AI in primary care runs into system design long before it reaches the exam room
Artificial intelligence may be advancing quickly in health care, but primary health care systems do not absorb new tools at the same speed. A comparative qualitative study in primary health care examined how readiness for AI differs between Quebec and Iran, and the central message was blunt: adoption depends less on whether the technology exists than on whether the surrounding system can support it.
Researchers used semi-structured interviews and focus group discussions with key informants in both settings, then applied a framework-guided content analysis built around the four domains of the Primary Care Evaluation Tool: stewardship, financing, resource generation, and service delivery. This design made the study less about abstract enthusiasm for AI and more about the practical conditions that decide whether a digital tool can be absorbed into routine care without destabilizing the work already being done at the front line.
Governance and financing emerged as the first barriers to overcome
Across both contexts, governance- and financing-related challenges surfaced repeatedly. Participants described those issues as central determinants of whether AI could be introduced responsibly into primary care, and the pattern was consistent enough to show that technical sophistication alone does not create implementation readiness.
In Quebec, the discussion centered more often on operational and ethical friction points. Workflow integration, transparency, and professional trust shaped how clinicians viewed the arrival of AI into daily practice. Even where digital capacity is stronger, a new algorithm still has to fit into appointment flow, documentation habits, referral patterns, and the expectations of clinicians who are responsible for the final decision. If the tool adds uncertainty, opacity, or extra steps, adoption slows quickly.
In Iran, the obstacle course was more basic. Participants emphasized weaknesses in governance stability, financing mechanisms, and digital infrastructure as primary barriers. The findings suggest that AI implementation in a middle-resource system can stall before clinical questions are even reached, because the foundation needed to support data flow, oversight, and sustained investment is incomplete.
Interoperability may be the quiet constraint that determines whether AI scales
Limited data interoperability appeared across both settings and deserves particular attention. AI systems depend on reliable, standardized, and transferable data. When records are fragmented or incompatible, the model may still function in theory, but the workflow around it becomes fragile, and the clinical value diminishes.
This is where the study becomes especially useful for policy design. A health system does not need more AI demonstrations if it cannot move information cleanly between services, sites, and providers. In primary care, the burden is amplified because the setting sits at the junction of prevention, chronic disease management, triage, and referral. AI tools that cannot draw from coherent data streams will struggle to support that complexity.
One sentence captures the core of the comparison: AI readiness tracked with system maturity, not with enthusiasm alone.
Ethical integration mattered in Quebec because the infrastructure gap was not the main problem
In Quebec, participants were less concerned about whether AI could be deployed at all and more concerned about how it would be deployed. That distinction is important. Once a system has a substantial digital base, the key hazards shift toward accountability, explainability, and the possibility that clinicians may be asked to trust recommendations that they cannot easily interpret.
Transparency becomes a clinical issue when a provider must explain why a machine-assisted recommendation should influence care. Trust becomes a workforce issue when clinicians feel they are being asked to adopt a tool without a clear understanding of its logic or its limits. Workflow integration becomes a safety issue when a system slows care or creates parallel processes instead of reducing administrative strain. In a high-resource environment, those are not peripheral concerns; they decide whether AI strengthens the consultation or complicates it.
Foundational investment remains the priority in lower-resource settings
The Iranian context pointed to a different threshold. Before a health system can debate sophisticated ethical guardrails, it needs stable governance structures, financing channels, and digital infrastructure capable of sustaining implementation. Those conditions shape whether AI can be tested, monitored, and maintained over time.
This is where phased implementation becomes clinically relevant. A system that lacks stable governance cannot reliably define who is accountable for oversight. A system without financing cannot support training, integration, maintenance, or iteration. A system without infrastructure cannot move the data required for training or use. In that setting, AI is not primarily a question of model performance; it is a question of whether the health system can absorb and sustain the technology without widening fragmentation.
Workforce capacity and standardization sit at the center of implementation
Across both countries, adaptive governance, sustainable investment, data standardization, and workforce capacity-building emerged as recurring requirements. None of those elements can be treated as optional extras. Each one affects whether AI is an aid to care or another disconnected layer on top of already strained primary care services.
Workforce capacity is especially important because AI implementation changes how clinicians interpret information, delegate tasks, and interact with digital systems. Even a useful tool can fail if the team using it has not been prepared for changes in data quality, workflow design, or decision support. Standardization matters for the same reason: inconsistent data inputs produce inconsistent outputs, and inconsistent outputs erode trust.
The study also makes clear that readiness is not a static label. A system can be strong in one domain and vulnerable in another. Quebec’s stronger operational base still left room for concern about trust and transparency. Iran’s more constrained environment faced barriers at the level of system stability itself. Those differences shaped what each system needed first.
A phased model fits primary care better than a universal rollout plan
The comparative framework argues against a single implementation playbook. Primary health care is too dependent on local governance, financing, and digital maturity for that approach to work well. Instead, AI integration has to follow the system’s actual level of readiness.
In practical terms, that means high-resource settings may spend more effort on ethics, transparency, and clinical workflow alignment, while middle-resource settings may need to invest first in governance, infrastructure, and financing continuity. The sequencing is not cosmetic. It determines whether AI is absorbed into routine care or remains an isolated experiment with limited reach.
The study’s message is straightforward and clinically consequential: AI can support primary care, but only after the system beneath it is prepared to carry the load. Where governance is unstable, the technology will struggle to stay in place. Where interoperability is weak, its output will be limited. Where clinicians do not trust the workflow, its use will remain partial.
The remaining challenge is not whether AI can enter primary care. It is whether the health system can support the conditions that make its use safe, durable, and clinically meaningful.