Filling the Gap Without Filling the Role: AI's Place in Mental Healthcare

A recent HRSA workforce brief made one number impossible to ignore: the clinicians who deliver evidence-based behavioral healthcare are running out faster than the population that needs them. Writing about that brief, one line drew more reaction internally than any other: when members cannot access evidence-based care, some of them turn to AI instead.  

What AI is actually doing in that gap right now 
Members are not waiting for health plans to decide whether AI belongs in behavioral health. They have already brought it in themselves. Someone lies awake at 2 a.m. and opens an app instead of calling a number that will put them on hold. Someone who has never said "I think I might have a problem" out loud types it into a chat window first, because it is easier to admit something to a screen than to a stranger with a clipboard.  

None of that is members being reckless. It's members doing the rational thing when 6 in 10 psychologists are not accepting new patients at all, and the ones who are, carry an average wait of 48 days.¹ AI didn't create that gap. It just became the fastest thing available to fill it, and people go where the door is open. We write more about what a shrinking behavioral health workforce means for members in this post. 

Where it stops being enough 
Availability isn't the same as competence, and competence in this field is specific. A member who says "I have anxiety" might have anxiety. They might also have OCD, unresolved trauma or a substance use disorder wearing anxiety as its surface symptom, and telling those apart requires a structured clinical assessment, not a well-phrased response to a prompt. Our clinical team has spent years reviewing utilization data on the health plan side, and the pattern holds up every time: the members who got better were the ones matched to the specific treatment their actual condition called for. The ones who didn't were usually stuck in a cycle of generic care that never got underneath the presenting symptom. 

AI also can't do the thing measurement-based care exists to do, which is track whether a person is actually improving over time and catch it early when they're not. A PHQ-9 score that quietly worsens over three visits is a clinical signal that a trained provider is watching for. A chat log isn't being read that way, and it shouldn't be mistaken for something that is. In this four-minute video clip, we talk more about the role of AI as a a powerful tool for support, navigation and access, but not a substitute for the clinical judgment and human connection that trained therapists provide.

The responsible place for AI is a bridge, not a destination 
We don't think the answer is to keep AI out of this picture, and we don't think that's realistic anyway. The honest answer is that AI is well suited in the waiting room, not the treatment room. It can absorb the moment between "something is wrong" and "I'm in evidence-based care," offer psychoeducation, reduce the stigma of the first admission and keep someone engaged instead of giving up. What it shouldn't do is become the place someone quietly stays because it was available and a licensed clinician wasn't. 

That distinction only matters if there's actually somewhere better to hand someone off to. This is why NovaOne is built the way it is: validated assessment data drives a personalized recommendation for evidence-based, in-network care that is typically available in under five days, so a member has an obvious right answer to turn to instead of a chatbot that may not address the real underlying concern. Additionally, our platform includes self-guided resources that members can use for support while they wait. Every member who lands there instead of staying in a well-meaning holding pattern is a member who gets better and moves on, which is exactly what a shrinking workforce needs more of. 

AI's place in mental healthcare is not to replace the clinician who can treat what is actually wrong. It's to make sure fewer people fall through the cracks on the way to that clinician. Filling the gap and filling the role have never been the same job, and the workforce data says we cannot afford to confuse them for the next twelve years. 

If you're thinking through where AI fits into your own behavioral health strategy and where it should hand off to something else, we'd be glad to share what we are seeing. Connect with the NovaOne team here.

Sources:
1. HRSA Bureau of Health Workforce, Behavioral Health Workforce Brief 2025