Artificial intelligence (AI) has been increasingly used in health care research. One primary area of application is literature review. Literature reviews are a key element of evidence-based healthcare and policy but they are often time consuming. Traditional guidelines on literature reviews are based on human searches and screening, which may be subject to biases, a lack of transparency and reproducibility, and human errors. With the development of AI, Machine Learning (ML), combined with Natural Language Processing (NLP) techniques, can be applied in literature reviews, enabling more transparent and efficient approaches, and more reproducible, reliable, and accurate results. They may potentially improve the way literature reviews are conducted in the future. However, it is critical to ensure that the new methods produce a high quality literature review that meets or exceeds the standards established by the research community and for health technology assessments. Therefore, it is important to establish standards on how to apply the NLP methods in literature reviews and have appropriate criteria to evaluate the quality of the reviews. In this presentation, we will discuss the key considerations in establishing the criteria, including the number of abstracts used in the training phase of the algorithm, the goodness of fit measures (or stopping rules) and the cross validation methodology, and the sensitivity and specificity thresholds for the classification of abstracts. In addition, we will also discuss how these criteria can be adjusted based on the type of literature review (i.e., systematic or targeted review), and whether the classification is performed for a single criterion (e.g. RCTs) or global criteria for relevance. Through these discussions, we hope to identify key criteria for implementing AI-assisted literature reviews and facilitate broader application of the methods through standardized approaches.