AI Can Give You an Answer. Therapy Helps You Understand Why You Needed One.

A lot of people do not start with a human therapist because they are avoiding growth. They start with AI because it feels safer than being seen…

One offers a soothing response; the other helps a person deepen, clarify, and stay with experience for longer than is immediately comfortable.

It is 11:43pm. You are in bed. The room is dark except for your phone. You have re-read the message three times.

“I think I need some space.”

You open ChatGPT and type:

“Am I overreacting?”“Is this relationship toxic?”“Why do I feel anxious all the time?”“Do I have trauma?”“Should I quit my job?”“Can you act like my therapist?”

None of that is strange anymore.

Plenty of thoughtful adults are now using AI for emotional support, relationship advice, anxiety, overthinking, low mood, loneliness, burnout, and decisions they feel oddly embarrassed to say out loud to another person. Research has caught up with that reality: consumers are increasingly turning to large language model chatbots for mental health conversations because they are easy to access and available when human care is harder to reach.

At the same time, recent reviews suggest that some chatbot interventions can produce small to moderate improvements in depressive and anxiety symptoms, particularly in structured or purpose-built tools rather than general-purpose chatbots. [1]

So this is not an anti-AI piece.

Sometimes AI is useful. Sometimes genuinely useful.

It can help you organise your thoughts, put language around a feeling, draft questions for therapy, summarise a messy situation, or even help you identify what you want to say before you finally say it properly.

But the deeper question is not only whether AI can give a quick or comforting answer.

The deeper question is whether being answered is the same as being therapeutically understood.

Usually, it is not.

And, I’m about to tell you why.

Why AI Can Feel So Therapeutic

AI can feel therapeutic for reasons that are completely understandable.

It is immediate. It is private. It does not flinch. It does not look tired. It does not make you book two weeks ahead or explain, out loud, why the same relationship keeps flattening you.

Recent reviews describe chatbot-based support as attractive partly because it offers round-the-clock access, reduced stigma, and an easier entry point for people who might otherwise avoid seeking help. Qualitative research on mental health chatbots also suggests that responsiveness, friendliness, and a non-judgmental tone can make people feel more willing to disclose emotionally sensitive material. [2]

That matters, especially if you are lonely, ashamed, burnt out, or unsure whether your problems are “serious enough” for therapy.

A lot of people do not start with a human therapist because they are avoiding growth. They start with AI because it feels safer than being seen.

To be fair… Some users do experience a kind of bond with mental health chatbots. Xu et al. (2025) explored what they called a digital therapeutic alliance and found that people can subjectively experience relationship-like processes with chatbots over time. Expectations around empathy, trust, conversational style, and fit all shaped whether the interaction felt supportive. In other words, the sense of connection can feel real, even when the other “party” is not a person. [3]

That is part of why this conversation needs nuance.

If something feels emotionally soothing, we tend to assume it is therapeutic.

Sometimes it is.

But, sometimes… It is just simply well-worded.

And these are not the same thing.

The Difference Between Advice and Therapy

A therapist is not simply a more expensive chatbot with a calendar.

Therapy is not the premium version of advice.

Therapy is an assessment process, a relational process, and an ethical process. It involves attunement, boundaries, informed consent, pacing, risk awareness, clinical judgement, and a conversation that unfolds over time rather than generating a polished answer on demand. In Australia, qualified counsellors and psychotherapists typically work within professional ethical frameworks and complaints processes designed to protect the public from harm and support safe practice. PACFA’s Code of Ethics explicitly frames counselling and psychotherapy around ethical standards, public protection, and complaints handling where those standards are not met. [4]

A good therapist is not only listening to what you say. They are also listening for what keeps happening.

They are listening for the moment your chest tightens when anger gets close. For the way every story about your partner becomes a story about your own worth. For the joke you make each time grief is about to arrive. For the way you speak beautifully about insight and then disappear the moment action is required. For whether your “clarity” is actually dissociation with a good vocabulary.

This is not advice-giving, but rather formulation.

It is also why validation alone is not the whole job.

Validation can be healing. It can reduce shame. It can help a person feel less ridiculous, less dramatic, less alone.

But constant validation is not the same as therapy.

In Scholich et al.’s 2025 comparison of licensed therapists and large language model chatbots, the chatbots used more affirming and reassuring language, along with more psychoeducation and suggestions, while therapists more often evoked elaboration. That difference is not trivial. It reflects two different tasks: one offers a soothing response; the other helps a person deepen, clarify, and stay with experience for longer than is immediately comfortable. [5]

Sometimes therapy validates.

Sometimes therapy slows you down.

Sometimes therapy says, very gently, “I can hear why that explanation feels true. I’m not convinced it is the whole truth.”

Sometimes therapy notices that the part of you asking the question is not actually looking for understanding. It is looking for certainty, rescue, permission, or a way to avoid feeling what comes next.

The most therapeutic response is not always the most immediately comforting one.

AI Responds. Therapy Relates.

This is the central contrast I keep coming back to.

AI responds.Therapy relates.

AI can validate.Therapy can validate, challenge, pace, repair, and hold accountability.

AI can generate language.Therapy helps you understand what is happening underneath the language.

The therapeutic relationship is not just a pleasant extra in counselling and psychotherapy. It is part of the treatment itself. Flückiger et al. (2018), in a meta-analytic synthesis spanning 295 independent studies and more than 30,000 patients, found that the therapeutic alliance was consistently associated with psychotherapy outcomes across treatment approaches, including internet-based psychotherapy. Saxler et al. (2024) similarly describe therapeutic alliance as a robust predictor of outcome in both face-to-face and online psychotherapy. That means the relationship is not simply the delivery system for therapy. It is one of the mechanisms through which therapy works. [6]

This matters if you are asking whether an AI therapist, a ChatGPT therapist prompt, or a mental health chatbot can replace therapy.

AI can simulate warmth, but simulation is not the same as mutual human relationship.

A chatbot does not actually know you. It does not carry responsibility for your care. It does not register the silence after you say, “I’m fine,” and realise you are not. It does not remember the shift in your face from three months ago when you touched something you were not ready to touch. It does not feel the subtle pull to rescue you, fear you, overjoin you, or misunderstand you and then repair that misunderstanding with you. It does not participate in the relational field. It outputs into it. Research on digital therapeutic alliance suggests that users can feel a sense of bond with chatbots, but that is still a different phenomenon from the accountable, mutual, clinically held therapeutic relationship between a client and a human therapist. [7]

What AI Cannot Reliably Track

Current AI mental health support tools can do some things surprisingly well. They can sound calm. They can offer structure. They can generate reflection prompts. They can mirror language back in a way that feels reassuringly coherent.

What they (AI) cannot yet do reliably is track the full complexity of a human being across time, context, embodiment, and risk.

That includes crisis risk. Self-harm risk. Suicidal thinking. Coercive control. Domestic abuse dynamics. Psychosis or delusional thinking. Eating disorder risk. Trauma responses. Dissociation. Manic states. Attachment patterns. Rapid changes in mood over time. Non-verbal cues. The emotional meaning of silence. The difference between insight and avoidance. The moment someone has begun intellectualising instead of feeling. The moment someone is using impressive self-awareness to avoid action. Those are not minor omissions. They sit right in the middle of real clinical work. [8]

Scholich et al. (2025) are especially useful here because they compared chatbot replies with those of licensed therapists rather than simply asking whether users liked the tool. Their conclusion was not that chatbots are useless. Their conclusion was that general-purpose chatbots were unsuitable as therapeutic agents, particularly in crisis situations, because they relied too heavily on generic validation, advice, and directive suggestions without enough inquiry. Here is a key distinction. Good therapy often asks before it answers. [5]

Clark’s 2025 simulation study adds another layer of concern. Across 60 scenarios involving distressed fictional adolescents, AI therapy and companion bots endorsed harmful or clearly unwise proposals in 19 cases, or 32% of opportunities. Four of the 10 bots endorsed half or more of the harmful ideas presented to them, and none opposed all of them.

That does not prove all therapy chatbots are unsafe across every situation. It does show that current systems can be worryingly inconsistent exactly where consistency matters most. [9]

Broader reviews are more balanced but still cautious. Sohn et al. (2026) found overall statistically significant benefits of chatbots for depressive and anxiety symptoms, but also noted substantial methodological heterogeneity and called for prespecified safety protocols, systematic reporting of adverse events, and longer-term follow-up. Zhang et al. (2025) found promising effects for generative AI mental health chatbots, yet emphasised wide prediction intervals, substantial between-study heterogeneity, and the need to view current findings as a starting point rather than definitive proof. Zhang et al. also explicitly argued that these tools should be positioned as supplementary rather than replacement treatments. Parks et al. (2025) make a related point from an evaluation perspective: if a tool is going to operate in mental health contexts, it needs to be assessed not only for sounding helpful but for safety, ethics, evidence basis, and appropriate safeguards. [10]

That is the issue in one sentence:

AI can sound therapeutic while still missing what therapy is actually doing.

When AI Might Help and When It Should Not Be Your Therapist

Used well, AI can still be useful.

Very useful, sometimes.

The best current use of AI mental health support is often as a supplement, not a substitute. Think of it as a reflective tool, not your primary clinician. That fits with the more cautious conclusions in current reviews: chatbots may serve as bridge interventions or adjuncts, especially for psychoeducation, self-guided exercises, between-session support, and improved access, but the evidence does not justify treating general-purpose chatbots as replacements for qualified therapists. [11]

Practical uses might sound like this:

“Help me organise what I want to bring to therapy.”“Give me journalling questions about this feeling.”“Help me separate the facts, feelings, and assumptions.”“Summarise this into three themes I can reflect on.”“Help me prepare a conversation with my therapist.”“Give me grounding exercises, and remind me to seek professional support if I feel unsafe.”

That kind of use makes sense. It is structured. It is bounded. It keeps the human at the centre.

There is also some evidence that generative chatbots may help with between-session cognitive behavioural tasks, self-reflection exercises, and other low-intensity mental health supports. Again, that is promising. It is not the same as saying a therapy chatbot can hold psychotherapy. [12]

Where AI should not be your therapist is clearer.

It should not be your main line of support for self-harm, suicidal thoughts, abuse, coercive control, psychosis, mania, severe depression, eating disorder symptoms, trauma flashbacks, dissociation, or situations where you feel yourself becoming dependent on the bot instead of reaching for real-world support.

Current studies raise enough concerns about crisis handling, harmful endorsement, inconsistent boundaries, and lack of evaluation standards that using AI as your primary mental health care in high-risk contexts is not a safe bet. [8]

If you or someone you know is in crisis in Australia, call 000 in an emergency. Healthdirect advises calling triple zero for immediate crisis situations, and directs people to Lifeline on 13 11 14 for 24/7 crisis support. The Australian Government also lists Beyond Blue on 1300 224 636 as a 24/7 support option. [13]

No shame. Just clarity.

There are times when convenience is good enough.

And there are times when you need care that can actually hold you.

The Real Question

By now, the internet is full of versions of the same headline: can AI replace therapy?

I think that is the wrong question.

The better question is:

What kind of support does this part of me actually need?

Sometimes you need information.Sometimes you need reflection.Sometimes you need a script for a hard conversation.Sometimes you need a calm place to put your thoughts at 11:43pm. And sometimes you need a regulated, trained, ethical human being who can sit with you long enough for the performance to drop and the truth to arrive.

That is what a human therapist is for.

Not because humans are magically superior at all things.

But because therapy is not just the transfer of helpful words. It is a clinically held relationship in which someone can assess risk, notice patterns, pace the work, challenge avoidance, track what changes over time, stay accountable to ethical standards, and help you understand why you needed the answer in the first place.

Current evidence does suggest that some AI tools can help with symptom reduction, emotional support, and access. It does not support the idea that general-purpose chatbots or AI companions can replace the relational, ethical, and clinically accountable work of psychotherapy. [14]

So here is the version I would keep:

AI can help you find words.Therapy can help you understand why those words are so hard to say.

AI can give you an answer.Therapy can help you build a different relationship with the part of you that keeps needing one.

If you have been using AI for anxiety, relationship stress, overthinking, emotional support, depression, loneliness, burnout, or decision-making, and you are starting to wonder whether you need something more personal, online counselling and psychotherapy can help. At Becoming On Point, I support adults navigating anxiety, pressure, relationships, self-doubt, burnout, and the quiet exhaustion of carrying everything alone. If you are looking for online therapy with a human therapist in Australia, you are welcome to book an online counselling or psychotherapy session and see whether this feels like the right fit.

FAQs

Can AI replace therapy?

AI can offer helpful language, reflection prompts, psychoeducation, and emotional support, but it is not the same as therapy. Therapy is not just receiving a thoughtful response. It is a relationship with a trained professional who can assess risk, notice patterns over time, pace the work, respond ethically, and help you understand what is happening underneath the problem you came in with.

Research on psychotherapy consistently shows that the therapeutic alliance — the relationship, trust, collaboration, and shared work between client and therapist — is strongly associated with therapy outcomes. That matters because therapy is not only about what is said. It is also about what happens between two people.

Is ChatGPT a therapist?

No. ChatGPT may sound therapeutic, but it is not a therapist. It does not hold clinical responsibility for you, diagnose you, maintain a therapeutic relationship, track your emotional patterns across sessions, or respond with the same duty of care as a qualified mental health professional.

That does not mean it is useless. It means it should be understood as a tool, not a replacement for counselling or psychotherapy. A chatbot can generate words that feel supportive. A therapist works with the person those words are trying to protect.

Why does AI feel like therapy?

AI can feel therapeutic because it is available, private, fast, articulate, and non-judgmental. You can ask it something vulnerable at midnight and receive an immediate response. For people who feel embarrassed, lonely, anxious, or unsure whether their problem is “serious enough,” that can feel relieving.

But feeling comforted is not always the same as being helped. A 2025 study comparing chatbot responses with licensed therapists found that chatbots used more affirmation, reassurance, psychoeducation, and suggestions, while therapists evoked more elaboration from the person. The researchers concluded that general-purpose chatbots were not suitable as therapeutic agents, especially in crisis situations.

Is it safe to use AI for mental health support?

It depends on how you use it. AI may be useful for low-risk support, such as journaling prompts, organising your thoughts before therapy, naming emotions, or preparing for a difficult conversation. A 2025 systematic review and meta-analysis found that generative AI mental health chatbots showed some promising effects in reducing mental health difficulties such as anxiety and depression, but that does not mean they should replace professional care.

AI becomes more risky when it is used as your main support for crisis, trauma, self-harm, abuse, severe depression, psychosis, eating disorder symptoms, or situations where you feel dependent on the chatbot.

What can a therapist do that AI cannot?

A therapist can notice not only what you say, but how you say it, what you avoid, what repeats, what changes, and what happens in the relationship between you. A therapist can assess risk, hold boundaries, challenge gently, repair misunderstandings, pace difficult work, and adapt support based on your history, nervous system, relationships, culture, and current safety.

AI can respond to the words you type. Therapy works with the whole person.

That difference matters. Sometimes the most helpful therapeutic moment is not a perfect answer. It is being met by another human being who can stay with you while you tell the truth.

When might AI be helpful between therapy sessions?

AI may be helpful as a bridge between therapy sessions, especially for simple reflective tasks. For example, you might ask it to help you organise what you want to bring to your therapist, turn a messy journal entry into themes, generate grounding exercises, or list the difference between facts, feelings, and assumptions.

A useful prompt might be:

“Help me organise this into three themes I could bring to therapy. Do not give me a diagnosis. Ask reflective questions instead of telling me what to do.”

The safest use of AI is often not “be my therapist.” It is “help me prepare to do deeper work with my therapist.”

When should I not use AI as my main support?

Do not use AI as your main support if you are feeling unsafe, having suicidal thoughts, wanting to hurt yourself or someone else, experiencing abuse, feeling detached from reality, struggling with severe depression, having trauma flashbacks, or feeling unable to stop using the chatbot for reassurance.

In those moments, you need real human support. Contact a mental health professional, crisis line, emergency service, trusted person, or local urgent support service. AI can produce caring words, but crisis care requires more than caring language. It requires responsibility, assessment, and real-world support.

Is online therapy better than AI therapy?

Online therapy and AI support are not the same thing. Online therapy still involves a real therapist, a therapeutic relationship, ethical responsibilities, clinical judgment, privacy standards, and ongoing care. AI may offer instant responses, but it does not provide the same relational or clinical structure.

For many people, online therapy gives the accessibility they like about digital support while still offering the human depth that AI cannot provide. You can speak from your own home, but you are not speaking into a system that only predicts the next sentence. You are speaking with a person who is listening for you.

How do I know if I need therapy instead of AI?

You may need therapy instead of AI if you keep asking the same questions and still feel stuck, if your distress is affecting your relationships or daily life, if you feel dependent on reassurance, or if you are dealing with trauma, grief, anxiety, depression, burnout, relationship stress, or emotional numbness.

A good question to ask is:

“Am I using AI to understand myself, or am I using it to avoid being known by another person?”

If the answer feels uncomfortable, that may be useful information. AI can help you find words. Therapy can help you understand why those words are hard to say.

References

Sohn, J. S., Ha, B.-G., et al. (2026). Systematic review and meta analysis of chatbots in the management of depressive and anxiety symptoms. npj Digital Medicine. [15]

Zhang, Q., Zhang, R., Xiong, Y., Sui, Y., Tong, C., & Lin, F.-H. (2025). Generative AI Mental Health Chatbots as Therapeutic Tools: Systematic Review and Meta-Analysis of Their Role in Reducing Mental Health Issues. Journal of Medical Internet Research. [16]

Scholich, T., Barr, M., Wiltsey Stirman, S., & Raj, S. (2025). A Comparison of Responses from Human Therapists and Large Language Model–Based Chatbots to Assess Therapeutic Communication: Mixed Methods Study. JMIR Mental Health. [17]

Xu, Z., Lee, J.-A., et al. (2025). The Digital Therapeutic Alliance With Mental Health Chatbots: Diary Study and Thematic Analysis. JMIR Mental Health. [18]

Clark, A. (2025). The Ability of AI Therapy Bots to Set Limits With Distressed Adolescents: Simulation-Based Comparison Study. JMIR Mental Health. [19]

Parks, A., et al. (2025). Is This Chatbot Safe and Evidence-Based? A Call for the Critical Evaluation of Generative AI Mental Health Chatbots. Journal of Participatory Medicine. [20]

Flückiger, C., Del Re, A. C., Wampold, B. E., & Horvath, A. O. (2018). The Alliance in Adult Psychotherapy: A Meta-Analytic Synthesis. [21]

Saxler, E., Schindler, T., Philipsen, A., Schulze, M., & Lux, S. (2024). Therapeutic alliance in individual adult psychotherapy: a systematic review of conceptualizations and measures for face-to-face- and online-psychotherapy. Frontiers in Psychology. [22]

PACFA. Code of Ethics. Psychotherapy and Counselling Federation of Australia. [4]

Healthdirect Australia. Mental health helplines. Australian Government. [23]

Australian Government Department of Health, Disability and Ageing. What we’re doing about mental health. [24]

Lifeline Australia. 24 hour crisis support and suicide prevention services. [2