psAico
EN

An AI assistant for psychological wellbeing, built on evidence

A review of the research on artificial intelligence for mental health: what the studies actually measured, and how psAico takes it into account.

Large language models can be instructed to help you explore how you feel, with psychoeducation, self-help strategies and practical tools drawn from CBT (cognitive behavioural therapy). Researchers have been studying this use of AI for several years, and psAico is built on what they have found.

Italy’s National Council of the Order of Psychologists, in a post published on 15 May 2025, made the point well: artificial intelligence “cannot replace the human psychologist”, but it can “promote wellbeing, collect and monitor data, share information”. That is exactly the spirit of psAico. It offers psychological help, not psychotherapy: it builds awareness, explains cognitive behavioural techniques and strategies that support wellbeing, and brings people closer to psychology and psychotherapy instead of standing in for them. It can also sit alongside therapy you are already having, for example by helping with what CBT calls homework, the exercises a therapist sets between sessions. So it can be useful both if you are not seeing anyone and if you are already in treatment. In that second case, it is always worth telling your therapist that you are using it.

On 3 July 2026, at the Italian Chamber of Deputies, the same National Council presented the first national survey on psychologists and generative AI, run with 5,906 professionals: 58.76% already use AI tools in their work, 55.04% say their own patients turn to AI for emotional matters or to look up diagnostic information, and 86% are asking for training and supervision on using it well. It is a snapshot of something that is already happening: the question is no longer whether people will use artificial intelligence to talk about themselves, but with what tools and what safeguards.

To read more about the cognitive behavioural approach, the NHS explains what CBT is, how sessions work and how to refer yourself to NHS Talking Therapies at this link.

What the research says so far

Research on AI assistants for psychological wellbeing is young and moving fast. Here is what the studies have measured, with the numbers and the sources: it is the clearest way to show what psAico rests on.

Effectiveness

Real effects, measured in thousands of people

A systematic review with meta-analysis published in npj Digital Medicine brought together 39 randomised controlled trials of chatbots for mental health. Across more than 7,400 participants for depression and 7,600 for anxiety, the chatbots produced statistically significant reductions in symptoms compared with control groups.

A second meta-analysis, published in May 2026 in the Journal of Medical Internet Research, isolated the 29 randomised trials of chatbots built explicitly on cognitive behavioural therapy, the approach psAico uses: a moderate reduction in depressive symptoms and a small reduction in anxiety symptoms, with the clearest benefit for people carrying more than one difficulty at once.

No digital tool replaces psychotherapy, and psAico does not claim to. But the effects are real and replicated, and they are clearer in people who start from more intense distress.

Therapeutic alliance

The first randomised trial of generative AI in therapy

In 2025 a team at Dartmouth College published in NEJM AI the first randomised controlled trial of a therapy chatbot built on generative AI. 210 adults with major depression, generalised anxiety disorder or a high risk of an eating disorder used it for four weeks: average symptom reductions of 51% for depression, 31% for anxiety and 19% for eating disorder concerns, still there at eight weeks.

The most striking figure is not a clinical one. Participants reported a therapeutic alliance comparable to outpatient psychotherapy, and they genuinely used the tool: on average 260 messages and more than six hours over four weeks.

We report it with its limits: the comparison was a waiting list rather than an active treatment, and on fifteen occasions a member of staff had to step in when someone expressed suicidal thoughts. The authors say it themselves: without clinical oversight these tools should not be left on their own. That is why psAico follows strict safety protocols that point people towards human professionals, or towards the emergency services when that is what is needed.

Staying with it

A tool people actually keep using

In a randomised controlled trial with 540 adults with symptoms of anxiety or depression, a cognitive behavioural therapy app built on generative AI was compared over six weeks with digital workbooks. The people using the app came back to it 2.4 times more often and stayed 3.8 times longer, with the same clinical results and the same safety profile.

This is the figure that matters most to us: a tool you drop after three days helps nobody. Analyses from the same trial also found that people who used the personalisation features saw a larger reduction in anxiety and better overall wellbeing, which is exactly the direction psAico is designed for.

A June 2026 meta-analysis of 39 studies points the same way from another angle: tools built on generative AI are abandoned far less often than scripted ones (9.25% against 23.12%), and less drop-out means potentially more people helped.

Design

Why a clinical project matters, not just any chatbot

A systematic review published in JMIR AI looked at 20 psychological counselling chatbots built on language models. The most instructive finding is not how well they hold a conversation, but how few of them say how they were validated, what safeguards they have in place and how they handle the safety of the people using them.

Nature Medicine showed in May 2026 how much the project matters more than the model: a language model with a layer of clinical reasoning on top, assessed blind by 22 expert clinicians, outperformed both state-of-the-art general models and human clinicians on the core competencies of cognitive behavioural therapy; and across 19,674 real conversations, the more that clinical layer was engaged, the greater the improvement in symptoms. The study is authored by the company that built the system, and we say so, but the direction is the one we follow too: not the bare model, the project around it.

That gap is exactly where psAico starts: created and trained by cognitive behavioural psychologists and psychotherapists, with a stated limit on use, clear guidance for emergencies, a minimum age and a privacy policy written for a psychological setting.

Safety

Risks can be measured, and designed for

In August 2026 Nature Medicine published a way of testing how general-purpose chatbots behave when the person writing is vulnerable: 810 conversations, 30 simulated user profiles, 13 clinically grounded risk dimensions. Concerning behaviour was widespread, less so in newer models, and above all it builds up turn after turn: risk peaks when apparently supportive replies end up reinforcing the very mechanism that is causing the person distress.

A review published in Global Mental Health in June 2026 reaches the same conclusion on the most delicate point: of 29 commercial chatbots put to the test, none responded adequately to a suicidal crisis, while a clinically designed, method-anchored chatbot proved effective across every domain assessed.

That is why psAico has explicit guidance for emergencies, instructions written by psychotherapists, a minimum age and automatic suspension of an account when a user turns out to be under age: not red tape, but an answer to risks somebody has measured.

Stated limits

How large the effect is, and why we set a limit

The most severe meta-analysis published so far brought together 22 randomised trials and more than 110,000 people across commercial chatbots taken as a whole: for depression the average improvement is about 1.6 points on the PHQ-9 questionnaire, below the 5-point threshold considered clinically meaningful, and for anxiety the picture is not settled. The authors conclude that these tools sit well as low-intensity adjuncts to psychotherapy, with safety designed in from the start.

That is exactly the place psAico has given itself, and it is why you will find no promises of a cure on this page: what makes the difference is how the tool is built, and what it sits alongside.

The same holds for time. A longitudinal study of more than 2,000 adults in four countries, followed for twelve months, found that heavy use of companion chatbots can go together with more loneliness, not less. That is why psAico has a stated limit on use and points people back to people: to those close to them, to professionals, to their own therapist.

“The results indicate that the development of AI in understanding emotions and social behavior related to social intelligence is very rapid. AI will help the psychotherapist a great deal in new ways. The psychotherapist needs to be aware of possible areas of further development of AI given their benefits in counseling and psychotherapy.”

Sufyan NS, Fadhel FH, Alkhathami SS, Mukhadi JYA - Artificial intelligence and social intelligence: preliminary comparison study between AI models and psychologists, Frontiers in Psychology, 2 February 2024.

In this 2024 study, ChatGPT-4 scored higher on a social intelligence scale than every one of the human participants: a striking result, and a measure of how quickly these models are learning to recognise emotions and social behaviour.

We report it precisely, because rigour is part of the project: it was a preliminary study, the comparison involved 180 counselling psychology students at a single university, and it measured social intelligence, not the effectiveness of a treatment. It is a strong signal about the direction of travel; the evidence on effectiveness is in the research above.

We are constantly researching and developing, and the direction is clear: the future of psychotherapy, like the future of most of what we do, will involve artificial intelligence. Our commitment is to use AI ethically and thoughtfully.

psAico was created in 2025, at a point where artificial intelligence (AI) is already widespread and more and more patients and professionals are turning to AI chats for help and support. The purpose of psAico is to make that use informed, healthy and ethical, with limits and in a way that makes the most of the technology safely.

The studies cited on this page:

  • Sohn JS, Ha BG, Park S, Kim JJ, Lee E, Oh H, Lee S, Kim E. Systematic review and meta analysis of chatbots in the management of depressive and anxiety symptoms. NPJ Digit Med. 2026 Mar 25;9(1):377. doi: 10.1038/s41746-026-02566-w. PMID: 41882250.
  • Gong B, Yao N, Xie H, Huang C, Kishimoto T, Berenbaum H, Mu W. Efficacy, User Engagement, and Acceptability of Cognitive Behavioral Therapy-Oriented Psychological Chatbots for Adults With Depressive and/or Anxiety Symptoms: Systematic Review and Meta-Analysis of Randomized Controlled Trials. J Med Internet Res. 2026 May 8;28:e82677. doi: 10.2196/82677. PMID: 42101333.
  • Heinz MV, Mackin DM, Trudeau BM, et al. Randomized Trial of a Generative AI Chatbot for Mental Health Treatment. NEJM AI. 2025;2(4). doi: 10.1056/AIoa2400802.
  • McFadyen J, Habicht J, Dina LM, Harper R, Hauser TU, Rollwage M. Increasing engagement with cognitive-behavioral therapy (CBT) using generative AI: a randomized controlled trial (RCT). Commun Med (Lond). 2026 Jan 15;6(1):129. doi: 10.1038/s43856-025-01321-8. PMID: 41540194.
  • Yang H, Liu J, Li L, Jiang R, Lou Q, Sun W, Zhou C. Ethical implications of high attrition in AI-based mental health interventions: a systematic review and meta-analysis. BMC Med Ethics. 2026 Jun 29. doi: 10.1186/s12910-026-01510-7. PMID: 42374513.
  • Cho HN, Wang J, Hu D, Zheng K. Large Language Model-Based Chatbots and Agentic AI for Mental Health Counseling: Systematic Review of Methodologies, Evaluation Frameworks, and Ethical Safeguards. JMIR AI. 2026 Mar 13;5:e80348. doi: 10.2196/80348. PMID: 41592221.
  • Rollwage M, McFadyen J, Juchems K, Balogh A, Pisupati S, Mircea MT, Hauser TU, Prichard G, Harper R. A cognitive layer architecture to support large-language model performance in psychotherapy interactions. Nat Med. 2026 May;32(5):1717-1725. doi: 10.1038/s41591-026-04278-w. PMID: 41820675.
  • Weilnhammer V, Hou KY, Luettgau L, Summerfield C, Dolan R, Nour MM. A clinically validated framework for auditing AI chatbot behavior in mental health interactions. Nat Med. 2026 Aug 7. doi: 10.1038/s41591-026-04577-2. PMID: 42567928.
  • Elbarbary K, Alfudoul A, Samir AA, et al. Are artificial intelligence chatbots safe for suicide risk assessment? A narratively synthesized review of current evidence. Glob Ment Health (Camb). 2026 Jun 25;13:e135. doi: 10.1017/gmh.2026.10260. PMID: 42466259.
  • Yang H, Chang F, Muroi F, Liu Z, Zhang W, Cai J. Commercial AI-Based Mental Health Chatbots as Low-Intensity Adjuncts to Psychotherapy: Effectiveness, Adherence, and Safety - A Systematic Review and Meta-Analysis. Psychother Psychosom. 2026 May 11:1-24. doi: 10.1159/000552072. PMID: 42113705.
  • Folk D, Dunn E. How Does Turning to AI for Companionship Predict Loneliness and Vice Versa? Psychol Sci. 2026 Apr;37(4):276-286. doi: 10.1177/09567976261427747. PMID: 41870975.
  • Sufyan NS, Fadhel FH, Alkhathami SS, Mukhadi JYA. Artificial intelligence and social intelligence: preliminary comparison study between AI models and psychologists. Front Psychol. 2024 Feb 2;15:1353022. doi: 10.3389/fpsyg.2024.1353022. PMID: 38379623.

A good deal of research has looked at the use of AI in mental health. These are some of the studies that inspired us:

  • Linardon J, Firth J, Carvalho AF, Bockting CL, Christensen H, Blease C, Torous J, et al. Multidisciplinary research priorities for artificial intelligence in mental health: a call to action. Lancet Psychiatry. 2026 Sep;13(9):792-803. doi: 10.1016/S2215-0366(26)00127-6. PMID: 42425118.
  • Bodner R, Lim K, Schneider R, Torous J. Efficacy and risks of artificial intelligence chatbots for anxiety and depression: a narrative review of recent clinical studies. Curr Opin Psychiatry. 2026 Jan 1;39(1):19-25. doi: 10.1097/YCO.0000000000001048. PMID: 41198140.
  • Fisher CE. The real ethical issues with AI for clinical psychiatry. Int Rev Psychiatry. 2025 Feb;37(1):14-20. doi: 10.1080/09540261.2024.2376575. Epub 2024 Jul 30. PMID: 40035378.
  • Raile, P. The usefulness of ChatGPT for psychotherapists and patients. Humanit Soc Sci Commun 11, 47 (2024). https://doi.org/10.1057/s41599-023-02567-0
  • Kolding S, Lundin RM, Hansen L, Østergaard SD. Use of generative artificial intelligence (AI) in psychiatry and mental health care: a systematic review. Acta Neuropsychiatr. 2024 Nov 11;37:e37. doi: 10.1017/neu.2024.50. PMID: 39523628.
  • Monteith S, Glenn T, Geddes JR, Whybrow PC, Achtyes ED, Bauer M. Implications of Online Self-Diagnosis in Psychiatry. Pharmacopsychiatry. 2024 Mar;57(2):45-52. doi: 10.1055/a-2268-5441. Epub 2024 Mar 12. PMID: 38471511.
  • Saqib H; Sumyia; Saqib A. AI Chatbots And Psychotherapy: A Match Made In Heaven? J Pak Med Assoc. 2023 Nov;73(11):2321. doi: 10.47391/JPMA.9608. PMID: 38013574.
  • van der Schyff EL, Ridout B, Amon KL, Forsyth R, Campbell AJ. Providing Self-Led Mental Health Support Through an Artificial Intelligence-Powered Chat Bot (Leora) to Meet the Demand of Mental Health Care. J Med Internet Res. 2023 Jun 19;25:e46448. doi: 10.2196/46448. PMID: 37335608; PMCID: PMC10337342.
  • Holohan M, Buyx A, Fiske A. Staying Curious With Conversational AI in Psychotherapy. Am J Bioeth. 2023 May;23(5):14-16. doi: 10.1080/15265161.2023.2191059. PMID: 37130403.
  • Sun J, Dong QX, Wang SW, Zheng YB, Liu XX, Lu TS, Yuan K, Shi J, Hu B, Lu L, Han Y. Artificial intelligence in psychiatry research, diagnosis, and therapy. Asian J Psychiatr. 2023 Sep;87:103705. doi: 10.1016/j.ajp.2023.103705. Epub 2023 Jul 22. PMID: 37506575.
  • Monteith S, Glenn T, Geddes J, Whybrow PC, Achtyes E, Bauer M. Expectations for Artificial Intelligence (AI) in Psychiatry. Curr Psychiatry Rep. 2022 Nov;24(11):709-721. doi: 10.1007/s11920-022-01378-5. Epub 2022 Oct 10. PMID: 36214931; PMCID: PMC9549456.
  • Pham KT, Nabizadeh A, Selek S. Artificial Intelligence and Chatbots in Psychiatry. Psychiatr Q. 2022 Mar;93(1):249-253. doi: 10.1007/s11126-022-09973-8. Epub 2022 Feb 25. PMID: 35212940; PMCID: PMC8873348.
  • Wiese W, Friston KJ. AI ethics in computational psychiatry: From the neuroscience of consciousness to the ethics of consciousness. Behav Brain Res. 2022 Feb 26;420:113704. doi: 10.1016/j.bbr.2021.113704. Epub 2021 Dec 4. PMID: 34871706; PMCID: PMC9125160.
  • Fiske A, Henningsen P, Buyx A. Your Robot Therapist Will See You Now: Ethical Implications of Embodied Artificial Intelligence in Psychiatry, Psychology, and Psychotherapy. J Med Internet Res. 2019 May 9;21(5):e13216. doi: 10.2196/13216. PMID: 31094356; PMCID: PMC6532335.

Clinically reviewed on 24 August 2026 by Dr Michele Facci, psychologist and psychotherapist - Ordine degli Psicologi della Provincia Autonoma di Trento (Italian register of psychologists) - no. 844.

Dr Michele Facci

Innovation and research

psAico comes out of the team at the psychology practice of Dr Michele Facci in Italy, a practice with a long record of professional innovation and research.

Find out more