A Comparison of Responses from Human Therapists and Large Language Model–Based Chatbots to Assess Therapeutic Communication: Mixed Methods Study
Bibliographic record
Abstract
BACKGROUND: Consumers are increasingly using large language model-based chatbots to seek mental health advice or intervention due to ease of access and limited availability of mental health professionals. However, their suitability and safety for mental health applications remain underexplored, particularly in comparison to professional therapeutic practices. OBJECTIVE: This study aimed to evaluate how general-purpose chatbots respond to mental health scenarios and compare their responses to those provided by licensed therapists. Specifically, we sought to identify chatbots' strengths and limitations, as well as the ethical and practical considerations necessary for their use in mental health care. METHODS: We conducted a mixed methods study to compare responses from chatbots and licensed therapists to scripted mental health scenarios. We created 2 fictional scenarios and prompted 3 chatbots to create 6 interaction logs. We recruited 17 therapists and conducted study sessions that consisted of 3 activities. First, therapists responded to the 2 scenarios using a Qualtrics form. Second, therapists went through the 6 interaction logs using a think-aloud procedure to highlight their thoughts about the chatbots' responses. Finally, we conducted a semistructured interview to explore subjective opinions on the use of chatbots for supporting mental health. The study sessions were analyzed using thematic analysis. The interaction logs from chatbot and therapist responses were coded using the Multitheoretical List of Therapeutic Interventions codes and then compared to each other. RESULTS: We identified 7 themes describing the strengths and limitations of the chatbots as compared to therapists. These include elements of good therapy in chatbot responses, conversational style of chatbots, insufficient inquiry and feedback seeking by chatbots, chatbot interventions, client engagement, chatbots' responses to crisis situations, and considerations for chatbot-based therapy. In the use of Multitheoretical List of Therapeutic Interventions codes, we found that therapists evoked more elaboration (Mann-Whitney U=9; P=.001) and used more self-disclosure (U=45.5; P=.37) as compared to the chatbots. The chatbots used affirming (U=28; P=.045) and reassuring (U=23; P=.02) language more often than the therapists. The chatbots also used psychoeducation (U=22.5; P=.02) and suggestions (U=12.5; P=.003) more often than the therapists. CONCLUSIONS: Our study demonstrates the unsuitability of general-purpose chatbots to safely engage in mental health conversations, particularly in crisis situations. While chatbots display elements of good therapy, such as validation and reassurance, overuse of directive advice without sufficient inquiry and use of generic interventions make them unsuitable as therapeutic agents. Careful research and evaluation will be necessary to determine the impact of chatbot interactions and to identify the most appropriate use cases related to mental health.
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 | 0.060 | 0.094 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.004 | 0.003 |
| Science and technology studies | 0.003 | 0.002 |
| Scholarly communication | 0.003 | 0.002 |
| Open science | 0.002 | 0.004 |
| Research integrity | 0.002 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".