+370 628 53379psichologe.iveta@gmail.com
A woman pauses at her laptop and looks through the window at home

Can artificial intelligence replace a psychologist?

Psychology and technologyEnglish edition updated

It is late at night, an unpleasant conversation keeps replaying in your mind, and you hesitate to message someone close to you. An AI chatbot replies immediately: calmly, at length and without any apparent tiredness. It is understandable that this availability can be appealing. But can artificial intelligence replace a psychologist? There is currently no basis for treating an ordinary AI conversation as equivalent to professional support. It may help you organise your thoughts, but a pleasant answer is not an assessment of your situation.

Why can it feel easier to open up to a screen?

When writing, you can pause, delete a sentence and start again. You do not have to watch another person's face or worry about an awkward silence. When it is difficult to admit that you feel exhausted, envious or lonely, this distance can make a first step feel less daunting. Trying to put an experience into words can sometimes help you notice what really hurts. That benefit need not be dismissed simply because the other participant is a program.

In the APA's 2026 survey, US psychologists reported noticing both the support their patients felt and concerning cases of attachment to chatbots. The survey involved 1,242 professionals. These were their reports about people they worked with, not a study of all AI users; the survey does not show that AI caused the difficulties described. Nor can these findings be directly applied to people living in Lithuania.

A useful personal question might be: what does being able to write right now give me? Perhaps you are looking for words, reassurance that your feelings make sense, or a way to avoid a conversation you dread. The answers may differ even within the same week. They help you understand the purpose of your use without labelling yourself.

What do studies of specialised tools demonstrate?

The broad term AI covers different tools. Some handle a wide range of questions; others are developed for a specific mental health intervention, with professionals involved in training and safety assessment. The label ‘therapeutic’ alone establishes neither that involvement nor effectiveness. Look for research on the particular tool and its intended purpose rather than relying on a general promise.

The 2025 randomised trial of Therabot involved 210 adults in the US with significant symptoms of depression or anxiety, or a high risk of eating disorders. Symptom scores improved more in the group using the specialised tool than in the waiting-list group. Follow-up lasted up to eight weeks. This is encouraging, but the control group did not receive a comparable intervention from a psychologist, so the study does not demonstrate equivalence to consultations. Its findings do not validate all general-purpose chatbots or long-term use.

How does a chat differ from a consultation?

The difference concerns the process of support, responsibility and opportunities to resolve misunderstandings. In a consultation, you can explore together why a suggestion does not suit you, what silence might mean or how difficulties change over time. A psychologist is not infallible either: what matters is being able to discuss and refine conclusions and seek other support when needed.

A general-purpose AI conversation and a psychologist's consultation
AreaAI chatbotPsychologist
Basis of the responseGenerates text from learned patterns and the information supplied.Applies professional knowledge and considers the individual context and how the consultation develops.
SupportMay offer encouragement, questions or a structure for reflection.Combines support with joint exploration and discussion of goals and change.
Boundaries and responsibilityThe provider defines the service's purpose; a response is not a professional commitment.Explains the limits of their competence, work and confidentiality.
Information storageDepends on the platform's data handling terms and settings.Information is stored in accordance with professional and applicable data protection requirements.
Support when wellbeing deterioratesCannot be relied on to recognise a crisis or arrange help reliably.Assesses support needs and may make referrals; a routine consultation is not an emergency service.

Choosing a psychologist and the relationship between you are explored in more detail in how to tell if a psychologist is right for you. The key point here is not to mistake fluent language for professional expertise: the World Health Organization warns about automation bias, where a technology's suggestion is trusted so much that errors go unnoticed.

When agreement starts a repeating cycle

Imagine asking, after a conflict, whether a colleague treated you unfairly. Agreement brings brief relief, but soon doubt returns and you add another detail. The conversation grows longer while you feel increasingly reluctant to speak to your colleague. This illustrative situation invites you to notice whether the exchange brings greater clarity and freedom to act, rather than counting the answers you receive.

In experiments reported in a 2026 study published in Science, overly agreeable models strengthened participants' belief that they were right and reduced their stated willingness to apologise or repair a conflict. These are findings from particular experiments, not proof that every AI conversation harms relationships. Still, they remind us that acknowledging a feeling and confirming one interpretation of events are different things.

You can ask yourself whether you are seeking a fresh perspective or another reassurance. If answers only briefly quieten doubt, it may help to return to the facts: what actually happened, what you are assuming and what you can discuss directly. For more on repeatedly analysing your thoughts, read when thoughts go round in circles. Asking AI to suggest alternatives does not guarantee a more accurate assessment.

Feeling private does not mean your data is private

Talking in an empty room can make it feel as though the conversation is yours alone. But your text is sent to the service provider, and how it is stored, used and deleted depends on the specific terms. Do not assume that every platform handles it in the same way, or equate an app conversation with the confidentiality of a consultation. The APA recommends checking privacy policies and settings and limiting the sensitive or personally identifying information you share.

In practice, this means leaving out surnames, addresses and personal identification numbers, and not uploading medical records or identifiable conversations with people close to you. A general description is often enough for reflection: ‘I find it hard to settle after an argument.’ If you have already shared more than you intended, you can review that service's deletion and data-use options. Removing a conversation from the visible list alone does not necessarily explain everything about how the data is retained.

Two women on a bench in a green garden talk quietly and listen to each other
Support can begin with a simple conversation with someone you trust.

A place for a tool that helps you prepare

If you choose to use AI, you can give it a small, clearly defined task: helping you list questions for a consultation or organising non-personal notes by theme. For example, when preparing for a session, separate when the difficulties began, how they interfere with life and what you hope support will offer. Read the final summary yourself: does it really convey your experience, or has the program added conclusions you never expressed?

The APA's guidance suggests treating AI as a limited supplementary tool, checking health information with a professional and preventing its use from interfering with sleep or relationships. One simple way to assess its value is to ask: when you close the window, is your next step clearer? If you stay at the screen seeking fresh reassurance, the task has already changed. Before starting, you could decide what you will do afterwards: note one question, talk to someone close to you or turn to an everyday activity. This is not a treatment method or a strict rule for use. It is a way to notice whether the tool helps you return to life beyond the screen or repeatedly delays that return.

Enjoying writing does not mean you have to choose an automated response. Written communication with a person is discussed in the article psychological support by email. What matters is who replies and what boundaries of support you have agreed, not simply whether the words appear on a screen.

When should you turn to a person?

Do not wait for a chatbot to decide that you are ‘struggling enough’. If anxiety, sadness or confusion interferes with everyday life, you are withdrawing from people, or AI conversations increasingly lead you to postpone help, that is reason enough to discuss how you feel. You could begin with: ‘I notice I talk to a program a lot, but I am still struggling.’ You do not need to know a diagnosis in advance or explain the cause precisely.

A 2025 study of language models found inappropriate responses to some dangerous mental health scenarios. Such testing does not establish the likelihood of harm to an individual, but it gives reason not to rely on an automated response in a crisis. If your life or someone else's is in immediate danger in Lithuania, call 112 or go to an emergency department; if possible, ask someone you trust to stay with you. Outside Lithuania, contact your local emergency service.

When there is no immediate danger, an online consultation with a psychologist can offer a way to discuss what is happening at your own pace and explore a suitable direction for support together. You can come to a session with unfinished thoughts. You do not need to ‘sort them out’ with AI first.

Common questions about AI and psychological support

Is it normal to feel heard after talking to AI?

Yes. An attentive-sounding response and the chance to express your thoughts can bring relief. That feeling deserves respect, but it does not confirm that the system accurately understood your situation or provided an appropriate psychological assessment.

Did the Therabot study prove that AI is as effective as a psychologist?

No. The 2025 study compared a specialised tool with a waiting-list group, not with consultations led by a psychologist. Improved symptom scores are promising, but they do not establish equivalence to human support or demonstrate the effectiveness of other chatbots.

What should I do if AI suggests a diagnosis based on what I tell it?

Treat it as an unverified assumption. Note what you have been experiencing, how long it has lasted and how it affects everyday life, then seek an assessment from an appropriate professional. Do not change medication or treatment based solely on a generated response.

Can I upload my partner's messages so AI can resolve a conflict?

It is better not to upload identifiable private conversations. They contain another person's data, and individual messages do not reveal the whole relationship. If you want to reflect on your reaction, describe in your own words what you felt and what you would like to discuss, without names or identifying details.

How can I tell whether I am repeatedly asking AI for reassurance?

Notice whether you return to the same question, seek increasingly firm confirmation and feel only brief relief after each answer. If this makes you put off sleep, decisions or a conversation with a person, it is worth pausing and discussing the pattern with a psychologist.

Does asking AI to disagree with me remove the risk?

No. Asking for another perspective may broaden an answer, but it does not guarantee accuracy, sensitivity or safety. A generated objection can also be wrong. Check important decisions against the facts and discuss them with someone you trust.

Should I tell my psychologist that I use AI for emotional support?

Yes. You can explain when you turn to it, what helps and what worries you. You do not need to provide your full conversation history. One response that mattered to you, or a summary of it, may be enough to discuss how this use fits with the goals of your consultations.

Does written support always mean talking to a bot?

No. In email consultations, a person responds, and you can agree on the boundaries, goals and confidentiality of the work. A written format alone does not tell you who provides the support, so it is important to check the service description and its provider.

Sources and further reading

Sources reviewed for the original article on 7 September 2026. Research findings apply to the tools and participants assessed in those studies.

  1. APA (2026). Survey of US psychologists about patients' use of AI chatbots.
  2. Heinz et al. (2025). Randomized Trial of a Generative AI Chatbot for Mental Health Treatment. NEJM AI.
  3. Stanford University (2026). Overview of research on overly agreeable AI published in Science.
  4. APA (2025). Advisory on AI chatbots and wellness apps.
  5. APA (2026). Guidance on assessing AI advice critically and more safely.
  6. WHO (2024). Ethics and risks of using large multimodal models in health care.
  7. Moore et al. (2025). Study of stigmatising and inappropriate responses from language models.