Halfway through a teletherapy session, you notice your therapist pause mid-sentence. They’ve just heard the soft clatter of your keyboard, or seen your eyes flick sideways to a second screen. You’re technically “there” on video, but a work email just popped up, laundry is buzzing, someone’s messaging you, and suddenly the session is sharing space with three other tabs in your brain.

On the therapist’s end, it can feel like talking into a room where the lights are on but no one is fully home: Reflections don’t land the same, important questions drift past, and the conversation stays on the surface instead of dropping into the real stuff you came to talk about. Remote sessions make it easier than ever to answer email, work, do housework, or scroll during therapy, but that “divided attention” quietly dilutes the experience for both client and counselor.

Adaptive vs. maladaptive divided attention

But divided attention in teletherapy isn’t always the villain. Sometimes it’s actually part of what makes the work possible. Think of the client who shares their screen while building a dinosaur world or sketching on their tablet during a session. They’re not disappearing from the room; they’re giving their nervous system something steady and predictable to do so the rest of them can stay present. It’s no different from playing Uno with a teenager in person while you talk about the hard stuff (hands busy, guard slightly lowered, words coming a bit easier).

That’s adaptive divided attention: The side activity supports the conversation instead of competing with it. In contrast, mindless scrolling, answering emails, or “I can only do laundry during therapy” energy usually pulls the person away from themselves. That’s when multitasking slips into avoidance, less a self-regulation tool and more a shield against feeling, a state of maladaptive divided attention. The question isn’t “Are they multitasking?” so much as “Is this helping them stay with the work, or helping them escape it?”

NeuroAI: treating attention as a dynamic system

In my NeuroAI work, I have been treating attention in digital therapy as a dynamic system, not a simple on/off switch. Unlike traditional engagement metrics that treat attention as binary, this framework recognizes attention as multidimensional and dynamic. It behaves like a marble moving across a landscape with deep valleys where it tends to settle, what complexity science calls “attractor states.”

An anxious patient’s attention keeps rolling back into the “catastrophe” valley; an ADHD patient may oscillate between the “game” valley (stimulation attractor) and the “conversation” valley (engagement attractor). Multimodal behavioral analysis lets us see this in real time by combining signals from video, audio, text, and interaction patterns instead of relying on a single channel. If a client is gaming during a session, the system can infer whether that behavior is anchoring them enough to stay present or pulling them out of the therapeutic landscape entirely. The goal isn’t to rip the marble out of the valley by force, but to gently reshape the landscape so attention settles more often in states that support insight, regulation, and connection.

Measuring presence and cognitive load

Under the hood, that means treating “presence” as something we can measure and optimize. When someone is truly engaged, their different channels tend to line up: their words, response latency, typing rhythm, and language complexity all tell a coherent story. During split-screen therapy, those channels often desynchronize; “I’m listening” in speech paired with long pauses, flattened language, or erratic interaction patterns.

A NeuroAI system can quantify this coherence, estimate how much mental “bandwidth” is going to the therapeutic channel versus the distraction channel, and recognize when a secondary activity (like sketching or a simple game) is actually stabilizing rather than fragmenting attention. Layered onto that is cognitive load: Every question, exercise, or exposure has a mental cost. By modeling interventions as steps with different cognitive demands, the system can recommend paths that stay within a person’s capacity, sometimes deliberately allowing a light, repetitive task in the background to offload anxiety so there is enough room left for emotional processing.

Redefining what being present looks like

In practice, what this kind of NeuroAI does is give clinicians permission to redefine what “being present” looks like. Instead of assuming that a neurodivergent patient who’s doodling or quietly gaming is checked out, we can actually see when that movement or background task is helping their nervous system stay online.

There’s growing evidence that for many people with ADHD, fidgeting and small repetitive movements improve focus rather than undermine it. A system that can read those patterns in real time turns what used to feel like “non-compliance” into usable clinical information: This is how this brain pays attention.

Meeting Gen Z patients where they live

It also meets patients where they already live. Gen Z and younger adults are used to parallel processing; most report using multiple screens or media streams at the same time as a normal part of daily life. For some of them, a light game or sketching during a session works like a built-in dual-attention task, similar in spirit to how EMDR pairs a traumatic memory with an external focus to make processing safer.

A NeuroAI layer can tell the difference between that kind of regulated split-screen and the kind that quietly drains the session, when the background activity starts pulling language, coherence, and emotional engagement away from the work.

Enhancing clinical judgment with AI

At a systems level, this doesn’t replace clinical judgment; it sharpens it. You still need clear telehealth boundaries (no sessions while driving or running errands, for example) as current safety guidance already stresses. But instead of guessing who’s “here” and who isn’t, clinicians get a live, nuanced picture of attention they can act on. That means fewer missed crises hiding behind “I’m fine,” more room for neurodivergent and Gen Z patients to show up as they are, and a version of digital care that feels less like fighting the medium and more like using it.

Ronke Lawal is the founder of Wolfe, a neuroadaptive AI platform engineering resilience at the synaptic level. From Bain & Company’s social impact and private equity practices to leading finance at tech startups, her three-year journey revealed a $20 billion blind spot in digital mental health: cultural incompetence at scale. Now both building and coding Wolfe’s AI architecture, Ronke combines her business acumen with self-taught engineering skills to tackle what she calls “algorithmic malpractice” in mental health care. Her work focuses on computational neuroscience applications that predict crises seventy-two hours before symptoms emerge and reverse trauma through precision-timed interventions. Currently an MBA candidate at the University of Notre Dame’s Mendoza College of Business, Ronke writes on AI, neuroscience, and health care equity. Her insights on cultural intelligence in digital health have been featured in KevinMD and discussed on major health care platforms. Connect with her on LinkedIn. Her most recent publication is “The End of the Unmeasured Mind: How AI-Driven Outcome Tracking is Eradicating the Data Desert in Mental Healthcare.”