Artificial intelligence is becoming part of everyday medical practice. Physicians use it to search clinical literature, review treatment options and find information about medications. Patients increasingly turn to AI chatbots to understand symptoms, discuss emotional difficulties and decide whether they need professional help.
In psychiatry, these developments raise questions that go beyond technological convenience. Mental health care depends on clinical judgment, an understanding of the patient's personal history and the ability to recognize risks that may not be immediately apparent. An AI system can provide a convincing explanation without necessarily providing a correct one.
A study published in JAMA on October 15, 2026, offers a clearer picture of how clinicians are using AI in practice. Its findings are important, but they also reveal a distinction that deserves particular attention. The growing use of artificial intelligence in medicine does not, by itself, demonstrate that it improves diagnosis, treatment outcomes or patient safety.
The researchers examined the use of OpenEvidence, an AI-enabled clinical information platform, among healthcare professionals in the United States.
The study covered the period from January 2024 through June 2026 and included 952,008 clinicians who submitted approximately 271 million queries. More than 556,000 of the users were physicians.
The scale of adoption was substantial. Monthly active users increased from 355 in January 2024 to more than 703,000 in June 2026. Over the same period, the average number of monthly queries per active user rose from 18 to 39.5.
The researchers also examined why clinicians were using the platform. Among physicians, 33.1% of queries concerned treatment and clinical management, while 13.9% concerned diagnosis.
These findings suggest that physicians are turning to AI primarily to support practical clinical decisions rather than simply to identify diseases.
For psychiatry, this distinction matters. Psychiatric treatment often involves several possible medications, differences in tolerability, coexisting medical conditions and decisions that must be reviewed over time. Rapid access to clinical information may be useful, particularly when a physician needs to check unfamiliar evidence or compare treatment options.
However, the study did not establish whether the answers generated by the platform were correct. It did not measure whether patients improved, whether prescribing errors declined or whether clinical decisions became safer.
The study was also limited to users of one platform. Its findings should not be interpreted as a representative measure of AI use among all psychiatrists or physicians.
The central conclusion is therefore straightforward. AI is being used extensively in clinical medicine, particularly for questions related to treatment. Whether this translates into better care remains a separate question.
Psychiatric diagnosis is not based on a single laboratory result or imaging finding. It requires an assessment of symptoms, their duration, the course of illness and their effect on daily functioning.
A patient reporting poor sleep, reduced concentration and persistent fatigue may be experiencing depression. Similar complaints can also occur in anxiety disorders, bipolar disorder, medical illnesses, substance-related conditions or as adverse effects of medication.
The distinction depends on clinical context.
A psychiatrist may need to establish whether the patient has experienced previous episodes of elevated mood, whether symptoms followed a medication change, whether there is a history of trauma and how the current condition differs from the patient's usual functioning.
AI can help organize information about these possibilities. It cannot reliably establish their significance without an adequate clinical assessment.
This is particularly important when evaluating depression, where an incomplete assessment of previous manic or hypomanic symptoms may lead to inappropriate treatment decisions.
The same principle applies to psychotic disorders, including schizophrenia. A description of unusual beliefs or perceptual experiences does not automatically establish a diagnosis. The clinician must consider the person's history, behavior, mental state, medical conditions and possible substance use.
Psychiatric assessment also involves observing how a patient communicates, responds to questions and describes experiences. These observations may influence diagnosis and treatment, yet they are difficult to capture through a brief written exchange with an AI system.
Medication management is one of the areas where AI-assisted clinical information tools may offer practical benefits.
Psychiatrists frequently need to consider drug interactions, adverse effects, dose adjustments and the implications of combining several medications. Treatment may also need to account for cardiovascular disease, metabolic disorders, pregnancy, neurological conditions and other medical factors.
A well-designed clinical information system can help a physician locate relevant prescribing information and identify issues that require closer examination.
That does not make it an independent prescribing authority.
AI-generated answers may omit important contraindications, rely on outdated information or present uncertain recommendations with excessive confidence. Even when an answer is factually correct, it may not be appropriate for a particular patient.
Consider a patient with treatment-resistant schizophrenia who is receiving clozapine. Safe management requires more than knowing the recommended dose. It involves appropriate laboratory monitoring, assessment of adverse effects, attention to interactions and ongoing evaluation of clinical response.
An AI tool may help retrieve information about these requirements. It cannot replace the treating psychiatrist's responsibility for interpreting results and deciding what action is necessary.
Similar considerations apply to complex antidepressant treatment, mood stabilizers and medication changes in patients with several psychiatric or medical diagnoses.
The safest role for AI in this setting is to support access to information that the clinician independently evaluates.
There is growing research interest in using machine learning to identify patterns associated with psychiatric disorders.
Researchers are investigating whether AI can analyze combinations of clinical symptoms, cognitive performance, electronic health records, neuroimaging findings and biological measurements.
The long-term objective is to improve diagnostic precision and identify patients who may respond differently to treatment.
These possibilities are scientifically important, but they should not be confused with established clinical capabilities.
Psychiatric disorders are heterogeneous. Two patients with the same diagnosis may have different symptoms, illness trajectories, medical conditions and treatment responses.
A model developed using data from one population may perform less accurately in another. Its predictions may also be affected by differences in language, culture, healthcare access and the quality of the information used during training.
Even strong performance in a research dataset does not necessarily mean that a system will improve outcomes in routine clinical practice.
Before a diagnostic AI tool can be relied upon, it needs appropriate external validation, assessment of potential bias and evidence that its use provides a meaningful clinical benefit.
At present, a comprehensive psychiatric assessment remains essential. AI may assist with selected tasks, but it should not be presented as a substitute for professional diagnosis.
AI chatbots are increasingly accessible to people seeking explanations for anxiety, depression, relationship difficulties or distressing thoughts.
Their appeal is understandable. They are available at any hour, do not require an appointment and may feel easier to approach than another person.
Some users may find them helpful for organizing thoughts, preparing questions for a consultation or obtaining general information about mental health.
Problems arise when a chatbot is treated as a clinician.
An AI system may respond sympathetically without recognizing that a person requires urgent assessment. It may fail to distinguish ordinary distress from a severe psychiatric condition or provide advice that does not account for the patient's medical history.
There are particular concerns when a person is experiencing suicidal thoughts, psychosis, mania or severe behavioral disturbance.
In such situations, the quality of a response cannot be judged solely by whether it sounds supportive. Appropriate care may require immediate risk assessment, direct clinical contact and a treatment plan adapted to the person's condition.
Chatbots can also produce inaccurate statements while presenting them confidently. This creates a risk that users will place more trust in an answer than its reliability warrants.
For these reasons, AI-generated mental health advice should not replace assessment by a qualified professional.
Psychotherapy presents a different set of questions.
Certain digital tools can support established psychological interventions. For example, structured applications may help patients record mood changes, complete therapeutic exercises or monitor progress between appointments.
Some automated systems are being studied as ways to deliver elements of psychological support.
However, providing information about a therapeutic technique is not equivalent to conducting psychotherapy.
Effective psychotherapy involves a clinical relationship, ongoing assessment and the ability to adapt treatment to the patient's changing needs.
In cognitive behavioral therapy, for example, a therapist does more than suggest alternative interpretations of distressing thoughts. Treatment requires an individualized formulation, identification of maintaining factors, carefully selected interventions and regular evaluation of progress.
A digital tool may assist with homework or symptom monitoring. It does not automatically provide the clinical judgment required to decide whether a particular intervention is suitable.
The distinction becomes especially important when treating patients with complex trauma, severe depression, personality difficulties or coexisting psychiatric conditions.
For such patients, psychotherapy may need to be coordinated with psychiatric treatment and other forms of care.
The relevant question is not whether AI can imitate a therapeutic conversation. It is whether a particular digital intervention has demonstrated safety and effectiveness for a clearly defined group of patients.
The introduction of AI into clinical care creates responsibilities for both healthcare organizations and individual clinicians.
One concern is the possibility of incorrect or incomplete information. Another is the handling of sensitive patient data.
Psychiatric records may contain details about trauma, family relationships, substance use, sexual experiences and other highly personal matters. Such information requires particularly careful protection.
Before using an AI platform with clinical information, healthcare professionals need to understand how the system processes data, what information may be retained and whether its use complies with applicable privacy and professional requirements.
Patient confidentiality cannot be assumed simply because a platform is marketed for healthcare.
There is also a risk of automation bias, which occurs when people give excessive weight to a computer-generated recommendation.
A clinician may be less likely to question an answer that is clearly written, detailed and supported by apparently credible references. Yet the presentation of an answer is not evidence of its accuracy.
Clinical decisions should therefore remain open to independent verification, especially when they concern diagnosis, medication changes or urgent safety issues.
International organizations, including the World Health Organization, have emphasized the need for appropriate governance, transparency and human oversight in the use of AI for health.
These principles are particularly relevant to psychiatry, where an error may have serious consequences for a patient's health, safety and autonomy.
One of the most promising directions in psychiatric research is the development of more individualized approaches to treatment.
In principle, AI could help clinicians interpret complex combinations of clinical information and identify patterns that are difficult to recognize through conventional analysis.
The potential is significant, but personalization should not be reduced to a prediction generated by software.
A treatment plan must reflect the patient's diagnosis, previous treatment response, medical history, psychological characteristics, personal circumstances and preferences.
It also needs to change when the patient's condition changes.
At IsraClinic, individualized assessment and coordinated treatment planning are central elements of clinical care.
The clinic's Psychoergonomic Method of Clinical Psychiatry, developed by Valery Kravitz, provides an internally developed framework for considering the patient's psychological characteristics, clinical presentation, life circumstances and response to treatment together.
The method emphasizes comprehensive assessment, collaboration among specialists and ongoing review of the treatment plan.
It is important to distinguish this clinical framework from independently validated AI-based diagnostic systems or treatments supported by randomized controlled trials. The method should not be presented as having demonstrated superiority over established psychiatric approaches without appropriate comparative evidence.
The broader principle remains relevant. Clinical information becomes useful when it is interpreted in the context of the individual patient.
AI may eventually improve parts of that process. It does not eliminate the need for a physician to understand the person receiving treatment.
Patients do not need to avoid AI altogether. General educational information can be useful, and digital tools may help people prepare for a medical appointment.
The important issue is knowing where the limits are.
A chatbot may explain what an antidepressant is commonly prescribed for. It cannot determine whether that medication is appropriate for a particular person.
It may describe the symptoms of bipolar disorder. It cannot establish the diagnosis from a short conversation.
It may suggest questions to ask a psychiatrist. It should not be used to independently stop medication, change doses or interpret potentially dangerous symptoms.
Patients should also be cautious about entering identifiable medical information into systems whose privacy protections they do not understand.
If symptoms are persistent, worsening or interfering with daily functioning, professional assessment is more appropriate than repeated attempts to obtain a diagnosis from a chatbot.
Anyone experiencing an immediate psychiatric emergency should seek urgent medical assistance rather than relying on an automated conversation.
The findings published in JAMA demonstrate that AI-assisted information retrieval is becoming a familiar part of clinical work.
The next stage of research must address a more demanding question.
Does the use of these systems lead to better decisions and better outcomes for patients?
Answering that question will require studies that evaluate diagnostic accuracy, prescribing safety, clinical effectiveness and the consequences of AI use in real healthcare settings.
Research must also examine whether these technologies perform consistently across different populations and whether their benefits outweigh the risks.
Psychiatry may eventually benefit from tools that help clinicians recognize complex patterns, identify medication-related concerns and make better use of emerging scientific evidence.
Those developments would be valuable, particularly for patients with complicated conditions or a history of unsuccessful treatment.
For now, the appropriate approach is cautious integration.
Artificial intelligence can make medical information easier to access. It may become an increasingly useful clinical resource. But responsible psychiatric care still depends on professional judgment, an accurate understanding of the patient's condition and a therapeutic relationship built on trust.
The purpose of technology should be to strengthen those foundations, not replace them.
Patel VR, Finlayson SG, Liu-Hass M, et al. Clinician Use of an Artificial Intelligence–Enabled Clinical Information Platform. JAMA. Published online October 15, 2026. doi:10.1001/jama.2026.21868. Read the study.
World Health Organization. Ethics and Governance of Artificial Intelligence for Health. WHO guidance and publications.
Dr. Karny Jabotinsky-Rubin, MD, MSc, BSc
Dr. Karny Jabotinsky-Rubin is a physician associated with IsraClinic, a private psychiatric clinic in Tel Aviv, Israel. Her clinical and professional interests include psychiatric assessment, individualized treatment planning and the integration of scientific evidence into patient care.
IsraClinic provides psychiatric assessment, treatment and follow-up for patients in Israel and internationally. Its clinical services include in-person and online consultations in English, Russian and Hebrew.
Dr. Karny Jabotinsky-Rubin, MD, MSc, BSc
Published October 19, 2026.
IsraClinic, Private Expert Psychiatric Clinic, Tel Aviv, Israel.
This article is intended for general educational purposes and does not constitute medical advice, diagnosis or treatment. The information presented reflects the scientific evidence available at the time of publication and should not be used to make individual treatment decisions without consulting a qualified healthcare professional.
Artificial intelligence tools should not be used as substitutes for psychiatric assessment, emergency care or professional decisions concerning psychiatric medication.
