A Prompt Engineering Framework for Large Language Model–Based Mental Health Chatbots: Conceptual Framework
Bibliographic record
Abstract
BACKGROUND: Artificial intelligence (AI), particularly large language models (LLMs), presents a significant opportunity to transform mental healthcare through scalable, on-demand support. While LLM-powered chatbots may help reduce barriers to care, their integration into clinical settings raises critical concerns regarding safety, reliability, and ethical oversight. A structured framework is needed to capture their benefits while addressing inherent risks. This paper introduces a conceptual model for prompt engineering, outlining core design principles for the responsible development of LLM-based mental health chatbots. OBJECTIVE: This paper proposes a comprehensive, layered framework for prompt engineering that integrates evidence-based therapeutic models, adaptive technology, and ethical safeguards. The objective is to propose and outline a practical foundation for developing AI-driven mental health interventions that are safe, effective, and clinically relevant. METHODS: We outline a layered architecture for an LLM-based mental health chatbot. The design incorporates: (1) an input layer with proactive risk detection; (2) a dialogue engine featuring a user state database for personalization and Retrieval-Augmented Generation (RAG) to ground responses in evidence-based therapies such as Cognitive Behavioral Therapy (CBT), Acceptance and Commitment Therapy (ACT), and Dialectical Behavior Therapy (DBT); and (3) a multi-tiered safety system, including a post-generation ethical filter and a continuous learning loop with therapist oversight. RESULTS: The primary contribution is the framework itself, which systematically embeds clinical principles and ethical safeguards into system design. We also propose a comparative validation strategy to evaluate the framework's added value against a baseline model. Its components are explicitly mapped to the FAITA-MH and READI frameworks, ensuring alignment with current scholarly standards for responsible AI development. CONCLUSIONS: The framework offers a practical foundation for the responsible development of LLM-based mental health support. By outlining a layered architecture and aligning it with established evaluation standards, this work offers guidance for developing AI tools that are technically capable, safe, effective, and ethically sound. Future research should prioritize empirical validation of the framework through the phased, comparative approach introduced in this paper.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.011 | 0.016 |
| Meta-epidemiology (narrow) | 0.001 | 0.002 |
| Meta-epidemiology (broad) | 0.001 | 0.003 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.002 | 0.005 |
| Scholarly communication | 0.007 | 0.006 |
| Open science | 0.005 | 0.007 |
| Research integrity | 0.003 | 0.004 |
| Insufficient payload (model declined to judge) | 0.008 | 0.003 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".