How Generative AI Can Make Literature Reviews Faster and More Reliable

“

The Authors’ Perspective: A Practical, Human-Centered Method The authors position their work as a direct response to a methodological gap: researchers lacked a structured way to integrate generative AI into reviews without compromising quality. Their perspective is decidedly pragmatic and human-centered. AI as a Controlled Assistant: They argue that AI does not replace the researcher’s critical judgment. Instead, when guided by experienced researchers, AI increases efficiency and helps make reviews more robust and methodologically rigorous. A Process, Not a Push-Button: The updated SSF method provides a step-by-step guide for using AI in search strategy, article selection, and scientific writing. The value lies in the structured workflow, not in automating the entire review. The Critical Perspective: The Risk of "Rigor for Volume" A significant counterpoint comes from researchers questioning whether AI-assisted reviews sacrifice methodological rigor for sheer output. The Transparency Gap: A 2026 viewpoint article argues that current reporting standards (like PRISMA) remain anchored in pre-AI workflows. They do not adequately address AI-specific challenges such as algorithmic bias, auditability, and the need to document human oversight, creating a risk of superficial or duplicated reviews. The Call for New Standards: This perspective advocates for urgent adaptations, including a PRISMA-AI extension and updated registry requirements (e.g., PROSPERO), to safeguard credibility. The Empirical Perspective: Promise, but with Significant Caveats Systematic reviews paint a mixed picture of AI's actual performance. Efficiency Gains, but Uneven Reliability: GenAI is most often used for screening, search, and writing, and many studies report efficiency gains. However, concerns remain regarding validity, transparency, and methodological rigor. Performance is stronger on easy, rule-governed tasks (e.g., extracting publication years) than on tasks requiring contextual interpretation or synthesis. Concrete Error Rates: One systematic review found that generative AI missed 68.0%–96.0% of relevant studies during searching and made incorrect exclusions in 1.0%–83.0% of cases. The Verification Burden: Researchers find LLMs useful for organizing ideas and exploring literature, but they do not trust them for accurate, verifiable content. A persistent verification burden and a lack of trust in outputs remain major pain points. The Researcher’s Perspective: A Tool, Not an Oracle From the vantage point of researchers actually using these tools, the experience is characterized by a tension between utility and distrust. Useful but Unreliable: LLMs help with organizing ideas, improving writing, and exploring literature, but researchers are clear that they cannot be trusted to provide accurate or verifiable scientific content. The "Press-Button" Warning: One analysis cautions that while AI can improve the span and quality of a review, the time savings are not as massive as expected, and a strategy of leaving AI to do the work is "a recipe for disaster". The Ethical and Integrity Perspective: New Risks for Academic Governance Finally, there is a strong perspective focused on the ethical and governance challenges that AI introduces into the review process. Core Ethical Concerns: The use of AI in evidence synthesis raises challenges related to authorship, research integrity, copyright, and plagiarism. Hallucinations and Bias: The most frequently cited risks are hallucinated references, unreliable or fabricated citations, and the amplification of bias. The Governance Deficit: Institutions and publishers are struggling to keep pace. There is a recognized need for transparent disclosure of AI involvement and clear guidelines to protect the legitimacy of scholarly work. Summary The perspectives on this article reveal a field in tension. The authors offer a practical, human-in-the-loop framework to harness AI’s efficiency. However, the broader research community cautions that this efficiency comes with significant empirical reliability problems (missed studies, hallucinations), a critical transparency gap in reporting standards, and unresolved ethical risks. The consensus perspective is that AI is a powerful but flawed assistant—one that demands structured oversight, verification, and new governance to prevent the pursuit of speed from eroding the rigor that gives literature reviews their value.

”