Artificial intelligence (AI) has substantial potential to improve service quality and health equity in primary healthcare institutions. China has introduced a series of policies to promote the development and application of AI in primary care. However, health data poverty, defined as the inability of individuals or populations to benefit from technological innovation because of insufficiently representative data, may widen service gaps between hospitals and primary healthcare institutions, as well as across primary healthcare institutions. To date, evidence on AI applications in China's primary healthcare institutions and their potential implications for health data poverty has not been systematically mapped.

To characterize AI models developed or validated using data from China's primary healthcare institutions, describe the datasets used in these studies, and assess the severity and distribution of health data poverty in this context.

We searched PubMed, Embase, Web of Science, CNKI and Wanfang Data for studies published between 2009 and 2025 on AI models in China's primary healthcare institutions. Data were extracted and synthesized across three domains: model characteristics, including algorithm type, validation strategy and performance; dataset characteristics, including data source, acquisition method and data content; and metadata distribution, including population coverage and handling of missing data.

57 studies were included. Existing AI models for primary care in China were mainly based on machine learning or deep learning and covered multiple application scenarios, including disease prediction and diagnosis. The most common disease areas were endocrine disorders, mainly diabetes, cardiovascular diseases, mainly hypertension, and mental health conditions. Datasets used to develop or validate these models showed limited accessibility, with no fully open dataset identified. Reporting quality was suboptimal, with a mean score of 6.53 out of 10; nearly one-third of studies did not report inclusion and exclusion criteria or methods for handling missing data. Dataset representativeness was also limited, with people younger than 45 years accounting for only 13.5% of reported study populations (5/37), and studies were predominantly conducted in more developed regions.

AI research in China's primary healthcare institutions is constrained by health data poverty, including limited data accessibility, suboptimal reporting quality and insufficient representativeness. Strengthening data linkage, developing high-quality research-ready datasets and establishing end-to-end governance may help build a stronger data foundation for primary healthcare.