Short answer: “AI therapy” is an umbrella term, not one kind of treatment. It can mean a general-purpose chatbot used for emotional support, a chatbot designed for mental-health or wellness tasks, or a digital service that performs screening, self-management or therapy-related functions. These products differ greatly in evidence, privacy, human supervision and crisis support.
Current professional guidance does not support presenting a general-purpose chatbot as a replacement for psychological treatment. Used cautiously, AI may help you organise questions for a clinician, learn basic psychoeducation or rehearse an exercise you have already been taught. It should not diagnose you, manage an acute crisis or become your only source of mental-health care.
As an Amazon Associate I earn from qualifying purchases.
What does “AI therapy” mean?
The phrase covers several technically and clinically different products. Treating them as interchangeable makes it impossible to judge whether a service is suitable for a particular need.
General-purpose generative chatbots
These systems were built for broad conversation, writing or information tasks rather than mental-health care. A person may nevertheless ask one about anxiety, depression, relationships or suicidal thoughts. The American Psychological Association (APA) describes this as a mismatch between product design and real-world use: a chatbot can produce fluent, empathetic-sounding text without having the clinical context, risk-assessment process or cultural competence expected in professional care.
Mental-health or emotional-wellness chatbots
These tools are designed around a narrower purpose, such as psychoeducation, mood check-ins, structured coping exercises or guided conversations. “Purpose-built” does not automatically mean clinically effective. You still need to ask what population was studied, which outcomes were measured, whether a clinician supervised the program and what happens when a user reports imminent danger.
Digital mental-health tools and wellness apps
A broader digital service may screen for symptoms, monitor treatment, deliver a structured intervention or support communication with a clinician. Other apps are intended only for daily wellness, reminders or symptom tracking. An app that records a mood is not necessarily providing therapy, and an app that presents information is not necessarily evaluating your condition.
How do AI systems used around therapy work?
Earlier mental-health chatbots commonly followed rule-based scripts: a particular phrase or answer led to a predefined response. Machine-learning systems added statistical classification, for example identifying likely topics or sentiment from text. Large language models (LLMs) generate responses from patterns learned during training and can maintain a more flexible conversation, but fluency is not proof that a statement is clinically correct.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallA service can combine these approaches. It might ask a standard safety question, classify a message, retrieve educational material and then generate a conversational reply. The important practical distinction is what the system is authorised and tested to do. A tool that offers a fixed breathing exercise has a narrower risk profile than one that improvises advice about medication, diagnosis or self-harm.
What does the current evidence show?
A systematic review by Hua and colleagues in World Psychiatry (2025) examined 160 studies published from 2020 through 2024. The review separates three evaluation stages:
| Evaluation stage | What it asks | Why it matters |
|---|---|---|
| Technical validation | Does the system perform its intended computational task? | A technically functional system can still give unsafe or clinically useless advice. |
| Feasibility and engagement | Can people use it, and do they continue using it? | Usage shows acceptability, not that symptoms or functioning improve. |
| Clinical efficacy | Does a defined intervention improve relevant outcomes in a studied population and design? | This is the evidence needed for a treatment claim, with the study’s limits kept in view. |
LLM-based chatbots accounted for 45% of new studies in 2024. Yet only 16% of the LLM studies in the review underwent clinical efficacy testing, while 77% remained in early validation. Across all chatbot studies in the review, 47% focused on clinical efficacy testing. These are characteristics of the research literature, not market share, guaranteed effectiveness or proof about any particular product.
Technical novelty and high engagement therefore need to be separated from demonstrated clinical benefit. A promising prototype, a popular consumer app and a clinician-supervised treatment program may all be called “AI therapy” while having very different evidence.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsCan AI replace a therapist?
For general-purpose generative chatbots, the APA’s professional guidance says no. Its discussion of AI use in therapy states: “Despite the appeal, APA does not recommend using general purpose generative AI chatbots for psychological treatment.” That position reflects the absence of reliable clinical context, the difficulty of assessing risk from a text exchange and uncertainty about whether responses are appropriate for a person’s history, culture and goals.
A licensed therapist does more than generate supportive sentences. Treatment involves assessment, a jointly developed formulation, informed consent, ongoing monitoring, adaptation to changing circumstances and responsibility for clinical decisions. A chatbot cannot provide that professional relationship or assume a clinician’s legal and ethical duties.
AI can still have a bounded role alongside care. APA describes uses such as learning general information about a mental-health issue, tracking moods or thoughts between appointments for discussion with a therapist, and practising a cognitive-behavioural exercise already taught in treatment. The safe boundary is that the clinician remains the source of assessment and treatment decisions.
Is AI therapy safe in a crisis?
Do not rely on a chatbot for imminent danger, suicidal intent, self-harm, overdose, abuse or another emergency. In an APA-reported study, six popular chatbots were tested against simulated mental-health crises and none met clinical standards for crisis response. Those findings apply to the systems and scenarios tested at that time; they should not be treated as a permanent result for every tool or future version. They do show why a polished conversational response is not a dependable crisis pathway.
Free tools Windows power users keep installed
One-click scans. No signup required.
If you or someone else may be in immediate danger, stop the chatbot conversation and contact local emergency services, a local crisis line or a qualified mental-health professional. Ask a trusted person to stay with you when possible. Crisis options differ by country, so use services available where you are rather than assuming a single international number.
Rank #4
Is an AI chatbot private?
Not automatically. The APA consumer guide warns that information shared with AI is not kept private and says, “AI should not be your only resource to relieve mental health symptoms.” A service may store prompts, associate them with an account, use human reviewers, retain them for a stated period or use data for product improvement. The exact practices depend on the provider and account type.
Before entering a disclosure, check the service’s current policy for:
- what text, files, voice recordings or metadata are collected;
- how long data are retained and whether you can delete them;
- whether conversations are used for training, analytics or advertising;
- which contractors or reviewers can access content;
- where data are stored and which law governs them; and
- whether you can turn off history, model training or other secondary uses.
Use the minimum information needed for a limited task. You can ask for help turning “I have panic attacks after a workplace incident” into questions for a clinician without supplying names, addresses, dates of birth, medical-record numbers or detailed trauma narratives. Never assume that a consumer chatbot has the confidentiality protections of a licensed therapist or a health-care provider.
What can AI reasonably help with?
If you choose to use an AI system, keep the task narrow and verifiable.
Preparing for an appointment
- Write your own symptoms, questions and goals first.
- Ask the tool to organise them into a short agenda, without identifying details.
- Check every suggestion and remove anything inaccurate or uncomfortable.
- Bring the list to your clinician and disclose that AI helped organise it.
Rehearsing a skill you already learned
You might ask for a reminder of the steps in a breathing, grounding or thought-record exercise taught by your therapist. Compare the output with your clinician’s instructions. Do not let a generated variation replace the treatment plan, especially if an exercise worsens symptoms.
Tracking information for discussion
A simple mood or sleep log can help you notice patterns to discuss with a professional. Keep the record in a tool whose storage and export settings you understand. Tracking is observation, not diagnosis; a change in a score does not establish a disorder or tell you which treatment to choose.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How should you evaluate a mental-health AI tool?
Use the following questions before creating an account or sharing sensitive information.
| Decision area | Questions to ask |
|---|---|
| Intended purpose | Is it general conversation, wellness, psychoeducation, screening, symptom monitoring or treatment? |
| Evidence | Does the provider report technical validation, feasibility data or clinical efficacy? Which population, outcome and study design were used? |
| Human involvement | Is it self-guided, supervised by a clinician or connected directly to a licensed professional? |
| Crisis pathway | What happens after a disclosure of self-harm or immediate danger? Is there a qualified human hand-off, and is it available in your location? |
| Privacy | What is collected, retained, reviewed, shared or used for secondary purposes, and which controls can you change? |
| Oversight | Which country’s regulator or professional standards apply, and what exactly has been assessed? |
Be sceptical of claims based only on a large user count, an engaging interface, a “therapist” persona or a single pilot study. Look for a clearly defined intervention, comparator, outcome measure, follow-up period and description of harms or dropouts.
What does regulation cover?
Regulatory status depends on the product’s intended purpose and the jurisdiction. Australia’s Therapeutic Goods Administration (TGA) says, “Some digital mental health tools (DMHT) are regulated as medical devices.” Its guidance, last updated 11 February 2026, indicates that software intended to diagnose or screen for a condition, monitor symptoms or treatment, or provide a therapeutic intervention such as cognitive-behavioural therapy may meet the medical-device definition.
The same guidance distinguishes software limited to non-clinical functions, such as displaying information, scheduling, symptom tracking without analysis or facilitating telehealth. Those functions are not regulated by the TGA as medical devices under that guidance, and the TGA says excluded or non-device products have not been assessed by it for quality, safety or performance. This classification is Australia-specific; it should not be transferred automatically to the United States, European Union, United Kingdom or another market.
A practical decision rule
- Define the job. If the job is diagnosis, medication advice, treatment selection or crisis management, use a qualified professional instead of a general chatbot.
- Check the evidence and supervision. Prefer a service that states its target population, outcomes, safety process and clinician involvement.
- Read privacy controls first. Do not disclose sensitive health information until you understand retention, access and deletion.
- Keep a human in the loop. Tell your therapist or doctor that you are using the tool and bring questionable outputs to them.
- Stop when advice is unsafe. APA advises stepping away from guidance that points toward harm and consulting trusted people and qualified providers.
AI can be a limited organisational or practice aid. It is not a licensed professional, a reliable diagnostic instrument or a substitute for a therapeutic relationship, and it should never be the only mental-health resource.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
