Sir, in evidence-based dentistry, systematic reviews play a crucial role by comprehensively analysing available data. However, searching numerous articles for relevance can be a time-consuming process. While Mahuli et al. acknowledged the utility of large language models (LLMs) in risk of bias analysis and data extraction for systematic reviews and meta-analyses, they overlooked the laborious process of article screening.1

In attempting to address this, I used two LLMs: ChatGPT 3.52 and Google Bard3 in the article screening process for a systematic review study. To conduct this systematic review, a dataset containing titles and abstracts of 1,111 articles underwent screening by two independent human reviewers. Concurrently, inclusion and exclusion criteria were defined for ChatGPT and Google Bard. Both AI models were prompted to evaluate articles, categorising them as ‘Yes' (relevant), ‘No' (irrelevant), or ‘Maybe' (uncertain), accompanied by brief reasonings for their decisions. Following this, the models underwent training using ten samples from the dataset, with a human operator correcting their responses. Subsequently, 100 randomly chosen article titles and abstracts were manually given to the AI models for screening.

ChatGPT aligned with the human reviewers' conclusions in 76% of cases, demonstrating a notably higher agreement compared to Google Bard, which aligned in only 47% of cases. This comparative analysis underscores ChatGPT's efficiency in determining article relevance during the screening process, suggesting its potential as a valuable tool for systematic review screening in evidence-based dentistry. In contrast, Google Bard exhibited a comparatively lower degree of concordance with the human reviewers and less favourable performance, indicating limitations in its accuracy for this specific task. This suggests a necessity for further refinement or cautious consideration of its applicability in similar contexts.

In conclusion, the application of LLMs, particularly ChatGPT 3.5, shows promise in enhancing evidence-based dentistry by optimising the screening process for systematic review studies, ensuring a more comprehensive scope, minimising the chances of critical articles being overlooked and thereby enhancing the robustness and reliability of the final review. However, it is crucial to acknowledge that human intervention and oversight are imperative to prevent errors.