Summary: A recent study from Lund University shows that an AI conversational assistant can perform psychiatric assessment interviews with greater diagnostic accuracy than commonly used mental health rating scales. In a sample of 303 people with confirmed psychiatric diagnoses, the AI assistant Alba conducted brief conversational interviews and provided DSM-5–based diagnostic suggestions. Alba’s assessments aligned more closely with participants’ confirmed diagnoses than the rating scales did, outperforming the scales in eight of nine disorders evaluated.
Alba was especially effective at distinguishing conditions that often have overlapping symptoms, such as depression and anxiety, where traditional rating scales frequently produce similar scores. Study participants also reported positive user experiences, describing the AI-led interviews as empathic, relevant, and supportive. The findings point to conversational AI as a scalable, person-centered tool to support clinical assessment, while preserving the central role of clinicians in treatment and care.
Key Facts:
- Higher diagnostic accuracy: The AI interview outperformed standard rating scales in eight of nine psychiatric conditions.
- Improved differentiation: Alba more clearly separated overlapping conditions such as anxiety and depression than conventional instruments.
- Positive user experience: Many participants described the AI interview as empathic, relevant, and supportive.
Source: Lund University
In the study, each of the 303 participants completed an online interview with the AI assistant Alba, which asked 15–20 open-ended questions about mental health symptoms and functioning. After the conversation, Alba produced diagnostic suggestions grounded in DSM-5 criteria. Participants also completed standardized rating scales that measure symptoms for the nine most common psychiatric diagnoses, allowing a direct comparison between AI-based conversational assessment and conventional scale-based screening.
The participant group included people with clinician-confirmed diagnoses of major depressive disorder, generalized anxiety disorder, obsessive-compulsive disorder, post-traumatic stress disorder, ADHD, autism spectrum disorder, eating disorders, substance use disorder and bipolar disorder, along with a healthy control group. Across this diverse sample, Alba’s interview-based assessments matched the confirmed diagnoses more consistently than standardized rating instruments.
One notable advantage of the conversational format was its ability to reduce diagnostic co-dependencies. Conventional rating scales often produce correlated high scores across multiple conditions—particularly anxiety and depression—making it harder to separate the primary problem. Alba’s DSM-5–aligned interview structure and follow-up probing improved differentiation between conditions with similar symptom profiles.
Participants evaluated the AI-led interview positively. Many described it as empathic and engaging, noting that an AI-conducted interview can be completed safely at home before a clinical appointment. That preliminary, standardized assessment can help streamline workflows in clinical settings, support triage, and provide clinicians with a well-documented starting point for evaluation while not replacing the clinical judgment of psychologists or physicians.
Analyses the entire diagnostic manual — not just single conditions
According to Professor Sverker Sikström, leader of the research team and founder of Talk To Alba, the study represents a step forward for digital mental health assessment. Unlike previous research that often focused on single diagnoses or lacked diagnostic justification, Alba evaluates and can justify potential diagnoses across the full range of DSM-5 categories included in the study, generating explanations tied to established diagnostic criteria.
Fact box: What is Talk To Alba?
Talk To Alba is an online AI platform designed to support assessment, treatment and administrative tasks in mental healthcare. Its features include AI-conducted clinical interviews, automated DSM-5–based diagnostic suggestions with documented justification, CBT-based patient tools, transcription and clinical note generation, and clinician-facing summaries to support case review and follow-up. Alba is in use at clinics in Sweden and internationally and is developed by TalkToAlba AB.
Key Questions Answered:
A: In this study, yes. The AI assistant showed higher diagnostic agreement with participants’ confirmed diagnoses than standardized scales in eight of nine evaluated conditions.
A: The AI demonstrated better separation between commonly overlapping conditions—most notably anxiety and depression—than conventional rating tools.
A: Most participants rated the experience as empathic, relevant, and supportive.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full for accuracy.
- Additional context was added by editorial staff where relevant.
About this AI and mental health research news
Author: Lotte Billing
Source: Lund University
Contact: Lotte Billing – Lund University
Image credit: Neuroscience News
Original research: Open access. “Generative AI-assisted clinical interviewing of mental health” by Sverker Sikström et al., published in Scientific Reports (DOI: 10.1038/s41598-025-13429-x).
Abstract
Generative AI-assisted clinical interviewing of mental health
Traditional mental health assessment relies on interviews by trained clinicians, an approach that is effective but resource-intensive and subject to variability in expertise and documentation. Advances in large language models and conversational AI offer a potential route to standardized, scalable interviews that emulate clinician-administered assessments. This study evaluated an AI interview system, TalkToAlba, designed to follow DSM-5 diagnostic criteria and to provide documented diagnostic reasoning.
Participants (N = 303) included people with clinician-confirmed diagnoses across nine common mental disorders and healthy controls. The AI conducted diagnostic interviews, estimated the likelihood of each disorder, and produced explanatory justifications. Results showed that the AI interview produced higher agreement (Cohen’s Kappa), sensitivity, and specificity for detecting clinician-diagnosed disorders compared with established symptom rating scales. The AI also showed lower diagnostic co-dependencies among categories and received high user ratings for empathy and relevance.
These findings indicate that AI-powered clinical interviews can be accurate, standardized, and user-friendly tools for assessing common mental disorders. Their scalability and potential to reduce clinician workload make them a promising complement to traditional diagnostic methods, supporting triage and preliminary assessment while preserving the clinician’s central evaluation and treatment roles.