Introduction Identifying difficult airways and avoiding unanticipated difficult airways through difficult airway assessment are crucial for patient safety prior to airway management. Therefore, accurately predicting difficult airways through airway assessment is a fundamental and significant technique in airway management by clinicians. Artificial intelligence (AI) is a rapidly evolving science with greater data processing ability than humans. AI, given its ever-expanding applications in medical diagnosis and disease prediction, has been employed to predict cases with difficult airways. Nevertheless, the diagnostic performance of AI algorithms for difficult airway assessment remains unclear due to the small sample sizes, insufficient image acquisition standards and poor predictive accuracies. Consequently, this study aims to formulate a protocol for a systematic review and meta-analysis to ascertain the diagnostic value of AI in assessing difficult airways.Methods and analysis English-language databases (Cochrane Library, Web of Science, PubMed, Ovid Medline and Embase), Chinese electronic databases (China National Knowledge Infrastructure, VIP and Wanfang ] and clinical trial registry databases will be searched from their inception until January 2025 to identify clinical trials of AI for difficult airway assessment. Sensitivities, specificities, areas under the receiver operating characteristic curve, diagnostic likelihood ratios and diagnostic ORs with 95% CIs will be presented as indicators of AI's diagnostic accuracy in assessing difficult airways. Depending on the level of statistical heterogeneity evaluated by the I-square test, the fixed-effects or random-effects model will be employed. The risk of bias will be evaluated using the Quality Assessment of Diagnostic Accuracy Studies 2 tool. Furthermore, the quality of evidence concerning the outcomes will be assessed based on the Grading of Recommendations Assessment, Development and Evaluation criteria for diagnostic tests. Heterogeneity will be investigated through sensitivity, meta-regression and subgroup analyses. Additionally, Deeks' funnel plot asymmetry test will be used to detect publication bias.Ethics and dissemination Ethical approval is not required for this systematic review protocol. The results will be disseminated through peer-reviewed publications.PROSPERO registration number CRD42023462926.