Research Methodology
How we conducted The State of AI Emotional Wellness in 2026 — our data sources, evaluation criteria, scoring frameworks, limitations, and disclosure.
Last updated: June 2026 | Author: Kelly Kuo
Research Approach
This report is a secondary research analysis — it aggregates, synthesizes, and contextualizes publicly available data from peer-reviewed research, government health advisories, and published application documentation. No proprietary user data was collected or analyzed.
The report covers the AI emotional wellness landscape as of May 2026, with the goal of providing an honest, data-backed assessment of where the industry stands — including Cherizh's position within it.
We chose secondary research over primary data collection for this initial report because the existing body of research from institutions like the WHO, APA, and Stanford is robust and well-validated. Our primary data collection survey is now live to supplement future editions.
Data Sources
All data in this report comes from publicly accessible, verifiable sources. We categorize them into three tiers based on authority and rigor:
Tier 1: Government & Institutional Research
Highest authority — peer-reviewed or government-published
- U.S. Surgeon General. Our Epidemic of Loneliness and Isolation: The U.S. Surgeon General's Advisory on the Healing Effects of Social Connection and Community. 2023.
- World Health Organization. Guidelines on Digital Mental Health Interventions. WHO, 2023.
- National Institute of Standards and Technology. Cybersecurity Framework 2.0. NIST, 2024.
Tier 2: Academic & Clinical Research
Peer-reviewed studies with published methodology
- Fitzpatrick KK, Darcy A, Vierhile M. Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression via a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial. JMIR Mental Health. 2017;4(2):e19.
- Inkster B, Sarda S, Subramanian V. An Empathy-Driven, Conversational Artificial Intelligence Agent (Wysa) for Digital Mental Well-Being. Journal of Affective Disorders. 2022;301:465-471.
- Harvard Graduate School of Education. Loneliness in America: How the Pandemic Has Deepened an Epidemic of Loneliness and What We Can Do About It. Making Caring Common Project, 2021.
- Stanford Institute for Human-Centered AI. Research on AI and Therapeutic Alliance. HAI, 2023.
- Waldinger RJ, Schulz MS. The Harvard Study of Adult Development. Harvard Medical School. (Ongoing since 1938.)
Tier 3: Industry & Advocacy Data
Published statistics from recognized organizations
- National Alliance on Mental Illness. Mental Health By the Numbers. NAMI, 2024.
- American Psychological Association. The Loneliness Epidemic Persists. APA Monitor on Psychology, 2023.
- Published privacy policies, product documentation, and app store listings for Cherizh, Woebot, Wysa, Replika, Pi, and Character.ai — accessed May 2026.
Evaluation Criteria
Applications were evaluated across five dimensions. Each dimension was chosen because it directly impacts the user's experience of emotional safety and support quality.
1. Memory Architecture
Does the app remember your conversations, relationships, and emotional patterns across sessions?
Why it matters: Research on therapeutic alliance shows that continuity — feeling known — is the strongest predictor of support quality. Memory is the technical foundation of continuity.
2. Privacy Standards
Encryption level, data sales policy, AI training policy, user deletion rights, and policy transparency.
Why it matters: Users share their most vulnerable thoughts with these apps. Privacy isn't a feature — it's the foundation of trust. Evaluated against NIST cybersecurity framework standards.
3. Crisis Detection
Pattern-based early warning, warm handoff to crisis resources, post-crisis continuity, and scope clarity.
Why it matters: Emotional wellness apps will inevitably encounter users in crisis. How an app handles this transition — from everyday support to crisis handoff — is a safety-critical design decision.
4. Clinical Evidence
Published peer-reviewed studies demonstrating measurable outcomes (PHQ-9, GAD-7 improvements).
Why it matters: Claims of effectiveness should be backed by evidence. We distinguish between apps with published RCTs and those making unsubstantiated wellness claims.
5. Support Gap Coverage
24/7 availability, the "2 AM problem," accessibility without appointments or waitlists, and supplementing (not replacing) professional care.
Why it matters: The Surgeon General's Advisory and NAMI data show that the average delay between symptom onset and treatment is 11 years. The gap between needing support and accessing it is the core problem this category exists to solve.
Scoring Framework
Memory Architecture Scoring
Each app was evaluated on four memory layers. Presence of each layer was scored binary (yes/no) based on publicly documented capabilities.
| Layer | Definition | Evidence Required |
|---|---|---|
| Conversation Recall | Remembers what you said in previous sessions | Product documentation or direct testing |
| Relationship Mapping | Tracks people in your life and how you talk about them | Feature documentation or observed behavior |
| Emotional Patterns | Identifies mood trends over weeks and months | Mood tracking feature with trend analysis |
| Identity Memory | Knows who you are, your values, your journey | Persistent user profile or identity model |
Privacy Scorecard Criteria
Five privacy standards were evaluated. Each scored as compliant or non-compliant based on published privacy policies as of May 2026.
| Standard | What We Checked |
|---|---|
| AES-256 Encryption | Data encrypted in transit and at rest using AES-256 or equivalent standard |
| No Data Sales | Explicit policy stating user data is never sold to third parties |
| No AI Training | User conversations are not used to train the company's AI models |
| User Deletion | Users can permanently delete all their data on request |
| Transparent Policy | Privacy policy is readable, accessible, and not buried in legal jargon |
Limitations & Constraints
Author Conflict of Interest
The author, Kelly Kuo, is the founder of Cherizh. While we have endeavored to present all applications fairly using consistent evaluation criteria, readers should account for this potential bias. We invite corrections and challenges at support@cherizh.com.
Point-in-Time Snapshot
Application features, privacy policies, and capabilities were evaluated as of May 2026. AI companion apps evolve rapidly — features described here may have changed since publication. We will update this report quarterly.
Secondary Data Only
This edition relies entirely on publicly available data. We did not conduct original user research, clinical trials, or controlled comparisons. Our 2026 survey will contribute primary data to future editions.
Feature Verification
Not all features could be independently verified through direct product testing. Some evaluations rely on published product documentation, which may differ from actual implementation. Where direct testing was possible, we note it explicitly.
Market Coverage
This report covers the six most prominent English-language AI emotional wellness and companion apps. Regional apps, non-English apps, and enterprise wellness platforms are outside scope. Inclusion does not imply endorsement.
Review & Update Process
This report follows a structured review cycle:
- 1.Initial publication: May 2026 — baseline landscape analysis
- 2.Quarterly updates: Feature changes, new entrants, policy updates, and corrections
- 3.Annual edition: Full re-evaluation with primary survey data incorporated
- 4.Corrections: Factual errors corrected within 48 hours of verification, with changelog noted
To submit a correction, dispute a finding, or request inclusion of an additional application, contact support@cherizh.com.
Read the Full Report
This methodology page supports the full research report. Read the findings, data visualizations, and competitive analysis.