Research Hub
A growing body of research is examining how conversational AI may support emotional well-being, self-reflection, and everyday coping. Early results are encouraging, but the evidence is still evolving. Many studies are small, short, product-funded, or use waitlist controls, and long-term benefits and safety remain open questions.
This library includes independent peer-reviewed research, responsible-AI guidance, and research involving Abby. Inclusion does not mean that a study evaluated or endorsed Abby. We summarize both what each study found and the limitations that matter.
The studies below are independent research. They were not conducted by Abby, and none of them evaluated or endorsed Abby. They examine conversational AI and emotional support more broadly, and many study clinical populations and outcomes. Abby itself is a subclinical, everyday emotional-support product, not a clinical treatment.
Systematic Reviews & Meta-Analyses
Syntheses that pool results across many trials. These are the strongest available summaries of the field, and also the clearest about its limits.
Effectiveness of AI and Rule-Based Conversational Agents for Depression, Anxiety and Stress: A Meta-Analysis
Across 48 randomized trials, conversational-agent interventions produced small-to-moderate improvements in depression, anxiety, and stress compared with control conditions. Effects were larger in clinical populations and shorter interventions.
What to keep in mind: The review combined AI and rule-based systems, and the publisher labels the current article as an early, unedited version. Long-term effectiveness and optimal integration with care remain uncertain.
Independent research. Not conducted by or affiliated with Abby.
The Effectiveness of CBT-Based NLP-Enabled AI Conversational Agents for Mental Health Intervention
CBT-based AI conversational agents showed a small-to-moderate effect on depressive symptoms and a small effect on negative affect. Results for generalized anxiety, stress, and positive affect were not statistically significant after adjustment for publication bias.
What to keep in mind: This is an early, unedited version of the article, and results were stronger for some outcomes than others. It is not proof that AI chatbots improve every mental-health measure.
Independent research. Not conducted by or affiliated with Abby.
Systematic Review and Meta-Analysis of AI-Based Conversational Agents for Promoting Mental Health and Well-Being
AI-based conversational agents significantly reduced symptoms of depression and distress across the pooled trials. The review did not find a significant improvement in overall psychological well-being, and the authors concluded that therapeutic relationship, content quality, and communication quality shape user experience.
What to keep in mind: Only six studies reported follow-up effects, so the review could not establish whether benefits persist over time. The studies and outcome measures were highly varied.
Independent research. Not conducted by or affiliated with Abby.
Conversational Agent Interventions for Mental Health Problems: Systematic Review and Meta-Analysis of Randomized Controlled Trials
Conversational-agent interventions produced statistically significant short-term improvements across depression, generalized anxiety, stress, distress, well-being, and several other outcomes. Personalization, empathic responses, and longer interaction duration were associated with stronger effects.
What to keep in mind: Long-term effects were not statistically significant for most outcomes. The review also included embodied agents, virtual reality, and avatar interventions, so not every included system is comparable to a text-based thought companion.
Independent research. Not conducted by or affiliated with Abby.
Effectiveness and Safety of Using Chatbots to Improve Mental Health: Systematic Review and Meta-Analysis
The review found preliminary evidence that chatbots may improve depression, distress, and stress. It also found conflicting results for anxiety and affect and no significant effect on overall psychological well-being.
What to keep in mind: The authors characterized the evidence as weak and insufficient for definitive conclusions. Only two studies assessed safety, and absence of reported harm is not strong proof of safety.
Independent research. Not conducted by or affiliated with Abby.
Randomized & Real-World Studies
Individual trials and real-world evaluations of specific conversational systems. None of these systems is Abby.
Randomized Trial of a Generative AI Chatbot for Mental Health Treatment (Therabot)
In an eight-week evaluation of adults with major depression, generalized anxiety, or eating-disorder risk, Therabot users reported meaningful reductions in symptoms across all three groups. Participants also reported a therapeutic alliance with the system comparable to levels reported in human-care settings.
What to keep in mind: The comparator was a no-access, waitlist-style control rather than active therapy or another chatbot. The researchers emphasized that generative AI still requires close clinical oversight and further safety validation.
Independent research. Not conducted by or affiliated with Abby.
Attachment, Loneliness, and Social Support as Moderators of Conversational AI-Based Mental Health Outcomes
The conversational-AI group showed greater reductions in generalized anxiety and greater gains in well-being and life satisfaction than both group therapy and a waitlist control at post-intervention and follow-up. Depression improved relative to the waitlist control, while PTSD symptoms did not differ between groups.
What to keep in mind: The population was university students, and one author was employed by the intervention company while another consulted for it and held stock options. It should not be generalized to all users or conditions.
Independent research. Not conducted by or affiliated with Abby.
Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot)
Participants assigned to Woebot showed a greater reduction in depression symptoms than participants receiving an NIMH information ebook. Anxiety improved in both groups rather than only in the chatbot group.
What to keep in mind: This was a small, short, unblinded study in young adults, with one author affiliated with the company that developed the intervention.
Independent research. Not conducted by or affiliated with Abby.
Using Psychological Artificial Intelligence (Tess) to Relieve Symptoms of Depression and Anxiety
Students with access to Tess reported reductions in depression and anxiety relative to an information-only control. The authors positioned the system as a scalable form of support rather than a replacement for a trained therapist.
What to keep in mind: The trial was registered retrospectively, the journal explicitly advised careful interpretation, company employees were authors, and the company funded participant incentives.
Independent research. Not conducted by or affiliated with Abby.
An Empathy-Driven Conversational AI Agent (Wysa) for Digital Mental Well-Being: Real-World Data Evaluation
Higher-engagement Wysa users reported greater average improvement in depression scores than lower-engagement users, and most submitted feedback described the app as helpful or encouraging. The findings suggest promise for conversational support in real-world use.
What to keep in mind: Users were not randomized, high engagement may reflect important pre-existing differences, the comparison groups were unbalanced, and all authors had advisory, employment, or consulting relationships with Wysa.
Independent research. Not conducted by or affiliated with Abby.
Evidence of Human-Level Bonds Established With a Digital Conversational Agent
Within the first five days of use, participants reported therapeutic-bond scores comparable to scores reported in prior studies of individual and group CBT. The study suggests that users can experience a working alliance with a transparently nonhuman conversational agent.
What to keep in mind: This study did not measure clinical improvement, had no direct randomized comparison group, and was conducted by the product’s developers. It demonstrates perceived bond, not equivalence to human therapy.
Independent research. Not conducted by or affiliated with Abby.
Preprints & emerging research
Early-stage work that has not completed peer review. We list it for transparency, not as established evidence.
Safety, Ethics & Healthy Use
Effectiveness is only half the question. This research informs how Abby approaches safety and responsible AI: boundaries, healthy use, and governance.
Trust-Driven Healthy Engagement With Conversational AI for Mental Health Support in Young Adults
Perceived psychological support was more strongly associated with goal-oriented, self-regulated, boundary-aware AI use than with perceived emotional warmth alone. The study supports designing for healthy use and user autonomy rather than simply maximizing emotional attachment or engagement.
What to keep in mind: The outcome was users’ perceived support, not clinically verified symptom improvement, and the measures were self-reported.
Independent research. Not conducted by or affiliated with Abby.
Ethics and Governance of Artificial Intelligence for Health: Guidance on Large Multi-Modal Models
WHO recommends transparent, task-bounded, accountable AI development with healthcare professionals and users involved throughout design and deployment. It highlights risks including inaccurate outputs, bias, privacy and cybersecurity failures, automation bias, and inequitable access.
Independent guidance. Not produced by or affiliated with Abby.
Research Involving Abby
This section is different from everything above: it contains only research that Abby conducted, commissioned, or directly participated in. We hold our own work to the same standard we apply to the independent literature, including design, sample size, and limitations.
Our research program is underway. Abby is built with practicing therapists, and we are actively evaluating how well Abby supports everyday emotional well-being, how it behaves in sensitive moments, and how reliably it points people to professional and crisis resources. We will publish summaries of that work here, with the same evidence labels and limitations we require of every other study on this page.
In the meantime, you can read about our research program and how studies involving Abby are reviewed, and about how Abby approaches safety and responsible AI.
