When AI replaces therapy: What are we gaining and losing

Lately, I have noticed a shift in the way people are seeking support. Not just clients, but family, friends, and strangers online. They’ll often say:

• “I don’t need to see anyone, AI figured it out for me.”
• “AI have me therapy techniques and I didn’t need to pay”

The rise in AI use for mental health support isn’t random because it probably is filling gaps in our current health system. Research shows that barriers like cost, stigma, long wait times, and accessibility prevent many people from seeking professional help in Australia (Kavanagh et al., 2023).

As a provisional psychologist I can see how it can help but I also realise it is not reflective of true therapy, and I think the difference matters more than people realise.

What we are gaining

AI is not inherently harmful. In fact, when used thoughtfully, it can support mental health in meaningful ways including:

1. Reflection and processing emotions

AI can help people organise and label their thoughts, feelings, behaviours and notice patterns that one misses sometimes. These are basic elements of therapies like Cognitive Behavioural Therapy (CBT).

2. Psychoeducation

When AI is accurate, people can learn about anxiety, depression, attachment styles, coping strategies, etc in a digestible way. That’s powerful, especially for those who’ve never had access to this knowledge before.

3. Between-session support

For people already in therapy, AI can act as a supplementary tool, helping reinforce skills, journal, or reflect between sessions.

4. Lowering the barrier to help-seeking

Sometimes AI becomes the first step or a way for someone to realise, “I might need more support than this.”

What we are losing

The concern is not that people are using AI but that some people are replacing it with professional mental help which can causes problems such as:

1. AI cannot truly assess risk

AI doesn’t have accountability or duty of care. It cannot intervene in crises the way a trained professional can. If someone is experiencing suicidal thoughts, self-harm or severe trauma, AI is not trained to safely manage that and cannot directly intervene.

2. It can reinforce but it cannot challenge

A good therapist doesn’t just validate you, but they challenge you especially in your patterns of thinking or behaviours like avoidance. AI can sometimes be agreeable and mirror what you’re already thinking, which reinforces unhelpful patterns, sometimes without you even knowing.

3. Lack of understanding

Therapy isn’t just about receiving what’s told to you, it is also important for you to understand and be aware of what is happening and describing it in your own words. AI can be a quick solution, giving you an answer immediately, but without you fully understanding the meaning and rationale behind it, which is not helpful long-term.

4. No therapeutic relationship

One of the strongest predictors of therapy outcomes is the therapeutic alliance, also known as the relationship between the client and their therapist (Flückiger et al., 2020). Being seen, understood, and connecting to another human being is not something AI can ever replicate.

5. Potential for misinformation or overconfidence

Even when AI provides helpful insights, it can sometimes present wrong information with confidence or misses important parts of the picture for you because at the end of the day, mental health is not a one-size-fits-all, treatment needs to be tailored.

So… what now?

AI is likely here to stay as part of our mental health landscape. The goal isn’t to reject it but to use it intentionally and thoughtfully.

You can think of AI as:

• A tool, not a therapist
• A starting point, not the whole journey
• A supplement, not a substitute

Because to be honest, at the end of the day, therapy is about connection and having someone to help you as a unique individual and only a human can do that.

Sarah Leung
Provisional Psychologist

References

Flückiger, C., Del Re, A. C., Wlodasch, D., Horvath, A. O., Solomonov, N., & Wampold, B. E. (2020). Assessing the alliance–outcome association adjusted for patient characteristics and treatment processes: A meta-analytic summary of direct comparisons. Journal of Counseling Psychology, 67(6), 706–711. https://doi.org/10.1037/cou0000424

Kavanagh, B. E., Corney, K. B., Beks, H., Williams, L. J., Quirk, S. E., & Versace, V. L. (2023). A scoping review of the barriers and facilitators to accessing and utilising mental health services across regional, rural, and remote Australia. BMC health services research, 23(1), 1060. https://doi.org/10.1186/s12913-023-10034-4