By Sara Eklove, RVP, AllOne Health
It’s impossible to talk about the future of Employee Assistance Programs without talking about AI. Many are excited about the potential of what can AI do, but the more important question is: what should it do?
That distinction matters more than anything else right now. Because in mental health, capability without intention can quickly miss the point, and may even have dangerous consequences.
Here’s what AI can do well.
It can meet people instantly, at any hour, in any state of mind. It removes friction at the most critical moment: the first step. It might be able to help someone articulate what they’re feeling when they don’t have the language. It can identify patterns, triage needs, and guide someone toward the right kind of support. In a world where access is often the barrier, AI can open the door.
Here’s what it can’t do (and shouldn’t try to).
AI cannot sit with someone in crisis. It cannot interpret silence, hold complexity, or build the kind of trust that changes outcomes. It cannot replace the clinical judgment, empathy, and human connection that define effective care.
And in a time when mental health support is imperative across organizations and our communities, it’s very important that we ensure AI is positioned as a tool to enhance care, but not replace the work that happens between clinician and individual.
So what should AI do? A clinical perspective on where this is headed
AI has been unleashed and is already shaping norms around how people engage with their own mental health—whether the industry is ready for it or not.
A growing number of individuals are turning to AI tools to ask deeply personal questions, process emotions, and even simulate therapeutic conversations. Recent national survey data shows that roughly one in three adults have used artificial intelligence chatbots for health-related information or advice, and about one in six adults have used them specifically for mental health information or support. According to a Kaiser Family Foundation tracking poll on health information, trust, and artificial intelligence use (2026). That ranges from “Why do I feel this way?” to role-playing difficult conversations, to seeking advice they may not yet feel comfortable asking another human.
These findings highlight a rapid shift in behavior: AI is already functioning as a first-touch point for health questions—and increasingly, for emotional and psychological support—well before many individuals ever engage with formal clinical care systems.
That trend should get our attention.
Because what sits underneath it is not just curiosity—it’s unmet need. It’s access gaps. It’s the reality that for many people, talking to something feels easier than talking to someone, at least at first.
But we need to be very clear about what exactly AI is in this context. It’s not a clinician, it’s a pattern recognizer. It is built on probabilities, trained on vast datasets. It aggregates information across millions of inputs – but it does so without clinical judgment, without nuance, without lived experience, and without the ability to understand context in the way a trained human provider does. It is not conscious, it doesn’t feel, and it does not hold space or responsibility.
And that distinction becomes critical when people begin to form perceived relationships with it.
We are already seeing early signals of this: individuals describing AI as a confidante, a companion, even a source of emotional validation. While that can reduce immediate feelings of isolation, it also introduces new clinical questions:
- What happens when affirmation replaces appropriate challenge?
- How do we ensure accuracy when AI is generating responses probabilistically, not diagnostically?
- Where is the line between support and over-reliance?
- And what safeguards are in place when someone is in distress or at risk?
It is also important to recognize the business model behind many widely used AI platforms. When a product is “free,” the user is not the customer in the traditional sense—the user is often the source of the value. Attention, engagement time, and conversational data become the underlying product being optimized. That optimization is typically designed around keeping users engaged, returning frequently, and continuing the interaction.
In a mental health context, that introduces a real concern: systems that are designed to be helpful and affirming can also unintentionally reinforce what a person is already feeling or expressing, creating a real concern and experience of “confirmation bias”. Because these models are trained to be responsive and coherent, they may reflect and amplify emotional tone in ways that feel validating in the moment—even when that validation is not clinically appropriate or balanced. This is where confirmation bias can emerge: a person expressing anxiety, distress, or distorted thinking may receive responses that mirror that framing rather than gently challenging or clinically grounding it.
Unlike a trained clinician, there is no built-in therapeutic responsibility to disrupt harmful thought patterns, introduce evidence-based reframing, or prioritize long-term wellbeing over immediate conversational continuity. Without appropriate guardrails, oversight, and escalation pathways, this can create a subtle but important risk: the experience of feeling understood in the short term without necessarily being guided toward healthier outcomes in the long term.
What to expect: How AI will show up in EAP
Within EAPs, AI will not—and should not—replace human care. But it will fundamentally reshape how people access it.
We should expect to see AI integrated in three primary ways:
- Gateway to access and guide to care
AI will increasingly serve as the first point of contact—helping individuals clarify what they’re experiencing, normalizing help-seeking, and guiding them toward the appropriate level of care. - Reduce administrative work and improve the counseling experience
From intake to follow-up, AI can streamline administrative steps that often delay care—removing barriers that cause people to drop off before they ever speak to a clinician. - Deliver real-time insights and proactive solutions (with safeguards)
When designed responsibly, AI can surface anonymized trends that help organizations understand where stress is building—by role, shift, or environment—allowing for more proactive intervention.
But the value of AI in EAP will ultimately be judged by one question:
Does it improve access to care, help connect people to human support sooner, and when used appropriately with clinical guidance, does it extend care between sessions as a supplemental mental health maintenance tool?
In this way, AI should be used to solve the right problems.
What to look out for: Questions leaders (and customers) should be asking
As AI capabilities expand, not all EAP or mental health solutions will be created equally. Employers and clinical leaders should be asking more rigorous questions, including:
- Where does AI stop and human care begin?
Is there a clear, immediate pathway to a licensed clinician? - What are the escalation protocols?
How does the system detect and respond to risk (e.g., suicidal ideation, crisis indicators)? - Is there clinical oversight?
Are licensed professionals involved in designing, training, reviewing, and continuously monitoring the system? - How is accuracy validated?
What processes are in place to audit outputs and correct errors? - How is data handled?
Are interactions confidential, and how is sensitive information protected? - Does this reduce or increase friction?
Is the experience actually making it easier for people to get help—or just adding another layer? - How often are these processes being reviewed and re-evaluated?
For AI to play a responsible role in mental health, it must operate within clearly defined clinical guardrails. That includes:
- Defined scope of use
AI should be positioned as a support and navigation tool—not a diagnostic or treatment provider. - Risk detection and escalation protocols
Systems should be designed to recognize language associated with distress or crisis and immediately escalate to human intervention. That may include warm handoffs to counselors, crisis lines, or emergency services when appropriate. - Human-in-the-loop review
Licensed clinicians should be actively involved in reviewing interactions, refining responses, and ensuring alignment with evidence-based practices. - Bias monitoring and continuous improvement
Because AI learns from data, it can also inherit bias. Ongoing evaluation is essential to ensure equitable and appropriate responses across populations. - Transparency with users
Individuals should understand when they are interacting with AI, what it can and cannot do, and how to access human support at any point.
AI’s Impact on the Future of EAPs & Mental Health
It’s not a question of if AI becomes part of mental health support, it is already here and in play. The real question is how we integrate it responsibly into care systems that are, at their core, human. And that shift won’t only affect how people access support. It will also change how clinicians are trained, how they practice, and how care is delivered.
We’re likely to see AI show up in three practical ways inside EAP and clinical environments:
First, as a training and preparation tool—helping clinicians think through cases, explore different approaches, and reflect on potential interventions before or after sessions. Not as a decision-maker, but as a structured “thinking partner” that expands perspective.
Second, as a real-time support layer in documentation and workflow—helping reduce administrative burden so clinicians can stay focused on the person in front of them, not the paperwork around them.
And in some settings, AI may even be present in the background of care itself—not directing sessions, but available to clinicians as a prompt or reference point when helpful. For example, surfacing relevant clinical frameworks, suggesting evidence-based considerations, or helping organize thoughts in complex cases. The clinician remains fully in control, but with an additional layer of support available when needed.
This is where the opportunity—and responsibility—becomes clear. AI is not entering mental health as a replacement for clinical judgment. It is entering as an augmentation tool for clinicians who remain accountable for care.