Implementing a delayed retirement policy is a crucial measure to advance China’s national strategy of actively addressing population aging in the new era. This study examines public attitudes, opinions, and perceptions of the delayed retirement policy by analyzing comments from major media accounts on Weibo and Douyin. Using Python-based data mining and natural language processing (NLP) sentiment analysis tools, the research identifies the characteristics and evolutionary patterns of online public opinion surrounding this policy, providing insights to refine its implementation. The results reveal that public evaluations of the delayed retirement policy are predominantly positive or neutral. High-frequency terms such as “support”“voluntary”“flexible” and “elastic” reflect widespread public recognition of the policy, with expectations for adjustments tailored to individual needs. Additionally, the public opinion dynamics exhibit distinct evolutionary features, including diverse discussion participants, deepened discourse, rapid diffusion of secondary topics, and a trend toward rational social attitudes. This “unexpected” online opinion phenomenon is closely linked to factors such as government expectation management and sentiment modulation, varying public interests and responses, flexible policy wording and emotional reassurance, official narrative guidance and curated comments, as well as netizens’ moral constraints and self-censorship. To foster constructive interactions between the policy and the public, the study recommends optimizing strategies in several areas: addressing public concerns, strengthening authoritative information dissemination, responding to core public demands, and building a robust public opinion monitoring system. These measures aim to enhance the effective implementation and iterative improvement of the delayed retirement policy.