By
Stuart Kerr, Technology Correspondent,
LiveAIWire

One in four adults in high-income countries will experience a
trauma-related mental health condition in their lifetime, according to the
World Health Organisation, yet the global shortage of mental health
practitioners means that most will not receive evidence-based treatment. The
gap is not a matter of willingness. It is a structural failure of capacity,
geography, and stigma that the current system cannot resolve at the pace that
the scale of unmet need demands. AI-assisted therapy tools are being
positioned as part of the answer, and the evidence base for some applications
is more substantial than the scepticism they attract from clinical
professionals might suggest.

Natural language processing tools can now detect markers of
depression, anxiety, and post-traumatic stress in speech patterns, syntax,
and rhythm with accuracy that in controlled studies has exceeded the
performance of general practitioners making initial assessments. These
diagnostic applications do not treat trauma. They identify it earlier and
more consistently than clinical systems that depend on patients initiating
contact, accurately describing their own symptoms, and being seen by a
professional with the training and time to make a reliable assessment.
Earlier identification translates directly into earlier intervention, and
earlier intervention in trauma is associated with significantly better
outcomes across multiple conditions. The World
Health Organisation’s mental health fact sheets document the scale
of the treatment gap that AI tools are being positioned to help
address.

What AI-Assisted Therapy Actually Looks Like

The therapeutic applications receiving the most clinical attention
fall into distinct categories with different levels of evidence behind them.
Cognitive behavioural therapy chatbots, including Woebot and Wysa, deliver
structured CBT exercises through conversational interfaces. Published
randomised controlled trial evidence shows modest but statistically
significant reductions in depression and anxiety symptoms compared to control
conditions. These are not chatbots simulating human empathy. They are
structured protocol delivery tools that make evidence-based exercises
available on demand, without appointment barriers or the stigma that prevents
many people from initiating contact with clinical services.

VR-based exposure therapy, in which AI-generated environments
allow individuals with PTSD to engage with trauma-related stimuli under
controlled conditions with therapeutic guidance, has stronger clinical
evidence than conversational interfaces across several documented
applications. A systematic
review published in the Journal of Psychiatric Research found that
VR exposure therapy produced significant reductions in PTSD symptom severity
in military veteran populations, with effect sizes comparable to traditional
in-person exposure therapy. The advantage over in-person exposure is control:
the therapeutic environment can be calibrated precisely, exposure can be
graduated more finely than real-world contexts permit, and the patient can
disengage immediately if distress exceeds a manageable
threshold.

What This Means for You

If you are considering AI-assisted therapy tools for yourself or a
family member, the most important distinction is between tools with published
clinical evidence and those without it. The digital mental health market
includes hundreds of apps making therapeutic claims, most of which have no
published efficacy data and some of which are actively harmful through their
design choices or the populations they are marketed to. The apps with
randomised controlled trial evidence, Woebot and Wysa among them for CBT
delivery, are a small subset of the market.

The appropriate use case for AI-assisted tools is as a complement
to professional care rather than a replacement for it. For individuals who
cannot access professional support because of cost, geography, waiting times,
or stigma, an evidence-based AI tool providing structured CBT exercises is
meaningfully better than no support. For individuals in acute crisis, AI
tools are not a substitute for emergency clinical intervention, and the most
responsible apps in the market are explicit about this limitation and provide
crisis resources prominently. If you are in crisis, contact a mental health
helpline rather than relying on an AI application.

The Governance Gap

The clinical evidence base for specific applications is stronger
than the governance framework around the broader market. Regulatory oversight
of digital mental health tools varies substantially by jurisdiction. In the
UK, the National Institute for Health and Care Excellence has developed
assessment frameworks for digital mental health interventions, but the pace
of market entry substantially exceeds the pace of assessment. In the United
States, the FDA’s regulatory approach to software-as-a-medical-device applies
to some therapeutic applications but not to the broader category of wellness
and self-help tools that most AI therapy products are classified
under.

The governance gap creates a market in which well-designed,
evidence-based tools compete with poorly designed apps making similar claims
without similar evidence, and in which patients have limited information to
distinguish between them. This is not a problem unique to AI mental health
tools, but the intimacy of the domain and the vulnerability of the population
make the consequences of poor quality tools more serious than in most other
areas of the digital health market.

What Clinical Professionals Think

Clinical scepticism toward AI therapy tools is partly justified by
the governance failure described above and partly reflects professional
concerns that extend beyond evidence quality. Therapists and psychiatrists
who have worked with trauma patients consistently raise concerns about the
relational dimension of effective trauma therapy: that healing from trauma
often depends specifically on the experience of a trustworthy human
relationship rather than on the delivery of therapeutic content through any
particular channel. This concern is empirically grounded. The therapeutic
alliance, the quality of the relationship between therapist and patient, is
one of the strongest predictors of therapy outcomes across modalities, and it
is not clear that AI tools can replicate this.

The strongest clinical case for AI-assisted therapy is therefore
in specific applications where the relational dimension is less central than
the delivery of structured content: psychoeducation, symptom monitoring, guided
relaxation, exposure hierarchy development. These are components of effective
trauma treatment that AI can deliver competently, freeing clinical time for
the relational and judgement-intensive components that AI cannot substitute.
For related coverage, see our look at whether
AI therapy actually works, our coverage of AI
detecting depression before doctors can, and the broader picture of
AI
in personal health monitoring.

The Regulatory Landscape

The governance gap in digital mental health is most visible in the
comparison between the pace of market entry and the pace of regulatory
assessment. In the UK, NICE has developed assessment frameworks for digital
mental health interventions, but the volume of new products entering the
market substantially exceeds NICE’s assessment capacity, meaning that most
products available to UK consumers carry no independent quality assessment.
The NHS App Library includes only products that have passed a quality
assessment, but it represents a small fraction of the apps available through
consumer app stores.

The regulatory challenge is partly definitional. General wellness
apps that help users manage stress or improve sleep sit outside most medical
device regulatory frameworks. Apps that claim to treat specific mental health
conditions fall within those frameworks in principle, but the boundaries are
contested and enforcement is inconsistent. This definitional ambiguity allows
the market to segment into a well-regulated and assessed tier of clinical
tools and a largely unregulated tier of wellness apps, with consumers having
limited ability to distinguish between them.

Closing this gap requires regulatory frameworks that extend
meaningful quality assessment to the wellness tier of digital mental health
tools rather than limiting oversight to clinical-grade applications. This is
a technically demanding regulatory challenge because the volume of products
is large, the evidence standards appropriate for wellness tools are genuinely
different from those for medical devices, and the pace of product iteration
means that assessments can become outdated quickly. But the alternative,
leaving millions of people to navigate a market of variable quality tools
without meaningful independent guidance, is a governance failure with real
consequences for mental health outcomes.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering
artificial intelligence, cybersecurity, and the social impact of emerging
technology. He publishes daily at LiveAIWire.com.