AI evidence extraction

Utilizing Artificial Intelligence to create narrative literature reviews

Auro del Giglio, Mateus Uerlei Pereira da Costa · Einstein (São Paulo) · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.31744/einstein_journal/2026rw1165

Methodology & findings

Study design

Narrative literature review examining the use of AI-powered tools in the composition of narrative reviews.

Main result

The study demonstrates that AI-powered tools can substantially enhance the narrative literature review process across multiple steps. Specifically, "researchers can quickly identify relevant research gaps, generate outlines, conduct literature searches, and organize retrieved articles. Additionally, Artificial Intelligence can aid in the writing process." The paper identifies seven key steps where AI integration is beneficial: identifying research gaps, creating outlines, searching for relevant literature, organizing articles, writing reviews, and managing references. Tools such as ChatGPT, Research Rabbit, Semantic Scholar, and Zotero are described as enabling more efficient literature review workflows while maintaining quality.

Research paradigm

Pragmatist/Applied research - exploring practical applications of AI tools in academic writing

Author conclusions

"The integration of Artificial Intelligence into narrative literature reviews has immense potential for revolutionizing the efficiency and effectiveness of this critical research endeavor." The authors conclude that "Artificial Intelligence not only saves time but also enhances the quality of the review by suggesting appropriate corrections to the text. However, it is essential to note that Artificial Intelligence tools should not replace the vital process of critically reading and analyzing scientific literature." They further state: "Instead, they should be used as complementary aids to improve the overall writing of literature reviews. As Artificial Intelligence technology advances, researchers can look forward to streamlining their work and producing high-quality narrative literature reviews more efficiently."

Risk of bias

Selection bias: Non-systematic article selection based on author judgment rather than explicit inclusion/exclusion criteria; Lack of reproducibility: Authors acknowledge search strategy and reference selection are not reproducible; Subjective relevance assessment: Article inclusion determined by authors' subjective judgment of relevance; Limited search scope: Only first 200 references from Google Scholar were reviewed; no date limits but limited source diversity; No systematic quality assessment of included articles; Selection bias: Non-systematic article selection based on author judgment; Reproducibility risk: Authors explicitly state the search strategy 'may not be reproducible'; Information bias: Reliance on articles authors deemed 'most relevant' rather than systematic inclusion criteria; Temporal bias: No date limits on searches may skew findings toward older literature; Search bias: Subjective interpretation of relevance; search strategy not fully reproducible; Database limitations: Only PubMed and Google Scholar searched; limited to first 200 Google Scholar results; Confirmation bias: Authors selected articles deemed most relevant to their objectives

Limitations

The authors acknowledge that "AI-powered tools may be several times more efficient than humans for various tasks, such as researching literature, summarizing articles, and creating bibliographies, they lack creativity, which is an exclusively human characteristic." Additional limitations include: "Although one can use ChatGPT to identify some of the gaps, future directions, or research in a particular field, because this LLM is built and trained on existing information, the novelty of these outputs is questionable and potentially biased." The authors also note that "the automatic process of summarizing an article using an AI tool does not substitute for the learning process involved in reading the original paper, manually selecting critical parts of the text, and creating commentaries." Furthermore, "An additional limitation of AI is the occurrence of hallucinations that can produce inaccurate results." The authors explicitly state: "Because this review was non-systematic, the articles selected for inclusion and citation were those deemed the most relevant by the authors
Therefore, the search strategy and reference list described herein may not be reproducible."

Open questions raised

Optimal integration of AI in the narrative review process, including investigation of effective workflows and methodologies that optimize human-AI collaboration
Quality assessment frameworks for AI-assisted narrative reviews, including metrics and frameworks specifically tailored to assess reliability and accuracy
Development of better methods to evaluate quality of AI-assisted narrative reviews
Need for research on prompt engineering optimization for improved LLM outputs
Investigation of how to prevent AI-generated content from merely satisfying 'publish or perish' requirements without contributing novel information
The paper identifies five potential research gaps in AI-assisted narrative reviews: (1) Optimal Integration of AI in the Narrative Review Process, investigating most effective integration methods and workflows; (2) Quality Assessment of AI-Assisted Narrative Reviews, developing metrics and frameworks specifically tailored for AI-assisted reviews; and implicit gaps regarding the evaluation of AI-generated content reliability, accuracy, bias in AI outputs, and ethical implications of AI use in scientific publishing.

Data: Zotero database entries used to draft the article are available in Table 1S, Supplementary Material; Authors note: "In Table 1S, Supplementary Material, we present all the saved entries in the Zotero database used to draft this article." However, no explicit public dataset URL is provided. Data availability statement: "Data are available to reviewers upon request."; Zotero database entries saved and presented in Table 1S, Supplementary MaterialExtracted from: pdfAgreement 79%

Explore related topics

Systematic Review Automation Literature Discovery Conceptual