According to FAIR Health, mental health conditions accounted for 52.1 percent of patients with a telehealth claim in the first quarter of 2026.
At the same time, a 2026 multicenter study published in JMIR Formative Research analyzing 361 people with mental disorders found that data security and privacy were the leading reported barriers among respondents who identified obstacles to online psychotherapy, cited by 35.6 percent. Difficulty establishing therapeutic rapport followed at 19.8 percent.
Together, these findings point to a central challenge for digital healthcare: patients need convenient access to care, but they also need to feel understood and trust how their most sensitive information is handled.
The Importance of Nonverbal Behavior
Telehealth has expanded access to clinicians, reduced travel, and made follow-up appointments easier to manage. Yet a video consultation is not identical to an in-person encounter. Clinicians receive a patient's words and a partial view of their behavior, often without the broader context that helps them interpret hesitation, discomfort or uncertainty.
The question is whether technology can help close that gap without creating a new privacy problem.
Nonverbal communication has long played a role in clinical interactions. Facial expressions, eye contact, posture and pauses can provide context for what someone says, although none offers a reliable or universal explanation of a person's internal state.
Research published in 2025 on nonverbal behavior in telehealth visits examined the challenges of interpreting facial expressions and eye contact remotely. Other research has associated cues such as nodding and leaning forward with positive communication outcomes in simulated telehealth encounters.
These findings do not establish that an algorithm can accurately identify a patient's emotions or improve clinical outcomes. They do, however, raise a legitimate question about whether clinicians could benefit from additional context when communicating through a screen.
AI Can Play a Role in Facial Awareness
Artificial intelligence tools designed to analyze facial movements, vocal characteristics or other behavioral signals are one possible approach. Often grouped under the term emotional AI, these systems attempt to identify patterns in observable behavior. In healthcare, their potential value lies not in replacing clinical judgment, but in helping clinicians identify moments that may warrant further conversation.
That distinction is essential because observable behavior is not the same as emotional truth. A patient who avoids eye contact may be anxious, distracted, culturally accustomed to communicating differently or simply looking at the screen rather than the camera.
A calm expression does not necessarily indicate that someone feels comfortable. Any system that turns such observations into definitive emotional labels risks misleading the people it is supposed to help.
Surveillance and Monitoring Concerns
Healthcare also presents a different set of considerations from other environments where emotional analysis is being introduced. In employment or education, automated emotion assessments can create concerns about surveillance, power imbalances and decisions made about people without their meaningful consent.
In clinical care, additional behavioral information might support a conversation between a patient and a trained professional. But the potential benefit does not remove the need for safeguards.
Regulation reflects these concerns. The European Union's AI Act prohibits certain AI systems that infer emotions in workplaces and educational institutions, subject to limited exceptions, including medical or safety reasons.
The restrictions illustrate an important principle: the acceptability of an AI system depends partly on where it is used, why it is deployed, and what consequences follow from its assessments.
Healthcare applications should not be treated as automatically beneficial simply because they serve a medical purpose. Nor should every application of behavioral analysis be treated as equivalent to workplace surveillance. The relevant questions concern necessity, consent, accuracy, accountability and the degree of control retained by the patient and clinician.
Privacy architecture is part of that assessment. Many digital services collect information centrally, transmit it to external servers or retain it for purposes beyond the immediate interaction. When the information concerns a person's health, even data that appears less sensitive than a medical record can become revealing when combined with other information.
One way to reduce privacy risks is to process behavioral signals on a user's device rather than routinely transferring raw data to external servers. This approach can limit the amount of sensitive information shared beyond the device, although it does not eliminate the need for clear consent, appropriate data handling and safeguards against misuse.
New Systems Must Respect Patient Confidentiality
Developers should also consider whether AI systems can be trained using existing research and appropriately sourced datasets, reducing unnecessary reliance on personal data. These measures cannot guarantee privacy, but they provide a practical starting point for designing systems that respect patient confidentiality.
Local processing does not eliminate every privacy risk. Applications still need to explain what information is collected, what outputs are generated, how those outputs are stored and who can access them.
Consent must be meaningful, and healthcare providers remain responsible for meeting the legal and regulatory requirements that apply to their services. Even derived information can be sensitive if it is retained or used to make consequential decisions about a patient.
The principle is straightforward: collecting less information should be the starting point, not an afterthought.
The Clinician Must Still Make the Interpretation
It is equally important to define what these tools cannot do. Behavioral analysis is not mind reading, a lie detector or a clinical diagnosis. A system may identify a change in observable signals, but it cannot independently establish why that change occurred.
The interpretation belongs to the clinician, who can consider the patient's medical history, circumstances and own account of their experience. Consider a mental health consultation in which a patient describes feeling better but displays behavioral patterns that differ from previous appointments.
An analytical tool might flag the change as a reason to ask additional questions. It should not conclude that the patient is concealing distress or that their account is inaccurate. The value lies in opening a conversation, not closing it with an automated judgment.
The same principle applies to medication consultations with older patients. Someone considering a treatment change may hesitate without immediately explaining why. An additional signal could encourage a clinician to explore concerns about side-effects, practical difficulties or uncertainty.
But those possibilities must be established through conversation rather than inferred from facial movements alone.
Repeated consultations may offer opportunities to examine changes over time, provided the information is collected appropriately and interpreted cautiously.
Even then, differences in lighting, camera positioning, disability, cultural expression and individual behavior can affect what a system observes. Clinical validation must establish whether a tool is sufficiently reliable for its intended use, rather than assuming that more data automatically produces better care.
Incorporating More Emotional AI Into Digital Consultations
Ultimately, telehealth does not need to reproduce every aspect of an in-person appointment to be effective. It needs to give clinicians the information they require while respecting patients' autonomy and privacy. Additional context is useful only if it improves understanding without introducing disproportionate risks.
As healthcare providers consider incorporating emotional AI into digital consultations, they should begin with three questions: Does the technology address a clearly defined clinical need? Has it been independently evaluated for that use? And can it deliver meaningful value while minimizing the collection, transmission and retention of sensitive information?
The next step for healthcare providers is to pilot these systems under clear consent, privacy and clinical-validation standards before deploying them at scale.
Patients should understand what is being analyzed, clinicians should retain responsibility for interpretation, and providers should measure whether the technology improves consultations in practice. Digital healthcare will earn trust not by measuring everything it can, but by demonstrating that the information it collects is necessary, responsibly handled and genuinely useful to patient care.
Jonathan Jenin is an executive working with MorphCast, a company developing browser-based emotional AI technology. His work focuses on the application of behavioral signal analysis in digital interactions.
The views expressed are the author's own.