GIMBHI's Q&A with Emily York, Head of Clinical at Soulside AI

Emily York is a clinical psychologist and healthcare operator who has spent her career translating clinical insight into better care delivery, healthcare programs, products, and technology. She currently serves as Head of Clinical at Soulside AI, where she advises product and leadership teams on designing AI solutions for behavioral health that fit real clinical workflows and earn clinician trust. Alongside that work, she practices as a clinical psychologist within an interdisciplinary medical specialty clinic, supporting patients as they navigate the behavioral, psychological, and social aspects of living with complex chronic medical conditions. PMP-certified and clinically trained, Emily brings a dual perspective grounded in how care actually works and what it takes to design, implement, and scale solutions across complex healthcare settings. Her experience spans integrated behavioral health, telehealth implementation, program development, and clinical operations across Epic-based health systems. She holds a PhD in Clinical Psychology from Louisiana State University and completed her medical psychology residency at Duke University School of Medicine.

1. You’re Head of Clinical (advisory) at Soulside AI, but you came to this work as a practicing clinician, not as a founder or technologist. What drew you in, and what does a clinician bring to the table that a tech team can easily overlook?

My interest in AI started with burnout, not technology. Like many clinicians, I was expected to document during the visit or complete my notes afterwards, often long after the patient left. As a psychologist, I never felt comfortable typing while someone shared intimate details about their trauma, grief, or substance use, so my notes almost always waited until after hours. Over time, that wasn’t sustainable, and I eventually stepped away from clinical practice for a period. When I later learned about ambient AI, I found myself wondering whether it could help solve a problem I’d experienced firsthand: reducing documentation burden without asking clinicians to sacrifice their presence with patients.

The more I explored AI, the more I appreciated that many of the general-purpose tools weren’t designed around the realities of behavioral health practice. The conversation is the data, but the clinical picture extends well beyond the transcript. A transcript captures what was said; a clinically useful note has to capture what it means: affect, engagement, clinical reasoning, longitudinal patterns, and what can appropriately be inferred. It also has to support continuity of care across very different treatment settings and modalities.

That’s where I think clinicians contribute something technology teams can’t easily recreate. We know which distinctions matter clinically, which problems are worth solving, and what it will take for colleagues to trust a tool in everyday practice, because we’ve wrestled with those same questions ourselves. Those perspectives are difficult to layer onto a product after it’s built; they shape the questions that get asked at the beginning.

That’s one of the reasons I chose to become involved. I know many behavioral health clinicians have thoughtful concerns about AI, and I understand why. Rather than standing outside those conversations, I think clinicians have an important role in helping shape them. If these technologies are going to become part of behavioral health care, they should reflect the realities of clinical practice, earn the trust of both clinicians and patients, and ultimately strengthen the care we provide.

2. You’ve pointed to research on why digital health initiatives fail, and the common finding is that it’s rarely about the technology itself. What separates a tool clinicians come to trust from one they quietly abandon?

Years ago, during an electronic health record build, I proposed what seemed like a minor workflow change. A physician pushed back because it required one extra click, then walked me through how many clicks a single visit already took. That conversation stayed with me because it changed how I think about implementation. What looks like a small change from the outside can feel very different to the person doing the work every day.

Since then, I’ve paid as much attention to what happens after implementation as to the technology itself. Every new tool asks clinicians to change habits and workflows they’ve developed over years of practice. That’s a meaningful request, so the questions they ask are remarkably practical: Does this fit into my workflow? Can I trust the output? Does it help me provide better care? Am I still in control?

In behavioral health, trust extends beyond the quality of the note. Clinicians want to understand how the technology works, how data are used, and how to explain those issues to patients as part of informed consent.

One thing I’ve learned is that tools are rarely abandoned all at once. More often, clinicians gradually return to familiar routines because the technology creates friction or because they don’t feel heard when they raise concerns. Sustained use depends on ongoing partnership, where clinician feedback shapes the product over time. That’s where I think trust is really measured.

3. A clear thread runs through your work, from your clinical practice to your writing on loneliness and the Clubhouse model to your nonprofit work: human connection and community. How does that show up day to day, and why does it matter so much for patients with complex chronic conditions?

One of the things I’ve come to appreciate over the course of my career is that healthcare often focuses on helping people manage disease, while patients are trying to rebuild a life around it. Working in an integrated medical specialty clinic, I’ve seen how chronic illness affects much more than physical health. It changes routines, relationships, identity, and the activities that once gave people a sense of purpose. Over time, many people’s worlds become smaller, often in ways that aren’t obvious unless you ask.

That realization changed the kinds of questions I ask. I’m interested in symptoms, of course, but I’m equally interested in what life looks like outside the clinic. Who do you spend time with? What have you stopped doing because of your illness? What do you miss? What still gives you a sense of purpose? Those conversations often tell me as much about someone’s health as a symptom checklist.

I’ve watched patients reconnect with old friends, return to volunteering, join community groups, or rediscover activities they thought they’d lost. Those changes don’t alter the diagnosis, but they often change the way people experience living with it. They reconnect with sources of meaning, become more engaged in their health, and participate more fully in their lives. Those observations are increasingly reflected in the growing evidence supporting social prescribing and interventions that strengthen social connection.

That perspective has also shaped my work outside the clinic. Through my work with Lift UP, I’ve seen how meeting a basic need like food security can also create connection and belonging, and how much those things matter to people’s wellbeing. It’s one of the reasons I’m so interested in social prescribing. I think it encourages healthcare to recognize something many clinicians have observed for years: some of the most meaningful opportunities to improve health don’t happen during an appointment. They happen outside the walls of the clinic, when people reconnect with relationships, purpose, and the communities that help sustain them.

4. In the conversation about AI in behavioral health right now, what do you think we’re most at risk of getting wrong or missing entirely?

Something I’ve noticed is that, outside the behavioral health technology community, AI is often discussed as though it’s a single technology. In reality, it’s a broad range of tools with very different purposes, levels of risk, and evidence. Those distinctions matter because they shape the questions we should be asking.

An ambient documentation tool, a clinical decision support tool, a therapy chatbot, and an AI agent designed to support care between sessions all serve different purposes. Rather than expecting the same level of evidence or oversight from every application, I think we should ask what level of evidence, transparency, clinical oversight, and ongoing evaluation is appropriate for each use case and its level of clinical risk.

I’m encouraged to see initiatives like CHAI, the APA Labs Digital Badge Program, and VERA-MH helping move the field toward more thoughtful evaluation. We’re also seeing a growing evidence base across several areas of behavioral health AI. At the same time, we need to pay attention to how people are actually using these technologies. Many people are already using general-purpose AI for emotional support and mental health questions, regardless of whether those tools were designed for that purpose. If we want research to inform real-world care, we need to understand both the technologies we’re intentionally building and the ones people are already choosing to use.

Ultimately, I hope we continue asking broader questions. Are we solving an important clinical problem? Do we understand the people and workflows we’re designing for? Have we identified the risks and how to mitigate them? Are we building something that clinicians and patients will genuinely choose to use? Those questions matter because the goal isn’t simply to build better AI. It’s to improve care in ways that are meaningful, trustworthy, and responsive to the people we serve.