New Framework Defines How AI Should Assist Research Reviews While Keeping Humans in Control

Systematic reviews and evidence syntheses are growing harder to manage through purely manual processes as research volumes surge and update demands intensify. Machine learning and large language models now permeate these workflows—from query design through screening, data extraction, and synthesis—yet evidence for their reliability remains fragmented and task-specific. Current guidelines like PRISMA and related standards don't adequately address when AI use is methodologically sound, how to validate it, or how to document it for readers.

The Problem with Unguided AI Adoption#

Existing review standards, while foundational, lack concrete operational guidance on AI integration. They don't specify which decisions should always remain under human judgment, how AI tools should be benchmarked, or what transparency obligations exist around algorithmic involvement. This gap has allowed AI adoption to outpace our ability to evaluate it safely.

Introducing ARISMA#

Researchers have now proposed ARISMA—an AI Reporting and Integration standard for Systematic Methods and Analysis—to fill this void. Rather than treating AI as an autonomous reviewer, ARISMA frames it as an inspected, tracked, logged, and reversible assistant. The framework rests on a single principle: every consequential decision in the review must remain human-interpretable, auditable, and accountable.

What ARISMA Delivers#

The standard provides a lifecycle taxonomy, actionable process guidance, and stepwise recommendations spanning the entire review pipeline. It includes a governance and provenance model for tracking AI involvement, a tool-support framework, an AI-informed reporting checklist, and a validation matrix for assessing AI performance. The framework also considers legal, privacy, infrastructure, and long-term sustainability questions.

Path to Adoption#

ARISMA was developed through iterative expert consultation and offers a practical, auditable approach to integrating AI into evidence synthesis without surrendering scientific rigor or human accountability.