No numbers to cherry-pick. Yet.
Peacefull is in BETA-1. Our own outcome evidence is not yet published — and the companion is deliberately scoped into supervised clinical care while that evidence is generated. This page describes the research program designed to produce it, the order in which studies are running, and the pre-registration commitments each one carries.
Status as of May 2026 · See the model card for the technical side.
- Study 01Peacefull-OBS-90IRB submission in progress
90-day observational study
Question. How do patients actually use the companion day-to-day?
Design. Prospective · observational · non-interventional. No randomization, no comparator arm.
- Study 02Hybrid-EIIn design · begins after OBS-90
Effectiveness-implementation trial
Question. Does it help, and can a practice actually run it?
Design. Hybrid type-I. Outcome change (PHQ-9, GAD-7, PROMIS, alliance) + implementation conditions.
- Study 03MRTMonths 4–6 · in design
Micro-randomized trial
Question. Which companion actions, at which moments, move measures?
Design. Within-person randomization at decision-point level. Proximal outcome measurement.
The evidence base is young.
The field of AI in mental health has produced a great deal of narrative and very little adjudicated outcome evidence. We decline to contribute to the narrative before we have contributed to the evidence. When we have pre-registered, IRB-approved data to share, we will share it — including the outcomes that flattened. Until then, this page describes the program, not the results.
A 90-day observational study of real use.
Question.
How do patients already under supervised clinical care actually use the Peacefull companion day-to-day? When, how often, for what, and with what apparent relationship to their visits?
Design.
Prospective, observational, non-interventional. 90 days per participant. No randomization, no comparator arm, no changes to care. The study describes what already happens. Pre/post survey at enrollment and 90 days; optional post-study interview subset.
Data.
App-collected engagement telemetry that the product already logs (session cadence, duration, tool use). Conversation content is not read by researchers unless a participant separately consents to de-identified excerpt review for the optional interview.
Governance.
Study sponsor: Peacefull-Behavioral, LLC (dba Peacefull-ai). Study is run by an academic research team. The researchers are not paid by Peacefull for their work on the study. A Certificate of Confidentiality has been applied for. HIPAA Authorization is a separate signed document.
What OBS-90 is not.
OBS-90 is not a test of whether Peacefull "works." It is not a comparison to anything. It does not change any participant's care. It is a descriptive study, run so future studies can ask better questions.
Status.
IRB submission in progress. The patient consent form (OBS-90 v0.1, IRB-draft) is published for review. Enrollment not yet open. For replication or protocol-review inquiries, write to clinical@peacefull-ai.io.
Does it help, and can a practice actually run it?
A hybrid type-I effectiveness-implementation design, with an academic medical center partner. Measures both clinical outcome change (PHQ-9, GAD-7, PROMIS short forms, patient-reported therapeutic alliance with the supervising clinician) and the implementation conditions that determine whether a typical outpatient practice can adopt and sustain Peacefull at all.
In design. Pre-registered analysis plan will be published before first enrollment. Timeline target: begins after OBS-90 completes.
Which companion actions, at which moments, move measures?
A micro-randomized trial (MRT) of specific companion decision points — e.g., when to encourage direct clinician contact, when to offer a DBT skill versus a Behavioral Activation prompt. Within-person randomization at the decision-point level with proximal outcome measurement. This is the design the mental-health AI field largely lacks and we intend to contribute.
Months 4–6 of the research program. In design.
What we will publish, and what we will not hide.
Pre-registration.
Every analysis is specified in advance and deposited with our clinical advisory board before data collection closes. No HARKing, no forked paths, no post-hoc rescoping.
Intent-to-treat.
Primary analyses include everyone enrolled, not only those who completed the full window. Completer-only analyses are reported as secondary sensitivity checks.
Bias and equity reporting.
Performance deltas across age, gender, race/ethnicity, primary language, and insurance status are reported for every named failure mode from the model card's six-layer evaluation. Constraints of our current scope — adults, English, US — are reported as findings, not buried as limitations.
Null and negative results.
We commit to publishing what did not move as prominently as what did. If a study contradicts an earlier claim, we correct the earlier claim publicly — in the model card version log and in the legal changelog.
Academic partners, reviewers, auditors.
We respond to good-faith replication and protocol-review requests with the IRB-approved materials and — once data is generated — the anonymized dataset and pre-registered analysis plan. Academic partnerships, protocol reviews, and independent audits — write to clinical@peacefull-ai.io. Concerns about the model card or the safety methodology — modelcard@peacefull-ai.io. Safety reports from patients, clinicians, educators, or researchers — safety@peacefull-ai.io.