Support in daily life
A place to reflect and return to clinician-approved goals when life happens between appointments.
THE CASE FOR CONNECTED CARE
More support.
More possibility.
What do the studies actually measure? Explore how people use digital skills, what guided support has achieved in research, and how feedback can help human care.
These studies evaluate other tools and care approaches. The charts below explain the groups, measures and limits. They are not measured Moonbridge outcomes.
People are already turning to AI in hard moments.
OpenAI describes ChatGPT as “used by more than a billion people in increasingly personal areas like emotional support, life advice, and help during moments of crisis.”
Its own clinician-built benchmark found today’s models uneven, weakest at asking the right questions and judging urgency. That support is happening with or without the profession. Moonbridge brings it inside your practice: the same day-or-night access your clients are already reaching for, connected to your care plan and supervised by you. You stay the decision-maker.
MentalHealthBench (OpenAI, 2026): 1,215 synthetic conversations scored against rubrics from more than 80 licensed psychologists and psychiatrists, graded by an OpenAI model; the top-scoring model scored about 57%. The billion figure is ChatGPT’s total weekly reach, not a count of mental health use. The benchmark tests bare models, not Moonbridge. Source and limits.
Do people return to the skills?
A 2026 trial compared interactive software with a static digital workbook: the same CBT curriculum, presented as conversation or as pages of exercises.
Returning to the skills
Resource openings, relative to workbook use (1×)
Static digital workbook1×
Interactive software2.4×
2.4× means more frequent use, not 2.4 times more improvement.
Spending time with the exercises
Time using the material, relative to workbook use (1×)
Static digital workbook1×
Interactive software3.8×
3.8× measures time spent, not completed exercises or clinical benefit.
For a practice, this raises a useful design question: how can clients return to agreed skills when they need them? It does not prove that more app use produces better care.
What changed in symptom scores?
In a 2025 research trial, a generative chatbot was compared with waiting for access, not with ongoing clinician-led therapy. The chart shows the depression subgroup.
Research intervention
Waitlist control
Digital skills with a person checking in.
A 2026 trial paired a structured, rule-based coping program with five weekly 15-minute calls from trained nonspecialist supporters. The comparison group received mental-health information and links to services.
Overall distress
Participants meeting the meaningful-improvement threshold (%)
Information + service links41.6%
Skills program + support calls68.6%
About 69 in 100 versus 42 in 100 met this symptom-score threshold. This is not a recovery rate.
Anxiety symptoms
Participants meeting the meaningful-improvement threshold (%)
Information + service links41.9%
Skills program + support calls65.9%
About 66 in 100 versus 42 in 100 met this symptom-score threshold. This is not a recovery rate.
Depression symptoms
Participants meeting the meaningful-improvement threshold (%)
Information + service links39.8%
Skills program + support calls59.1%
About 59 in 100 versus 40 in 100 met this symptom-score threshold. This is not a recovery rate.
344 young adults in Jordan; 220 completed three-month follow-up. Missing data were estimated statistically. ELHRA-funded; no competing interests declared. The trial tested the program and calls together, not the software alone. Three months does not establish long-term benefit. Study identity and methods.
Making the request for care easier.
This UK study concerns the front door to NHS treatment: people asking a service for help through a conversational intake tool. It is not about redirecting an existing client to another clinician, a crisis line, or a treatment such as TMS.
More people requesting treatment
Change in service referral volume (%)
Comparison services+6%
Services with conversational intake+15%
Referral means a request to enter treatment. These are increases from earlier service volumes, not percentages of patients sent elsewhere.
Who used the intake route
Referral growth within each reported group (%)
Ethnic-minority patients+29%
Nonbinary patients+179%
Each group has its own starting volume. A 179% increase means about 2.79 times that earlier volume, not 179% of patients.
Why it is here: intake design can affect who asks for help. This is context about access, not a Moonbridge referral feature. Our companion supports clients within their existing practice relationship.
Useful information for the person providing support.
These studies ask different questions. One measures the empathy of peer-support messages; the other pools studies of symptom feedback to clinicians. Neither tests replacing a clinician.
Helping people write empathetic replies
Increase in rated message empathy with software assistance (%)
All peer supporters+19.6%
Those who found support difficult+38.9%
People wrote the replies with suggested edits. The second row is a subgroup, not a separate treatment arm. Ratings of empathy are not symptom outcomes.
Giving clinicians feedback on progress
Standardized symptom benefit (d), with 95% confidence intervals
All patients0.15 (0.10–0.20)
Patients not on track0.17 (0.11–0.22)
These are small average differences on a standardized scale, not 15% or 17% improvement. The ranges show statistical uncertainty; the second row is a subgroup.
The design implication for Moonbridge: give clinicians useful, permitted context and keep clinical decisions with them.
A more connected future for therapy.
Our direction is practical: make the days between sessions more useful to the people on both sides of the relationship.
Context chosen with care
Clients choose material to share. Clinicians can bring that context into the next conversation.
Software built around clinical judgment
Improve the instructions, safeguards and evaluation cases that connect advancing models to clinical workflows. Measure usefulness and safety as that system develops.
This is our product direction. Improvement means refining the instructions, safeguards and evaluation around the underlying models. It does not mean training those models on client conversations. Any separate review using client content needs the applicable authorization and agreements.
Read the date alongside the question.
Older care research and newer software trials serve different purposes. We do not rank unlike interventions on one converted scale.
Care principles
2018 and 2021 reviews examine feedback in human care. They can inform workflow design without establishing what a current language model can do.
Recent software studies
2025 and 2026 trials evaluate particular systems, populations and time periods. Newer capabilities do not automatically inherit those results.
Long-term questions
Pooled follow-up findings remain limited. Weeks or a few months cannot establish durable benefit or safety in ordinary practice.
Where a companion fits in care.
A practical comparison of workflows, not evidence that one produces better outcomes.
| Your practice today | With a companion | Your client, on their own | |
|---|---|---|---|
| Between-session support | Your existing resources and contact arrangements | A companion connected to approved goals | A separate consumer service |
| You get a timestamped record of what you were told | Your existing record process | Selected messages and bounded review items | Only what a client brings to you |
| Crisis response | Your existing crisis protocol | Automated screening and your practice protocol | Varies by provider |
| Expects you to watch it live | Your agreed contact and coverage arrangements | No continuous human monitoring promised | Not connected to practice coverage by default |
| Conversations train a model | Depends on the tools your practice uses | Not used for training under the service terms | Depends on provider, product and settings |
| Engagement design | Your existing approach | No streaks or engagement badges | Varies by product |
| Your practice holds the clinical record | ✓ | ✓ | Separate from the practice record |
What still needs to be established for Moonbridge.
A promising field and a validated product are different things.
Moonbridge has published no trial of its own and makes no claim about its own outcomes. Product testing, clinical review and real-world research answer different questions. Practices should assess the exact software, its safeguards, the work it supports, and the evidence for that version.