How Acolher stays scientifically grounded
The framework we use to keep wellbeing content evidence-informed, ethically designed, clinically reviewed and continuously improved.
Clinical principles
Ten commitments that govern every exercise, learning module, prompt and safety pathway.
Evidence-based
Guidance draws on established literature; where evidence is thin, we say so.
Person-centered
The person sets the pace, the goal and what is relevant to them.
Trauma-informed
No forced disclosure, no pressure, no re-exposure to painful detail for its own sake.
Culturally sensitive
Language, examples and resources adapt to region and culture, not a single default.
Ethically designed
No manipulation, no engagement tactics, no dependency-building.
Scientifically reviewed
Content is checked against sources before release and re-checked on a schedule.
Privacy first
Clinical value never justifies collecting more personal data than needed.
Human oversight
People approve safety, governance and knowledge changes — automation does not.
Continuous improvement
New evidence, safety findings and feedback feed back into the content.
Transparency
We state what the system is, what it is not, and how confident we are.
Scientific foundations
Wellbeing guidance in Acolher is informed by these fields. Techniques are adapted as self-help and psychoeducation — they are not delivered as therapy.
| Field | How it is used in Acolher | Evidence maturity |
|---|---|---|
| Cognitive Behavioral Therapy (CBT) | Thought–emotion–behavior mapping, cognitive reappraisal, behavioral activation exercises | Well established |
| Acceptance and Commitment Therapy (ACT) | Values clarification, defusion and acceptance language in reflection prompts | Well established |
| Motivational Interviewing | Non-confrontational, autonomy-supportive conversation style and change talk | Well established |
| Positive Psychology | Strengths, gratitude and meaning practices in the wellbeing center | Established |
| Mindfulness-based approaches | Breathing, grounding, body scan and attention practices | Well established |
| Emotion regulation research | Labeling, reappraisal and distress-tolerance strategies | Well established |
| Stress management research | Recovery, micro-break and workload-boundary guidance | Established |
| Resilience research | Support networks, meaning-making and recovery routines | Established |
| Behavior change science | Habit design, cues, streaks and implementation intentions | Established |
| Sleep science | Sleep hygiene, wind-down routines and circadian guidance | Well established |
| Communication science | Difficult conversations, feedback and boundary-setting | Established |
| Leadership and organizational psychology | Psychological safety, workload design and manager coaching | Established, context-dependent |
We avoid unsupported or pseudoscientific claims. Anything experimental is labeled as a hypothesis, and no exercise is presented as a treatment or a cure.
Clinical Review Board
In progressA multidisciplinary board is how clinical judgment enters the product. We describe the intended composition and duties openly, including the parts that are not in place yet.
Intended composition
- Licensed psychologists
- Psychiatrists
- Behavioral scientists
- Clinical researchers
- Organizational psychologists
- Ethics specialists
- AI safety specialists
- Medical advisors, when a topic requires them
Responsibilities
- Review clinical and educational content before publication
- Approve new wellbeing programs and campaign templates
- Review AI system prompts and specialist instructions
- Evaluate safety behavior and escalation pathways
- Review the scientific evidence behind each claim
- Approve content updates and version changes
- Review crisis protocols and regional resources
Content lifecycle
Every clinical content item — exercise, learning module, knowledge entry or specialist prompt — carries a governance record.
Record fields
Unique identifier
Stable ID used across the app, analytics and review logs.
Author
Who wrote the content.
Clinical reviewer
Who reviewed it for clinical appropriateness.
Scientific references
The literature or guideline the content is based on.
Evidence level
Well established, established, emerging or hypothesis.
Review date
When it was last examined.
Approval date
When it was cleared for release.
Version history
What changed, when and why.
Status
Draft, in review, approved, needs update or archived.
Next review date
Scheduled re-examination, at least annually.
Archived versions
Superseded versions kept for traceability.
Stages
- 1
Draft
Written against a defined need with candidate references.
- 2
Evidence check
Claims traced to sources; unsupported claims removed or downgraded.
- 3
Clinical review
Reviewed for appropriateness, tone, safety and stigma-free language.
- 4
Accessibility and localization
Plain language, reading level, WCAG checks and PT-BR/EN-US parity.
- 5
Approval
Explicit human approval before it can reach production.
- 6
Monitoring
Usage, safety signals and quality scores watched after release.
- 7
Scheduled review
Re-examined on its next review date or sooner if evidence changes.
- 8
Archive
Retired content is archived with its history rather than deleted silently.
AI clinical safety boundaries
These limits sit at the top of the assistant's instructions and take precedence over anything else, including a direct user request.
The AI never
- Diagnoses psychiatric or medical disorders
- Prescribes medication
- Recommends starting, stopping or changing medication
- Guarantees outcomes
- Claims certainty it does not have
- Replaces emergency care
- Replaces psychotherapy
- Replaces psychiatric treatment
- Invents studies, statistics, sources or credentials
- Pretends to be a human or a licensed professional
The AI may
- Provide emotional support and non-judgmental listening
- Teach coping and emotion-regulation strategies
- Encourage healthy habits and recovery routines
- Offer evidence-informed exercises and reflection prompts
- Promote emotional awareness and self-understanding
- Support resilience, meaning and connection
- Explain what research generally suggests, with its uncertainty
- Encourage professional care and help the person prepare for it
High-risk situations
Risk signals are evaluated on every turn. When risk is elevated, exercises and reframing are suppressed in favor of presence and a real human path forward.
- Stay calm and stay present — no abrupt ending, no cold hand-off
- Use supportive, non-judgmental, non-alarming language
- Encourage immediate help when the situation warrants it
- Surface emergency resources appropriate to the person's region
- Encourage contacting a trusted person nearby
- Activate internal safety handling for the rest of the conversation
- Never leave the conversation without a next step
United States
988 Suicide & Crisis Lifeline (call or text 988); 911 for emergencies.
Brazil
CVV 188 (24h, free); SAMU 192 and emergency services for immediate danger.
Elsewhere
The assistant points to local emergency numbers and encourages contacting nearby support.
Content quality standards
Every educational item is held to the same bar before it ships.
Scientifically accurate
Claims are traceable to literature or clearly framed as general practice.
Easy to understand
Plain language, short sentences, no jargon without explanation.
Inclusive
Examples span different life situations, family shapes, cultures and work contexts.
Accessible
WCAG-oriented contrast, focus states, semantic structure and keyboard use.
Culturally appropriate
PT-BR and EN-US content is written natively, not machine-translated.
Free of stigmatizing language
No labels, no 'crazy', no reduction of a person to a condition.
Regularly updated
Scheduled re-review with a next-review date on each item.
Clinical monitoring
What we watch continuously, using aggregate and non-content signals.
AI response quality
An internal quality gate runs before each reply, and routing quality scores are recorded without message content.
Clinical appropriateness
Sampled reviews of specialist behavior against the boundaries on this page.
User satisfaction
In-product feedback on suggestions and exercises.
Safety events
Elevated-risk turns are counted as metrics, never stored as transcript content for review.
Escalation events
How often crisis resources are surfaced, tracked in aggregate.
Evidence updates
Watchlist for guideline changes affecting existing content.
Clinical review status
Dashboard of items approved, due for review and overdue.
Ethical principles
Ethics constraints are enforced in the assistant's instructions, not only in policy documents.
Respect autonomy
The person decides; the assistant offers options, never directives.
Avoid manipulation
No guilt, fear, artificial urgency or retention tactics.
Protect dignity
No mockery, no minimizing, no moral judgment of what a person shares.
Promote inclusion
Language and examples do not assume one identity, ability or family model.
Reduce stigma
Emotional difficulty is framed as human, not as a defect.
Respect diversity
Cultural and religious context is respected without proselytizing.
Maintain transparency
The assistant discloses that it is AI and explains how it works when asked.
Protect vulnerable users
Higher caution and lower-intensity techniques when distress is high.
Clinical documentation
What we maintain internally and can share with enterprise reviewers under NDA.
Clinical methodology
How approaches are adapted into self-help exercises and psychoeducation.
Scientific references
Source list per content item and per knowledge entry.
Evidence reviews
Notes from each evidence check, including claims that were removed.
Review schedules
Next-review dates and owners for each content family.
Governance policies
The inviolable rules applied to every AI response.
Safety protocols
Risk evaluation, escalation and regional crisis resources.
Version history
Versioned governance, prompt and knowledge changes.
Continuous improvement
Review cadence and the inputs that trigger an earlier revision.
Annual review baseline
Every content item has a scheduled review at least once a year.
New scientific evidence
A relevant study or meta-analysis can trigger an out-of-cycle revision.
Updated clinical guidelines
Guideline changes are treated as a mandatory review trigger.
User feedback
Reports of unclear, unhelpful or uncomfortable content are reviewed.
Safety findings
Any safety concern moves to the front of the queue.
AI quality evaluations
Quality-gate and routing metrics inform prompt and content changes.
Clinical and scientific contact
If you are a clinician, researcher or enterprise reviewer, we would rather answer a hard question than publish a vague claim.
Clinical review
clinical@acolher.liveContent review, evidence questions and Clinical Review Board participation.
Reviewing Acolher for a clinical or enterprise assessment? Ask us for the methodology and reference list behind any exercise.
Contact clinical team