Sources, with their limits.

Citations · AI UX Playground

Connects claims to inspectable sources. Precise verse, document and media evidence.

Limit: Citation presence alone does not establish claim support.

Progressive Disclosure · AI UX Playground

Offers layers of information instead of simultaneous detail. Answer depth and passage preview.

Limit: The proposed reader transition is a domain adaptation.

Memory Scope Toggle · AI UX Playground

Gives memory persistence an explicit scope. Reviewable optional memory with clear purpose.

Limit: A faith app still needs data governance and lawful processing.

Suggest Confirm Execute · AI UX Playground

Separates suggestions, approved changes and execution. Action previews and bounded authority.

Limit: Approval thresholds depend on consequence and previously granted permissions.

Interrupt and Resume · AI UX Playground

Makes interruption and continuation visible. Partial output and recoverable long-answer reading.

Limit: Backend recovery must match the UI promise.

Control the level of automation · Projects by IF

Lets people select an appropriate level of automation. Permissions by capability and consequence.

Limit: Standing authority must be bounded and revocable.

Scripture structure documentation · USFM documentation

Preserves distinctions in Scripture, paragraphs, notes and editorial content. Derive user scenarios from content meaning, not marker names.

Limit: A format is not a complete model of app behaviour, annotations or formation.

Footnote structure · USFM documentation

Documents note caller, origin, quotation and explanatory parts. Keep note parts and anchors intact.

Limit: Does not prescribe inline, sheet or panel presentation.

Cross-reference structure · USFM documentation

Documents cross-reference origins, targets and groups. Preview, compare and return while preserving grouping.

Limit: Reference relation does not imply theological agreement.

Translation additions · USFM documentation

Describes translator-supplied wording that clarifies meaning. Explain the edition convention.

Limit: Translation additions are not AI additions.

Poetic paragraphs · USFM documentation

Describes poetic paragraphs and relative indentation. Preserve poetic relationships through reflow.

Limit: Accessible presentation needs testing beyond import correctness.

Extended study footnotes · USFM documentation

Describes extended study content attached to the text. Publisher note with clear provenance.

Limit: A study note is distinct from a personal annotation.

Study sidebars · USFM documentation

Describes a bounded editorial sidebar. Separate longer excursions from verse text.

Limit: Sidebar content remains subject to edition rights.

Internal editorial remarks · USFM documentation

Describes remarks not intended for publication. Keep internal records out of reader-facing commentary.

Limit: Requires import and publishing controls, not only styling.

Accessible modal dialog pattern · W3C WAI

Specifies modal focus and keyboard behaviour. Accessible note sheets and modal interactions on the web.

Limit: Native apps require corresponding platform mechanisms.

Understanding Reflow · W3C WAI

Explains usable content reflow at constrained width and enlargement. Readable cards, references and comparison layouts.

Limit: Some content has defined exceptions; test the actual experience.

AI Chatbots Design Guidelines · Nielsen Norman Group

Discusses concise output, expanded detail and streaming-scroll problems. Long answers, deliberate expansion and reader-controlled scrolling.

Limit: General chatbot research, not validation of these Bible-app patterns.

Android Audio Bible · YouVersion

Documents playback, navigation, speed, timer and edition availability. Audio-first controls and accurate availability states.

Limit: Documented behaviour; no direct product usability audit performed.

Dwell Read Along · Dwell

Describes combined listening and reading. Continuous read-along as an interaction reference.

Limit: Publisher claims about comprehension are not independent study results.

Return to the playing verse · Dwell

Describes returning from browsing to the verse currently playing. Separate viewport and playback positions.

Limit: Behaviour reference, not a fresh hands-on audit.

Guided Scripture · YouVersion

Documents a guided sequence with media and Scripture. Named stages, navigation and optional media.

Limit: The proposed wireframe is an original generalisation.

How to pray Lectio Divina · Hallow

Explains reading, meditation, prayer and contemplation in this practice. Named phases and intentional silence.

Limit: A tradition-specific practice guide, not a universal or clinical protocol.

EPUB Media Overlays · W3C

Provides a model for synchronised text and audio. Honest timing granularity and playback anchors.

Limit: Not proof of a particular Bible app implementation.

Record your insights using Notes · Logos

Explains notes and selection or reference anchoring. Annotation scope, library and edition changes.

Limit: Documentation does not validate all proposed sync or overlap behaviour.

Resource Guide · Olive Tree

Describes context-related study resources. Contextual previews and linked study panels.

Limit: A product reference; its UI was not independently audited here.

Evidence-bounded study method · This atlas

Separate evidence, interpretation and missing data. Apply provenance and uncertainty to lexical study.

Limit: Not an external empirical source.

Digital Spiritual Formation research agenda · This atlas

Defines questions about actual reading, motivation, autonomy and formation. Proposed habit, progress and community patterns.

Limit: Kairos studies have not yet been supplied or evaluated; no findings are attributed to them.

Wysa FAQ · Wysa

Describes curated exercises, role boundaries and different rule-based or generative offerings. Reference for controlled content and explicit service scope.

Limit: Marketing and provider documentation, not independent efficacy evidence. Wysa is not a crisis service.

Woebot FAQ · Woebot Health

Records retirement of the app on 30 June 2025. Historical reference for curated conversational content.

Limit: Not a currently available app recommendation.

Therabot trial overview · Dartmouth · 27 March 2025

Reports a randomised study with 106 participants given Therabot and 104 controls without it, with symptom improvements and clinical oversight. Shows why system-specific evidence and oversight matter for health claims.

Limit: Not a head-to-head comparison with therapy. Full NEJM AI text was unavailable in this review. Findings do not transfer automatically to general or faith chatbots.

Randomized Trial of a Generative AI Chatbot for Mental Health Treatment · Heinz and colleagues · NEJM AI · 2025

Original publication associated with the Dartmouth trial. Further review is needed before citing detailed methods or numerical effect estimates from the paper.

Limit: Access failure is explicitly recorded; the institutional summary supports the limited account above.

Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers · Moore and colleagues · FAccT 2025

Experiments found stigma and inappropriate responses, including unsafe handling of delusional beliefs, in tested models. Inform adversarial and multi-turn evaluation for distress and dangerous claims.

Limit: Specific models and experimental scenarios; not a measurement of all present or future systems.

Concept of Trauma and Guidance for a Trauma-Informed Approach · SAMHSA · 2014

Principles include safety, transparency, collaboration, empowerment and cultural sensitivity. Design lens for predictable transitions, voluntary disclosure and accountable data practices.

Limit: An organisational approach, not a validated AI script. The full guide needs deeper review before claiming implementation fidelity.

Psychological first aid Guide for field workers · WHO, War Trauma Foundation and World Vision International · 2011

Frames humane practical support with respect for dignity, culture and abilities. Prioritise practical orientation, user choice and access to real support.

Limit: Designed for human helpers, not authorisation of autonomous AI crisis care.

Evidence standards framework for digital health technologies · NICE · updated 2022

Relates evidence expectations to purpose and potential risk. Distinguish usability, wellness and treatment claims and their required evidence.

Limit: A UK evaluation framework, not German law or a product certification.

For Faith Leaders · International OCD Foundation

Explains how repeated reassurance and symptom accommodation can maintain OCD cycles. Inform specialist-reviewed scenarios involving repetitive moral certainty requests.

Limit: Do not diagnose ordinary doubt or use a chatbot to perform exposure therapy. Guidance for human faith leaders requires careful adaptation.

Dealing with psychological crises · German Federal Ministry of Health portal

Names emergency support including 112 for acute danger in Germany. Maintain locally appropriate, verified contact cards.

Limit: Country-specific guidance; routing proposals are not guaranteed crisis detection.

Process personal data lawfully · European Data Protection Board

Special-category processing needs an appropriate Article 9 exception in addition to an Article 6 basis. Separate conversation processing, durable memory, training and sharing purposes.

Limit: Specific lawful basis, retention and DPIA decisions require the actual data flows.

AI Act Article 50 · European Commission AI Act Service Desk

Sets transparency duties for specified systems, including direct AI interaction unless obvious. Make the AI role understandable by the first interaction or exposure as applicable.

Limit: Duties differ by role and content; the whole article must be assessed for the product.

AI Act Article 5 · European Commission AI Act Service Desk

Defines prohibited practices subject to specific statutory conditions. Review manipulation and exploitation of vulnerability in nudges, monetisation and dialogue.

Limit: Not every reminder or personalisation feature is prohibited; the statutory harm criteria matter.

AI Act Article 6 · European Commission AI Act Service Desk

Defines high-risk classification conditions. Assess intended function, regulated product relationship and applicable conformity assessment.

Limit: A mental health topic alone does not make every chatbot high-risk.

AI Act Article 113 · European Commission AI Act Service Desk

Lists general application and staged provisions for specified high-risk systems. Check the current schedule for the product’s actual classification.

Limit: Do not substitute an old blanket deadline; legal and guidance updates need checking before launch.

Guidance on software qualification and classification MDR and IVDR · MDCG 2019-11 Rev. 1 · June 2025

Software qualification depends on intended medical purpose and functions; claims and promotion matter. Align product behaviour and marketing with the actual purpose.

Limit: A disclaimer does not override medical functionality or contradictory claims. Guidance is not a product-specific legal opinion.

Psychotherapeutengesetz Section 1 · German federal legislation

Regulates the protected professional title and scope of psychotherapeutic practice. Review persona, professional representation and service claims.

Limit: This is one part of the assessment; it does not alone settle all rules on practising medicine.

Heilmittelwerbegesetz · German federal legislation

Regulates health-related advertising within its scope, including misleading claims. Avoid unsupported or guaranteed treatment claims in copy and onboarding.

Limit: Applicability depends on the advertised product and claims.

Barrierefreiheitsstärkungsgesetz Section 1 · German federal legislation

Defines covered products and services including relevant e-books and e-commerce services. Assess accessible reading, audio, notes and purchase flows.

Limit: Does not apply automatically to every faith app; scope and exemptions require checking.

Understanding Use of Color · W3C WAI

Colour must not be the only means of conveying information. Playback cues and overlapping reading states.

Limit: Accessibility guidance supports cue design; it does not validate a particular faith-app layout.

Understanding Reflow · W3C WAI

Content should remain usable as available width changes. Reading anchors, passage cards and translation reflow.

Limit: Semantic scroll restoration is an original design proposal, not a requirement stated by this criterion.

Towards understanding sycophancy in language models · Anthropic researchers · 2023

Studied assistants sometimes preferred agreement with user views over truthfulness. Test belief reversal, flattering moral endorsement and harmful agreement.

Limit: Specific studied models and tasks; not a faith-app intervention study.

Five action steps for helping someone having thoughts of suicide · NIMH · Revised 2024

Direct questions, nonjudgmental listening, safety and connection to people are part of human support. Inform specialist-reviewed suicide response and human-help routes.

Limit: Human-helper guidance, not a validated chatbot risk assessment. US contact information must not be exported to other jurisdictions.

Caring for women subjected to violence: health-care provider curriculum · WHO · Revised 2021

LIVES organises listening, inquiry, validation, safety and support. Design disclosure responses around needs and choice rather than interrogation.

Limit: Clinical training for human providers focused on violence against women; transfer to other populations or AI requires separate review.

Section 4.2: Spiritual Abuse · Church of England safeguarding e-manual

Addresses coercive control using spiritual authority and warns against minimisation or directing victims to confront an abuser. Independent support and a response that does not defend the institution.

Limit: Institutional safeguarding guidance, not universal law or an automated abuse classifier; full-page review remains outstanding.

Ruth AI Chatbot FAQ · National Domestic Violence Hotline

Describes bounded tech-safety assistance, human support and limitations of its quick-exit feature. Compare AI versus staffed support, retention explanations and exit controls.

Limit: Provider documentation, not independent safety or efficacy evidence; no assumption that another product shares its safeguards.

Technology-Facilitated Abuse · National Domestic Violence Hotline

Describes abuse involving devices and accounts. Consider monitored devices, notifications and safe support access.

Limit: US service context; device-specific advice needs specialist review.