In the Middle, Not on Top: AI-Mediated Communication for Patient-Provider Care Relationships

Abstract: Relationship-centered care relies on trust and meaningful connection. As AI enters clinical settings, we must ask not just what it can do, but how it should be positioned to support these values. We examine a "middle, not top" approach where AI mediates communication without usurping human judgment. Through studies of CLEAR, an asynchronous messaging system, we show how this configuration addresses real-world constraints like time pressure and uneven health literacy. We find that mediator affordances (e.g., availability, neutrality) redistribute interpretive work and reduce relational friction. Ultimately, we frame AI mediation as relational infrastructure, highlighting critical design tensions around framing power and privacy.

• The paper demonstrates that mediating AI redistributes interpretive work, enhancing communication despite time constraints in clinical settings.
• It introduces the CLEAR system to translate, clarify, and summarize clinical information, reducing social friction and maintaining continuity in care.
• Empirical findings reveal increased patient confidence and decreased provider burden, alongside challenges like narrative stabilization and privacy risks.

AI as Relational Infrastructure in Patient-Provider Communication

Motivation and Positioning

The paper "In the Middle, Not on Top: AI-Mediated Communication for Patient-Provider Care Relationships" (2604.00643) examines a critical question in the integration of AI within healthcare: how to position AI such that it addresses operational constraints while sustaining the core relational dynamics essential to patient-provider interactions. Rather than privileging prescriptive or autonomous AI roles, the authors empirically articulate a "middle, not top" approach, where AI mediates but does not supplant human judgment or accountability. The study focuses on resource-limited care contexts characterized by high clinical throughput, brief and fragmented encounters, and heterogeneous patient health literacy and language skills.

CLEAR System: Mediated Communication in Practice

The authors instantiated their approach through CLEAR, an AI-mediated asynchronous communication system for patients and providers. CLEAR's design explicitly avoids the proxy clinician paradigm. Instead, its affordances center around translation, clarification, contextual continuity, and work redistribution for both patients and providers:

• For patients, CLEAR enables post-visit review of instructions, translation of clinical terminology, and iterative drafting of follow-up queries. This asocial mode allows repeated clarification—minimizing social cost, embarrassment, and hesitation typically incurred when confronting providers with "basic" questions.
• For providers, CLEAR summarizes patient interactions, aggregates clarifications, and maintains continuity artifacts to reduce preparation overhead, particularly in settings with discontinuous care.

Empirical studies (formative: 6 providers, 20 patients; user: 8 providers, 20 patients) demonstrate that CLEAR's mediator design increases patient confidence in communication and understanding, while also alleviating provider burden by shifting interpretive tasks outside the time constraints of clinical encounters.

Empirical Findings

The deployment and analysis of CLEAR yield several mechanisms and tensions inherent to the "AI in the middle" model:

• Temporal Redistribution: Shifting interpretive labor (clarification, preparation) into pre- and post-encounter periods leads to higher-quality interactions during time-limited consultations.
• Reduction of Social Friction: The asocial mediation channel enables patients to refine and pose questions without the affective barriers of direct clinician interaction, supporting more nuanced engagement over time.
• Continuity and Narrative Artifacts: Persistent, structured summaries maintain context across multiple providers and visits, mitigating fragmentation in care.
• Consistency, Neutrality, and Trust: Patients and providers perceive AI mediation as more consistent than direct interactions, especially concerning technical jargon. While this consistency fosters trust, it can also obscure clinical uncertainty and nuance, leading to potential narrative stabilization where authoritative summaries omit key complexities.
• Relational Risks—Privacy and Framing Power: AI mediation introduces risks related to persistent records and the prioritization of certain details over others, thereby amplifying privacy concerns and the possibility of "fixing" clinical narratives in ways that may not map to underlying uncertainty or patient intent.

Design Implications and Theoretical Impacts

The "in the middle" construct has significant theoretical and practical implications for the development and deployment of AI in healthcare communication:

• Preservation of Human Authority: By confining AI to an interpretive, mediating role, the model maintains accountability and decision-making with providers, opposing trends in fully autonomous or recommendation-driven health AI.
• Support for Relationship-Centered Care: The system bolsters relational continuity and scaffolds communication, rather than substituting for relational work—a misalignment often observed in optimization-centric AI deployments.
• Risks of Structural Mediation: The centralization and persistence of interpretive summaries introduce new axes of framing, stabilization, and privacy exposure. System design must therefore explicitly architect controls to support patient consent, avoid over-authoritative summaries, and ensure clinicians remain the locus of interpretation.
• Scalability and Applicability: The mediation stance is especially promising for resource-limited, high-volume care environments characterized by episodic provider contact and diverse patient populations. The underlying mechanisms—temporal redistribution, continuity scaffolding, and social friction reduction—are generalizable across similar communication-constrained contexts.

Future Directions

The paper's findings invite several avenues for subsequent inquiry:

• Safeguards Against Overstabilization: Methods for encoding clinical uncertainty, dynamically updating narrative summaries, and supporting context-sensitive privacy warrants further research.
• Boundary Work: Determining for which clinical situations or specialties "AI in the middle" is appropriate, and developing robust guardrails to prevent role slippage toward prescriptive or autonomous AI deployments.
• Expanding Modalities: Extending mediation beyond text-based messaging (e.g., multimodal explanations, adaptive literacy targeting (Huang et al., 10 Nov 2025)) and examining their impact on relationship-centered metrics.

Conclusion

The "middle, not top" positioning for AI-mediated communication offers a theoretically nuanced and practically viable design stance for relationship-centered care. By mediating but not displacing human-human communication, systems like CLEAR can redistribute interpretive work, ameliorate social barriers, and maintain continuity in resource-constrained environments, while surfacing new challenges around privacy, framing, and the stabilization of clinical narratives. The approach reframes healthcare AI as relational infrastructure, emphasizing the necessity for explicit design constraints and ongoing critical evaluation of AI's role in clinical relationships.