How we review evidence

What this library covers

The literature cutoff and search date is 11 November 2026. The core question concerns adults: how do defined CBT-based interventions compare with inactive controls, usual care, credible active interventions, or other delivery formats on symptoms, functioning, quality of life, remission, relapse, engagement, and adverse outcomes? Follow-up is considered separately from the immediate endpoint. Related therapies and additional-condition pages are selective evidence maps rather than comprehensive efficacy reviews.

Searches and selection

Public web searches targeted PubMed records, PMC, Cochrane evidence summaries, journal publishers, author-institution repositories, and official guideline issuers. Exact-title and DOI checks, 2024–2026 searches, and selected citation follow-up supplemented topic searches. No fixed lower publication-year cutoff was used; important earlier work was retained. Direct PubMed search endpoints were unavailable, and licensed PsycINFO and Embase were not searched independently. Selection was purposeful, not exhaustive. The search log records 225 query entries, not a count of all screened papers.

What was accessible

Full papers, abstracts, official indexed excerpts, guidelines, institutional pages, and company documentation are distinguished in the source library. Some publishers and guideline endpoints blocked access. Funding, attrition, harms, diagnostic details, and other fields were not fully extracted from every source. “Not extracted” does not mean that an outcome was absent or that no event occurred. Product descriptions were checked against supplied local documentation; they are not clinical outcome research.

Appraisal and certainty

PRISMA reporting principles, AMSTAR 2 questions, and Cochrane guidance informed transparency and appraisal. This is an AI-assisted, structured selective review. It was not preregistered and did not use independent duplicate screening, a new meta-analysis, or completed independent AMSTAR 2, RoB 2, or GRADE assessments. Published outcome-specific GRADE or CINeMA judgments are reported where accessible; other limitations are narrative assessments. No paper-count quality score has been invented.

Sources: PRISMA, n.d.; Shea et al., 2017; Schünemann, Higgins, et al., 2024; Schünemann, Vist, et al., 2024

Comparators, overlap, and corrections

Waitlist, usual-care, and active-treatment comparisons answer different questions. Within-group improvement is separate from a randomized between-group effect, and a nonsignificant difference does not establish equivalence. Reviews often contain the same underlying trials; their samples are not added together. Material GAD, CHAMP, and insomnia corrections were incorporated. Integrity checks were targeted to accessible records and selected notices, not an exhaustive Crossmark or retraction-database audit.

Sources: Papola et al., 2024; "Errors in network meta-analysis", 2024; Tyrer et al., 2021a; Tyrer et al., 2021b; Steinmetz et al., 2024b; Steinmetz et al., 2024a; Schünemann, Vist, et al., 2024

Authorship, review, and company interests

These educational summaries were prepared with AI assistance for Umbrella Journal, a company that develops journaling and digital-support products. They describe published research by the named investigators, not studies conducted by Umbrella Journal. Independent clinical review and approval have not been completed; no clinician byline or institutional endorsement is claimed. Incubator membership and institutional listings are not evidence of clinical effectiveness.

Research on a treatment and research on an app

A therapeutic approach, a delivery format, and a particular product are different units of evidence. A benefit for clinician-supported CBT does not establish a treatment effect for Umbrella Journal, Ella, or Nuggets. No identity-matched efficacy study was identified in the documented product searches and supplied material, but database and registry endpoint limits prevent a definitive statement that no study exists. Direct studies, if supplied, would need appraisal of their population, comparator, outcomes, and product version.

Updating this library

The proposed maintenance plan is quarterly targeted literature checks and a fuller annual review, with earlier updates for guideline revisions, corrections, retractions, adverse findings, or material product changes. Independent reviewers would check the clinical wording, sources, populations, and claims and disclose their qualifications and conflicts. Reviewer names and approval dates will be added only after review actually occurs. These are proposed procedures, not completed clinical approval.

Proposed research agenda

The following work is proposed. It is not a record of clinical studies already conducted by Umbrella Journal.

Establish intended use and an auditable intervention

Define which product, version, population, and feature is being evaluated. Specify whether the intended purpose is reflection, habit support, guided self-help, or treatment. Record the therapeutic content, clinician involvement, model version and prompts where relevant, exposure to the intervention, and version changes. A study of one frozen programme does not validate all later updates.

Commission independent content and safety evaluation

Arrange a scoped review by qualified clinical and ethics specialists with disclosed conflicts, alongside people with relevant lived experience. Examine clinical boundaries, coercive or stigmatizing wording, handling of uncertainty, reassurance loops, privacy, consent, accessibility, age appropriateness, and responses to deterioration. For generative features, test repeated multi-turn conversations, model changes, language variants, and adversarial or ambiguous inputs. Record findings, fixes, and re-review of the actual candidate build.

Run feasibility research before making efficacy claims

A prospective feasibility study should assess recruitment, consent comprehension, adherence, retention, burden, acceptability, adverse events, and whether the proposed measures work. Choose feasibility thresholds in advance. Do not interpret satisfaction, session counts, or pre-post improvement as proof of clinical benefit. Obtain appropriate ethics review, a data-management plan, and consent before participant research. Sample size must be justified by the study question; no generic participant target is proposed here.

Test meaningful comparative outcomes

If a clinical benefit claim is intended, develop a preregistered, adequately powered randomized study with independent collaborators and a credible comparator. Preserve participants’ usual care where appropriate. Prespecify primary outcomes and follow-up, use validated measures suited to the population, and assess functioning, quality of life, deterioration, dropout, and adverse effects in addition to symptoms. Distinguish the effect of the whole app from the added effect of AI or a specific feature through the design rather than post hoc feature correlations.

Measure access and generalizability

Recruit beyond highly engaged volunteers when appropriate and record language, disability access, digital literacy, concurrent care, age, and relevant clinical complexity. Analyze prespecified subgroup questions with adequate power and uncertainty, avoiding overinterpretation of small groups. Youth need a separate protocol and developmental safeguards. Trial retention should not be confused with real-world continued use, and missing outcomes should not disappear into a favourable completer-only analysis.

Maintain the evidence and claims record

Keep a versioned claim ledger linking exact wording to evidence and the product version it concerns. Run quarterly targeted literature checks and a fuller annual review, with earlier review triggered by guideline revisions, corrections, retractions, adverse findings, or material product changes. Log searches, decisions, changed claims, and reviewer approval. Dates for completed reviews must reflect actual completion. These activities and study designs are proposed work, not research already conducted by Umbrella Journal.