AI-Facilitated Cognitive Reappraisal via Socrates 2.0: Mixed Methods Feasibility Study.
Open accessThe DOI matches its Crossref record; 35 of the 35 references it cites were found in a registry or an index; 5 of the 13 sources its claims cite were checked for support.
Abstract
Background: Innovative, scalable mental health tools are needed to address systemic provider shortages and accessibility barriers. Large language model–based tools can provide real-time, tailored feedback to help users engage in cognitive reappraisal outside traditional therapy sessions. Socrates 2.0 (Rush University Medical Center) is a multiagent artificial intelligence tool that guides users through Socratic dialogue. Objective: The study aimed to examine the feasibility, acceptability, and potential for symptom reduction of Socrates 2.0. Methods: A total of 61 adult participants enrolled in a 4-week mixed methods preclinical feasibility study. The participants used Socrates 2.0 as desired and completed the self-report measures of depression, social anxiety, posttraumatic stress, and obsessive-compulsive symptoms at baseline and 1-month follow-up. Feasibility, acceptability, and appropriateness, along with usability and working alliance, were assessed via validated measures. The semistructured interviews explored user experiences and perceptions. Results: Participants engaged with Socrates 2.0 an average of 6.70 (SD 4.57) times over 4 weeks. Feasibility (mean 4.26, SD 0.67), acceptability (mean 4.16, SD 0.84), and usability ratings were high. Participants reported small-to-moderate reductions in depression (effect size d =0.30), social anxiety (d =0.25), obsessive-compulsive (d =0.33), and posttraumatic stress (d =0.28) symptoms. Working alliance scores suggested a moderately strong perceived bond with the artificial intelligence tool. Qualitative feedback indicated that the nonjudgmental, on-demand nature of Socrates 2.0 encouraged self-reflection and exploration. Some users critiqued the repeated questioning style and limited conversation depth. Conclusions: Socrates 2.0 was perceived as feasible, acceptable, and moderately helpful for self-guided cognitive reappraisal, demonstrating potential as an adjunct to traditional therapy. Further research, including randomized trials, is needed to determine effectiveness across different populations, optimize personalization, and address the repetitive conversational nature.
Overview
CoChat Summary
## TL;DR
This 4-week mixed methods feasibility study found that 61 adults generally rated Socrates 2.0 as feasible, acceptable, and usable, and some reported small-to-moderate symptom reductions after using the AI Socratic dialogue tool. The main concerns were repetitive questioning and limited conversational depth, suggesting the tool may be best as an adjunct to therapy rather than a replacement.
## Key findings
- Among 61 participants, engagement averaged 6.70 (SD 4.57) uses over 4 weeks and 10.44 (SD 7.65) exchanges per conversation.
- Feasibility, acceptability, and implementation were rated highly: Feasibility of Intervention Measure mean 4.26 (SD 0.67), Acceptability of Intervention Measure mean 4.16 (SD 0.84), and Intervention Appropriateness Measure mean 4.13 (SD 0.91).
- Participants showed small-to-moderate symptom reductions from baseline to 1 month in depression (d=0.30), social anxiety (d=0.25), obsessive-compulsive symptoms (d=0.33), and posttraumatic stress (d=0.28).
- Qualitative feedback was mostly positive: users said the tool encouraged reflection and felt nonjudgmental, but some criticized its repetitive questioning and limited depth.
## Methods
- Design: 4-week mixed methods, preclinical feasibility study with baseline and 1-month follow-up; quantitative pre-post comparison plus semistructured interviews.
- Sample/data: 61 adults recruited via social media and word of mouth; no diagnosis or current treatment required, and participants needed to be 18+, able to read/write English at a sixth-grade level or higher, and have internet access.
- Analysis: Paired-sample 2-tailed t tests and Cohen d for symptom change; qualitative data analyzed by thematic analysis informed by Consensual Qualitative Research principles.
## Limitations
- The paper states that participants self-selected, the study had no control or comparison group, the sample was not clinically diagnosed and had generally low symptom severity, the sample was highly educated, and digital literacy was not systematically assessed.
- The text also makes evident that findings are limited by the preclinical feasibility design and short 4-week duration, so effectiveness and durability of effects remain uncertain.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
5 Valid · 0 Not valid · 5 Not checked · 39 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Compared to an earlier single-agent version of the tool, the addition of the 2 supervisory AI agents appears to improve dialogue quality by reducing the risk of redundant, off-topic, or overly lengthy answers and repetitive conversational loops (ie, segments in which similar questions or statements are repeated without progressing the dialogue) [13].”
§ AI-Facilitated Cognitive Reappraisal via Socrates 2.0: Mixed Methods Feasibility Study > Introduction
What the sources say
- Partly supportedcites [13] Held, A novel cognitive behavioral therapy–based generative AI tool (Socrates 2.0) to facilitate Socratic dialogue: protocol for a mixed methods feasibility study, JMIR Res Protoc sourceExperimental — accuracy not yet measured
The Introduction’s “Socrates 1.0” section describes the earlier single-agent tool as verbose, prone to getting off task, and caught in conversational loops. The “Socrates 2.0” section says the added agents appeared to improve dialogue quality and reduce looping, with supervisor feedback helping keep conversations focused. Our copy does not establish that the added agents reduced redundant or overly lengthy answers specifically. (LLM, conf=0.84)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“It is plausible that the human-like responses that LLMs produce lead to a greater sense of being understood compared to rule-based algorithms where responses often did not feel tailored to the individual [3].”
§ AI-Facilitated Cognitive Reappraisal via Socrates 2.0: Mixed Methods Feasibility Study > Discussion This study evaluated the feasibility and acceptability of Socrates 2.0, a multiagent AI tool designed to engage individuals in Socratic dialogue. Participants generally reported high levels of satisfaction with Socrates 2.0’s accessibility and a nonjudgmental interaction style. In line with other digital mental health tools, it appears that the 24/7 access can be a key factor for high user engagement, particularly among those who avoid traditional in-person therapy due to logistical constraints or concerns about stigma [42930]. In this study, participants described Socrates 2.0 as a safe space to explore personal struggles. Some participants even reported disclosing things to the tool that they would have been hesitant to discuss with a human therapist for fear of judgment. In other cases, participants remarked that working with Socrates 2.0 was a way for them to practice what to say to their human therapist at a later time. These insights contrast common fears associated with the use of AI in the context of mental health care, where privacy is commonly listed as a key issue [56]. Notably, none of the participants in this study raised privacy or ethical concerns regarding their use of Socrates 2.0. This may have been due to the detailed informed consent process, which clearly explained the data-handling procedures, including where data were stored and who had access to the data, and emphasized that the tool was not a substitute for therapy. Although self-selection may be a key factor, the absence of reported concerns may also suggest that transparency around privacy protocols is important when engaging with AI-based mental health tools. Many participants praised the human-like quality of Socrates 2.0’s responses, which was further supported by above-average ratings of the working alliance. It is plausible that the human-like responses that LLMs produce lead to a greater sense of being understood compared to rule-based algorithms where responses often did not feel tailored to the individual [3]. To optimize future tools, research should compare theorized therapeutic mechanisms, such as the working alliance [31], between algorithms to identify the exact mechanisms that contribute to positive experiences and treatment outcomes. Interestingly, some individuals in our sample initially expected Socrates 2.0 to function like a general-purpose LLMs, such as ChatGPT, but found Socrates 2.0’s focus on cognitive reappraisal offered more depth and made exchanges more focused and meaningful. These findings support literature suggesting that specialized AI tools, rather than broad all-purpose models, may be more successful at engaging users in structured therapeutic exercises [4732]. Participants also described several areas for improvement. In contrast with the positive quantitative and qualitative feedback, some participants described interactions as repetitive and said that they felt caught in circular conversations or that they were receiving basic follow-up questions instead of genuinely novel insights. Others remarked that conversations lacked depth at times. This perception likely reflects the inherent structure of Socratic dialogue, which depends on a sequence of open-ended, reflective questions intended to guide users towards deeper cognitive reappraisal rather than to provide direct suggestions or advice. While this repetition served a therapeutic purpose, it may contrast with user expectations of conversational variety typically associated with general-purpose language models (eg, Open AI’s ChatGPT). In line with this, some users who had these criticisms remarked that they had hoped for specific advice and guidance from the AI therapist rather than continued questions. Another element that may have contributed to the repetitive nature is the supervisory agents within Socrates 2.0, which are designed to monitor the dialogue and ensure that the AI therapist maintains focus on the user’s stated belief. This keeps the conversation on task and aligned with Socratic principles, but may also increase what some users perceive as repetitive statements when users attempt to shift topics or seek broader discussion. In addition to refining the hidden supervisory AI agents, it will be important to evaluate the impact of different AI therapist response styles as well as the importance of setting expectations for the interactions with tools like Socrates 2.0. This may be especially relevant for individuals who have firm ideas about what therapy should look and feel like and expect interactions with an AI therapist to mimic such expectations. It may also be important to leverage LLMs to detect possible conversational fatigue and build an adaptive algorithm that uses such data to signal to the AI therapist to adjust the response style. Although accessibility was a key reason for developing Socrates 2.0, only a subset of participants endorsed survey items reflecting improvement in access to care. Possible reasons for this may include items being worded too broadly, such as by asking about health care access, whereas Socrates 2.0 was only designed to facilitate one concrete therapeutic task rather than function as an alternative to mental healthcare or even health care as a whole. At the same time, the tested version of Socrates 2.0 had some accessibility limitations that may have contributed to the lower survey item endorsement. For example, the tested version was designed to be used on a desktop or laptop and was not optimized for use on smartphones. In addition, this study required users to log in manually with a pregenerated username and password, which may have impacted the perceived convenience. Future iterations of the tool have already begun to address accessibility limitations through streamlining login procedures for users and providing a better mobile experience. As part of this study, we also gathered descriptive data on how frequently and when participants chose to interact with Socrates 2.0. Whereas some participants reported logging in for short “bursts” of conversation during moments of stress or uncertainty, others preferred extended sessions in the evenings or weekends, which are perhaps more akin to traditional therapy appointments. These data suggest that individuals use digital mental health tools differently compared to traditional 50-minute therapy appointments and underscore the need for such tools to be designed differently than to simply mimic human-delivered therapy sessions. Despite promising indications, the study carries notable limitations that shape the interpretation and applicability of its findings. First, participants self-selected into the study and may have been more comfortable with technology and AI than the general population. Second, the feasibility-focused design lacked a control or comparison group and only included exploratory quantitative analyses; therefore, the significance of these findings is limited. Although participants reported a range of mental health symptoms, the sample was not selected based on clinical diagnoses, and symptom severity was generally low across all screening measures. Future research should assess feasibility, acceptability, and symptom impact in clinical populations with confirmed mental health diagnoses. The sample was also highly educated, which may limit generalizability to broader populations. Though LLMs are flexible at accommodating a range of reading levels and communication styles, future research should evaluate the tools' usability and acceptability across more diverse educational and demographic backgrounds. Finally, digital literacy was not systematically assessed and may have influenced the frequency and effectiveness of user engagement with Socrates 2.0. Socrates 2.0 holds clear potential to facilitate cognitive reappraisal outside of therapy sessions, largely due to its 24/7 availability, sense of anonymity, and engaging conversational style. Participants’ remarks often framed Socrates 2.0 as a potential adjunct rather than a replacement for human-delivered therapy, which reflects a stance increasingly recommended in the literature: AI tools may be best used within a stepped-care or blended-care model, where AI-driven conversation augments but does not supplant clinician-led treatment [4]. Future research should examine multi-month engagement patterns and involve randomized controlled trials comparing Socrates 2.0 with standard CBT worksheets to determine whether the AI-driven dialogue can produce measurable clinical advantages. Research should also examine different subgroups, including participants with severe symptoms, culturally diverse backgrounds, or varying degrees of digital literacy. Overall, this work underscores the promise and challenges of LLM-driven tools in the domain of mental health and highlights an important and exciting avenue for future research and development as AI technologies continue to evolve.
What the sources say
- Partly supportedcites [3] Abd-Alrazaq, Perceptions and opinions of patients about mental health chatbots: scoping review, J Med Internet Res sourceExperimental — accuracy not yet measured
The Discussion (Practical and Research Implications; page unavailable) says chatbots relying on fixed rule bases cannot adapt to specific user needs, supporting the concern about responses not feeling tailored. The review does not assess LLM responses or compare whether LLMs give users a greater sense of being understood than rule-based chatbots. (LLM, conf=0.84)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Participants reported small-to-moderate reductions in depression (effect size d=0.30), social anxiety (d=0.25), obsessive-compulsive (d=0.33), and posttraumatic stress (d=0.28) symptoms.”
§ AI-Facilitated Cognitive Reappraisal via Socrates 2.0: Mixed Methods Feasibility Study > Abstract > Results
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Participants reported small-to-moderate reductions in depression (effect size d=0.30), social anxiety (d=0.25), obsessive-compulsive (d=0.33), and posttraumatic stress (d=0.28) symptoms.”
§ AI-Facilitated Cognitive Reappraisal via Socrates 2.0: Mixed Methods Feasibility Study > Abstract > Results
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“A subset of users felt that Socrates 2.0 was unable to thoroughly address complex emotional concerns or trauma-related issues, describing some responses as superficial or generic.”
§ AI-Facilitated Cognitive Reappraisal via Socrates 2.0: Mixed Methods Feasibility Study > Results > Negative Themes
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
19 more claims
Create a free account to see all 49 claims and every source verdict.
- It may state two separate claims as one. 18
- Probably not a substantial claim. 14
- The quote only partly states this claim. 2
Citations
Cites
Using thematic analysis in psychology
Braun, V., & Clarke, V. (2006) · Qualitative Research in Psychology · doi:10.1191/1478088706qp063oa · Cited by 194,414
A Guide to Conducting Consensual Qualitative Research
Hill, C. E., Thompson, B. J., & Williams, E. N. (1997) · The Counseling Psychologist · doi:10.1177/0011000097254001 · Cited by 2,884 · cited 1×
“Qualitative responses were analyzed using thematic analysis [27], informed by principles of Consensual Qualitative Research to minimize individual bias [28].”
Psychometric assessment of three newly developed implementation outcome measures
Weiner, B. J., Lewis, C. C., Stanick, C., Powell, B. J., Dorsey, C. N., Clary, A. S., Boynton, M. H., & Halko, H. (2017) · Implementation Science · doi:10.1186/s13012-017-0635-3 · Cited by 2,819 · cited 1×
“…and also completed a set of validated usability and acceptability measures, such as the System Usability Scale [23], the Acceptability of Intervention Measure, the Intervention Appropriateness Measure, the Feasibility of Intervention Measure [24], and the mHealth App Usability Questionnaire [25].”
Löwe, B., Wahl, I., Rose, M., Spitzer, C., Glaesmer, H., Wingenfeld, K., Schneider, A., & Brähler, E. (2010) · Journal of Affective Disorders · doi:10.1016/j.jad.2009.06.019 · Cited by 2,473 · cited 2×
“…clinical cut-off scores indicating probable clinical range are approximately ≥6 on the Patient Health Questionnaire-4 for depression or anxiety [15], ≥10 on the IAS-3 for social anxiety [16], ≥4 on the PTSD Checklist-4 for PTSD [20], and ≥4 on the Obsessive-Compulsive Inventory-4 for…”
The System Usability Scale: Past, Present, and Future
Lewis, J. R. (2018) · International Journal of Human–Computer Interaction · doi:10.1080/10447318.2018.1455307 · Cited by 1,958 · cited 1×
“…after the 4-week study period and also completed a set of validated usability and acceptability measures, such as the System Usability Scale [23], the Acceptability of Intervention Measure, the Intervention Appropriateness Measure, the Feasibility of Intervention Measure [24], and the…”
The mHealth App Usability Questionnaire (MAUQ): Development and Validation Study
Zhou, L., Bao, J., Setiawan, I. M. A., Saptono, A., & Parmanto, B. (2019) · JMIR mHealth and uHealth · doi:10.2196/11500 · Cited by 675 · cited 1×
“…and also completed a set of validated usability and acceptability measures, such as the System Usability Scale [23], the Acceptability of Intervention Measure, the Intervention Appropriateness Measure, the Feasibility of Intervention Measure [24], and the mHealth App Usability Questionnaire [25].”
Li, H., Zhang, R., Lee, Y.-C., Kraut, R. E., & Mohr, D. C. (2023) · npj Digital Medicine · doi:10.1038/s41746-023-00979-5 · Cited by 598 · cited 2×
“These findings support literature suggesting that specialized AI tools, rather than broad all-purpose models, may be more successful at engaging users in structured therapeutic exercises [4,7,32].”
Munder, T., Wilmers, F., Leonhart, R., Linster, H. W., & Barth, J. (2010) · Clinical Psychology & Psychotherapy · doi:10.1002/cpp.658 · Cited by 538 · cited 1×
“Participants were also asked to complete the Working Alliance Inventory–Short Revised [26] to evaluate whether they experienced any sense of collaborative alliance with the AI system, a validated assessment traditionally associated with measuring therapeutic rapport.”
Stade, E. C., Stirman, S. W., Ungar, L. H., Boland, C. L., Schwartz, H. A., Yaden, D. B., Sedoc, J., DeRubeis, R. J., Willer, R., & Eichstaedt, J. C. (2024) · npj Mental Health Research · doi:10.1038/s44184-024-00056-z · Cited by 391 · cited 5×
“…adjunct rather than a replacement for human-delivered therapy, which reflects a stance increasingly recommended in the literature: AI tools may be best used within a stepped-care or blended-care model, where AI-driven conversation augments but does not supplant clinician-led treatment [4].”
InfluentialCoombs, N. C., Meriwether, W. E., Caringi, J., & Newcomer, S. R. (2021) · SSM - Population Health · doi:10.1016/j.ssmph.2021.100847 · Cited by 333 · cited 1×
“In line with other digital mental health tools, it appears that the 24/7 access can be a key factor for high user engagement, particularly among those who avoid traditional in-person therapy due to logistical constraints or concerns about stigma [4,29,30].”
The Relationship Between Homework Compliance and Therapy Outcomes: An Updated Meta-Analysis
Mausbach, B. T., Moore, R., Roesch, S., Cardenas, V., & Patterson, T. L. (2010) · Cognitive Therapy and Research · doi:10.1007/s10608-010-9297-z · Cited by 309
Perceptions and opinions of patients about mental health chatbots: scoping review
Abd-Alrazaq, A. A., Alajlani, M., Ali, N., Denecke, K., Bewick, B. M., & Househ, M. (2021) · Journal of Medical Internet Research · doi:10.2196/17828 · Cited by 301 · cited 2×
“It is plausible that the human-like responses that LLMs produce lead to a greater sense of being understood compared to rule-based algorithms where responses often did not feel tailored to the individual [3].”
Kantor, V., Knefel, M., & Lueger-Schuster, B. (2017) · Clinical Psychology Review · doi:10.1016/j.cpr.2016.12.001 · Cited by 294 · cited 1×
“In line with other digital mental health tools, it appears that the 24/7 access can be a key factor for high user engagement, particularly among those who avoid traditional in-person therapy due to logistical constraints or concerns about stigma [4,29,30].”
Kazantzis, N., Whittington, C., Zelencich, L., Kyrios, M., Norton, P. J., & Hofmann, S. G. (2016) · Behavior Therapy · doi:10.1016/j.beth.2016.05.002 · Cited by 280 · cited 1×
“However, worksheets lack the interactivity and immediate feedback patients may need to remain engaged and may at least partially explain the high homework noncompletion rates that have been observed [11,12].”
Andersen, F. A. (1999) · International Journal of Toxicology · doi:10.1177/109158189901800306 · Cited by 258
The alliance in mental health care: conceptualization, evidence and clinical applications
Wampold, B. E., & Flückiger, C. (2023) · World Psychiatry · doi:10.1002/wps.21035 · Cited by 213 · cited 1×
“To optimize future tools, research should compare theorized therapeutic mechanisms, such as the working alliance [31], between algorithms to identify the exact mechanisms that contribute to positive experiences and treatment outcomes.”
Developing an optimal short-form of the PTSD Checklist for DSM-5 (PCL-5)
Zuromski, K. L., Ustun, B., Hwang, I., Keane, T. M., Marx, B. P., Stein, M. B., Ursano, R. J., & Kessler, R. C. (2019) · Depression and Anxiety · doi:10.1002/da.22942 · Cited by 206 · cited 1×
“…included the Patient Health Questionnaire-4 [15], Interaction Anxiousness Scale-3 (IAS-3) [16], Posttraumatic Stress Disorder (PTSD) Checklist-4 [17], Obsessive-Compulsive Inventory-4 [18], and Washington Early Recognition Center Affectivity and Psychosis Screen [19], and asked to participate…”
Clark, D. A. (2013) · The Wiley Handbook of Cognitive Behavioral Therapy · doi:10.1002/9781118528563.wbcbt02 · Cited by 192
Encyclopedia of Cognitive Behavior Therapy
Freeman, A., Reinecke, M. A., Nezu, C. M., Nezu, A. M., & Felgoise, S. H. (2005) · Springer eBooks · doi:10.1007/b99240 · Cited by 118
Homework Compliance Counts in Cognitive-Behavioral Therapy
LeBeau, R. T., Davies, C. D., Culver, N. C., & Craske, M. G. (2013) · Cognitive Behaviour Therapy · doi:10.1080/16506073.2013.763286 · Cited by 105 · cited 1×
“However, worksheets lack the interactivity and immediate feedback patients may need to remain engaged and may at least partially explain the high homework noncompletion rates that have been observed [11,12].”
Generative artificial intelligence in mental health care: potential benefits and current challenges
Torous, J., & Blease, C. (2024) · World Psychiatry · doi:10.1002/wps.21148 · Cited by 90 · cited 2×
“Numerous LLM-based therapy tools are currently in development [7], but rigorous safety and feasibility data are needed before such tools can be widely adopted [4,5].”
Bunyi, J., Ringland, K. E., & Schueller, S. M. (2021) · Frontiers in Digital Health · doi:10.3389/fdgth.2021.742196 · Cited by 82 · cited 2×
“Digital mental health interventions offer promising avenues to deliver support, overcome wait times, and meet a wide variety of patient preferences [1, 2].”
Geier, T. J., Hunt, J. C., Hanson, J. L., Heyrman, K., Larsen, S. E., Brasel, K. J., & deRoon‐Cassini, T. A. (2020) · Journal of Traumatic Stress · doi:10.1002/jts.22478 · Cited by 75 · cited 1×
“…indicating probable clinical range are approximately ≥6 on the Patient Health Questionnaire-4 for depression or anxiety [15], ≥10 on the IAS-3 for social anxiety [16], ≥4 on the PTSD Checklist-4 for PTSD [20], and ≥4 on the Obsessive-Compulsive Inventory-4 for obsessive-compulsive symptoms [21].”
The Challenges in Designing a Prevention Chatbot for Eating Disorders: Observational Study
Chan, W. W., Fitzsimmons‐Craft, E. E., Smith, A. C., Firebaugh, M., Fowler, L. A., DePietro, B., Topooco, N., Wilfley, D. E., Taylor, C. B., & Jacobson, N. C. (2022) · JMIR Formative Research · doi:10.2196/28003 · Cited by 72
King, D. R., Nanda, G., Stoddard, J., Dempsey, A., Hergert, S., Shore, J. H., & Torous, J. (2023) · Current Psychiatry Reports · doi:10.1007/s11920-023-01477-x · Cited by 63 · cited 1×
“These insights contrast common fears associated with the use of AI in the context of mental health care, where privacy is commonly listed as a key issue [5,6].”
Stirman, S. W., Gutner, C. A., Suvak, M. K., Adler, A., Calloway, A., & Resick, P. (2018) · Behavior Therapy · doi:10.1016/j.beth.2017.12.001 · Cited by 53 · cited 1×
“Outside of therapy sessions, patients are assigned to complete worksheets designed to facilitate similar cognitive reappraisal processes [9,10].”
Mamah, D. T., Owoso, A., Sheffield, J. M., & Bayer, C. (2014) · Comprehensive Psychiatry · doi:10.1016/j.comppsych.2014.07.004 · Cited by 42 · cited 1×
“…Stress Disorder (PTSD) Checklist-4 [17], Obsessive-Compulsive Inventory-4 [18], and Washington Early Recognition Center Affectivity and Psychosis Screen [19], and asked to participate in a short semistructured interview that assessed for their familiarity with AI in mental health care.”
The OCI-4: An ultra-brief screening scale for obsessive-compulsive disorder
Abramovitch, A., Abramowitz, J. S., & McKay, D. (2021) · Journal of Anxiety Disorders · doi:10.1016/j.janxdis.2021.102354 · Cited by 37 · cited 1×
“…[15], Interaction Anxiousness Scale-3 (IAS-3) [16], Posttraumatic Stress Disorder (PTSD) Checklist-4 [17], Obsessive-Compulsive Inventory-4 [18], and Washington Early Recognition Center Affectivity and Psychosis Screen [19], and asked to participate in a short semistructured interview that…”
Digital mental health care: five lessons from Act 1 and a preview of Acts 2–5
Insel, T. (2023) · npj Digital Medicine · doi:10.1038/s41746-023-00760-8 · Cited by 35 · cited 2×
“Digital mental health interventions offer promising avenues to deliver support, overcome wait times, and meet a wide variety of patient preferences [1, 2].”
Held, P., Pridgen, S. A., Chen, Y., Akhtar, Z., Amin, D., & Pohorence, S. (2024) · JMIR Research Protocols · doi:10.2196/58195 · Cited by 28 · cited 4×
“As described in a study by Held et al [13], the AI interface also displayed safety notices, reminding participants of the nonclinical nature of the tool and providing information on how to seek professional help if needed.”
InfluentialNichols, A. L., & Webster, G. D. (2015) · Personality and Individual Differences · doi:10.1016/j.paid.2015.01.043 · Cited by 22
Hsieh, C. J., Godwin, D., & Mamah, D. T. (2016) · Frontiers in Psychiatry · doi:10.3389/fpsyt.2016.00149 · Cited by 21
Carey, C. (2002) · Routledge eBooks · doi:10.4324/9780203130476-6 · Cited by 10
Abramowitz, J. S., Myers, N. S., Friedman, J. B., Juel, E. K., Nestadt, G., Kimmel, M., Osborne, L. M., Storch, E. A., SAMUELS, J. F., Nestadt, P. S., et al. (2024) · Archives of Women s Mental Health · doi:10.1007/s00737-024-01539-w · Cited by 6
Held, P., Pridgen, S. A., Szoke, D. R., Chen, Y., Akhtar, Z., & Amin, D. (2025) · Cognitive and Behavioral Practice · doi:10.1016/j.cbpra.2025.10.001 · Cited by 2 · cited 2×
“Clinicians remarked that the AI interface could make conventional CBT homework feel more approachable and interactive [14], but data from end users (eg, patients) are needed.”
InfluentialSelf-citation
Cited by
Shankar, R., Lim, A., & Xu, Q. (2026) · Healthcare · doi:10.3390/healthcare14101267 · Cited by 9
Szoke, D. R., Hutzler, I., Liu, J., Addante, S., Akhtar, Z., Smith, D. L., Dickens, K., Small, C., Pridgen, S., & Held, P. (2026) · JMIR Mental Health · doi:10.2196/91367 · Cited by 5
“Each clinician had experience providing talk therapy and reviewing real human user–AI therapist mental health transcripts from a previous study [29].”
Self-citationRidings, L. E., Held, P., Kuhn, E. R., & Stirman, S. W. (2026) · Journal of Traumatic Stress · doi:10.1002/jts.70092 · Cited by 2
“…including the delivery of psychoeducation (Yilanli et al., 2024), behavior and emotion identification (Chen et al., 2022; Shah et al., 2022; Tanana et al., 2021), mental health risk detection (Bantilan et al., 2021), and the delivery of specific therapeutic components (Held et al., 2025a).”
Self-citationHeld, P., Smith, D. L., Pridgen, S., Hutzler, I., & Szoke, D. R. (2026) · doi:10.2196/preprints.108739
Self-citationAkbarzadeh, M. M., & Akbarzadeh, M. (2026) · International Journal of Human-Computer Interaction · doi:10.1080/10447318.2026.2735780 · Cited by 1
Metadata
- Authors
- Held, P., Pridgen, S. A., Szoke, D. R., Chen, Y., Akhtar, Z., & Amin, D.
- Affiliations
- Rush University Medical Center (US)
- Corresponding author
- Held, Philip
- Venue
- JMIR Mental Health
- Publisher
- JMIR Publications Inc.
- Published
- 2025-12-05
- Volume
- 12
- Pages
- e80461-e80461
- ISSN
- 2368-7959 (ISSN-L)
- PMC
- PMC12680128
- OpenAlex
- W4415584816
- Semantic Scholar
- a27dbb6c66ac93740e2e8df517863ec359ee7adf
- Cited by
- 7
- References
- 29
- Citations by year
- 2026: 7
- FWCI
- 5.59
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- Locations
- JMIR Mental Health (publishedVersion, cc-by)
- PubMed (publishedVersion)
- PubMed Central (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Social Sciences › Psychology › Applied Psychology › Digital Mental Health Interventions
- Social Sciences › Psychology › Experimental and Cognitive Psychology › Anxiety, Depression, Psychometrics, Treatment, Cognitive Processes
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Keywords
- cognitive reappraisal, Socratic dialogue, large language models, depression, social anxiety, post-traumatic stress, obsessive-compulsive symptoms, therapeutic relationship, acceptability, feasibility, usability, user experience, digital mental health tools
- MeSH terms
- Adult, Artificial Intelligence, Depression, Feasibility Studies, Female, Humans, Male, Middle Aged, Stress Disorders, Post-Traumatic, Cognitive Behavioral Therapy, Young Adult
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.