scientific decision must remain human-interpretable, human-auditable, and human-accountable," meaning AI tools like large language models should function as inspected assistants rather than autonomous decision-makers in systematic reviews.

Current AI evidence is insufficient to justify widespread automation: The paper notes that "empirical evidence remains uneven, task-dependent, and insufficient to justify unconstrained automation," indicating that organizations should be cautious about deploying AI across all stages of evidence synthesis without proper validation for each specific task.

AI use in reviews requires transparent reporting and auditability: ARISMA proposes new reporting standards so that "readers can audit" AI involvement in research synthesis, addressing a gap where existing standards like PRISMA 2020 don't specify how AI use should be documented and made transparent to the scientific community.

Human oversight must be maintained at critical decision points: The framework emphasizes that certain review decisions "must remain human-led," establishing guardrails for where AI assistance is appropriate versus where human expertise and judgment are irreplaceable in evaluating research quality and relevance.

AI tools need benchmarking and validation before deployment: ARISMA treats AI as "inspected, benchmarked, logged, and reversible," meaning organizations should test AI systems against known standards and maintain the ability to reverse AI-assisted decisions before using them in actual research synthesis workflows.