Barriers to the implementation of artificial intelligence in primary care: a scoping review, 2020–2025

Background: Artificial intelligence (AI) is transforming healthcare, yet its implementation in primary care faces numerous barriers. This scoping review synthesizes recent evidence to identify and categorize these barriers, examining their manifestation across different implementation phases.

Methods: This scoping review was conducted according to Joanna Briggs Institute (JBI) methodology and reported in compliance with PRISMA-ScR guidelines. A comprehensive search strategy was implemented across PubMed, IEEE Xplore, and Google Scholar from January 2020 to June 2025, supplemented by citation chasing. Eligibility criteria were structured using the Population–Concept-Context framework, focusing on stakeholders in primary care, barriers to AI implementation, and primary care settings. Following deduplication, title and abstract screening was performed using ASReview, a machine learning-based active learning tool. Full-text screening was conducted independently by four reviewers, with disagreements resolved through discussion. Data extraction employed a structured form adapted from the Unified Theory of Acceptance and Use of Technology framework, operationalized through AI-Assisted Data Extraction (AIDE)—a novel human-in-the-loop tool integrating large language model (LLM) capabilities with mandatory human verification. All extracted data underwent explicit human approval before finalization. Data were analyzed using Python (3.13.5) and visualized to identify patterns in implementation barriers, stakeholder engagement, and AI applications across primary care contexts.

Results: The review included 141 studies. Qualitative studies were most prevalent (36.8%). A sizeable proportion of studies (53.6%) did not specify the AI model under evaluation. The most common objectives of implementation were Diagnosis (35.3%) and Quality Improvement (25.2%). Primary care physicians were the most frequent stakeholders (30.0%). “Attitude Toward Using” (73.8%), and “Performance Expectancy” (69.5%) were the most prominent barrier domains represented in literature. Pre-implementation studies constituted 58.0% of empirically evaluated studies, followed by implementation (30.7%) and post-implementation (11.4%).

Conclusions: The adoption of AI in healthcare continues to face significant barriers. Trust, data quality, and concerns around equity and bias remain central challenges. Our review highlights a shift toward pre-implementation research, provides a structured approach to anticipate challenges and guide more ethical, equitable, and context-sensitive AI integration in primary care.