Study Suggests AI Could Help Adapt Mental Health Care for Refugees and Migrants

AI enters a practical corner of mental health care

A new study from Karolinska Institutet suggests artificial intelligence could help speed up one of the hardest and slowest parts of global mental health care: culturally adapting treatment materials for people who do not share the language or assumptions of the original intervention.

The research focused on two common cognitive behavioral therapy techniques and tested whether AI-generated adaptations would be seen as culturally relevant and acceptable when compared with versions prepared by a human psychologist. The participants were Arabic-speaking refugees and migrants living in Sweden, Denmark and Germany, a group for whom access to evidence-based treatment in a familiar linguistic and cultural frame can be limited.

The result was striking. Participants judged the AI-adapted texts to be at least as acceptable as the human-adapted ones, and initially perceived them as even more culturally relevant. That does not mean AI is ready to replace clinicians or cultural expertise. It does suggest, however, that generative systems may be useful in narrowing a persistent access gap.

The bottleneck is not only therapy. It is translation and context.

Evidence-based psychological treatments are often developed, tested and distributed first in English. Adapting them for other languages and social settings is possible, but it can be slow, labor-intensive and expensive. Materials need not only translation but also adjustment in examples, tone and assumptions so they make sense to the people receiving them.

That challenge is especially acute for refugees and migrants, whose mental health needs may be substantial while service capacity remains thin. In practice, many patients encounter either long waits, materials that feel alien or a complete lack of treatment in a language they can use comfortably.

The Karolinska study points to a narrower, more operational role for AI than many public debates emphasize. Instead of promising diagnosis, counseling or autonomous psychiatric judgment, the technology here is being tested as an adaptation tool. That is a more concrete and arguably more defensible use case.

How the study was structured

Participants read CBT materials that had been translated and culturally adapted either by AI or by a psychologist, without being told which version they were seeing. The findings were published in JMIR Formative Research and reported by the researchers as an early indication that AI-generated adaptations can perform competitively on measures that matter to patients.

There were no significant differences in acceptability between the AI and human versions. On cultural relevance, the AI texts were initially rated higher. That outcome is important because culturally mismatched materials can reduce engagement even when the underlying therapeutic method is sound. If patients do not recognize themselves in the examples, language or framing, the intervention may never get a fair chance to work.

The researchers were careful not to overstate the conclusion. They describe the field as being in an early stage and stress that psychiatric uses of AI must be developed within clear quality and safety frameworks. That caution is appropriate. Mental health treatment involves trust, interpretation and risk, and errors in language or context can matter a great deal.

Why the result matters

The significance of the study lies less in proving that AI is better than psychologists and more in showing that AI may be good enough to accelerate a process that currently limits access. If a clinician or service provider can use AI to generate a credible first draft of culturally adapted material, then human review can focus on quality control instead of starting from scratch every time.

That could be especially valuable in public health systems and humanitarian settings, where resources are finite and demand is high. It may also help smaller organizations that lack dedicated translation and adaptation teams but still want to offer interventions responsibly across multiple languages.

There is also a broader systems implication. Mental health inequity is often discussed in terms of clinician shortages, but informational infrastructure matters too. Treatment cannot scale if it exists only in one cultural register. Tools that reduce the cost of adaptation could widen the reach of therapies that already have evidence behind them.

The limits are as important as the promise

None of this removes the need for professional oversight. A text can feel culturally fluent and still miss clinical nuance. It can read naturally while introducing subtle bias or unsafe suggestions. That is why the study’s most credible takeaway is not automation, but augmentation.

The best near-term use of AI in this setting may be as a force multiplier for clinicians, researchers and service designers who already understand the treatment model and the target population. In that role, the technology could reduce delays, expand language coverage and improve the odds that patients receive materials that feel relevant rather than imported.

For refugee and migrant populations, that would be a meaningful change. Mental health access is often constrained long before a therapy session starts. If AI can help close the gap between a proven treatment and a patient’s lived context, it may offer one of the more practical and immediately useful applications of the technology in health care.

This article is based on reporting by Medical Xpress. Read the original article.