Skip to main content
NCBI home page
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2025 Jul 30;38(5):813–820. doi: 10.1002/jts.23188

Generative artificial intelligence in posttraumatic stress disorder treatment: Exploring five different use cases

Philip Held 1,✉, Elizabeth C Stade 2, Katherine Dondanville 3, Shannon Wiltsey Stirman 4,5
PMCID: PMC12551617  PMID: 40736259

Abstract

Posttraumatic stress disorder (PTSD) is a prevalent and debilitating condition, yet many individuals face substantial barriers to accessing evidence‐based interventions. Advances in generative artificial intelligence (AI), particularly large language models (LLMs), have generated optimism about improving access and care. We present five emerging use cases for clinical AI tools in the context of PTSD treatment, some of which were presented as part of a symposium at the 40th Annual Meeting of the International Society for Traumatic Stress Studies. The first two use cases involve AI‐assisted training tools. The third use case focuses on an AI‐assisted automated fidelity rating system aimed at improving adherence to evidence‐based PTSD protocols. The last two use cases feature AI‐assisted therapy tools. Although AI‐based innovations hold the promise of enhancing the reach and consistency of evidence‐based PTSD interventions, they also raise important ethical and safety challenges, including risk of bias, threats to patient privacy, and the question of how to incorporate clinical oversight. Ongoing collaboration among multidisciplinary teams involving clinicians, researchers, and technology developers will be essential to ensuring that AI tools remain patient‐centered, ethically sound, and effective.

Trauma‐focused interventions such as cognitive processing therapy (CPT; Resick et al., 2024), prolonged exposure (PE; Foa et al., 2019), and written exposure therapy (WET; Sloan & Marx, 2025 ) have demonstrated short‐ and long‐term effectiveness for the treatment of posttraumatic stress disorder (PTSD; U.S. Department of Veterans Affairs [VA]/Department of Defense [DoD], 2023). However, logistical barriers, financial constraints, and clinician shortages can limit access to these interventions (Coombs et al., 2021; Kantor et al., 2017). In recent years, advances in generative artificial intelligence (AI), which uses large‐scale computational models to generate human‐like text, images, and other outputs, have sparked hopes of addressing these gaps in mental health care (Torous & Blease, 2024). In particular, large language models (LLMs) can emulate nuanced human dialogue and may be well‐suited to provide on‐demand therapeutic support as well as to support clinicians and potentially enhance existing PTSD treatments (Li et al., 2023; Stade, Stirman, Ungar, et al., 2024).

Individuals are already engaging with commercially available LLMs (e.g., ChatGPT, Gemini) to seek guidance on mental health concerns and, in some cases, to simulate therapeutic interactions. These “foundation” or general‐purpose models are trained on vast amounts of internet data, which enables them to generate human‐like responses based on language patterns. Foundation models are not explicitly designed for clinical purposes, although some of their training data may contain mental health–related information. When used in mental health contexts, available LLMs typically issue disclaimers or offer general referrals. In contrast, purpose‐built, clinically oriented AI tools differ significantly from commercially available general‐purpose LLMs. Although these tools may leverage foundation models (e.g., ChatGPT4o), they are intentionally developed, prompt engineered, and validated for mental health training and interventions. Specifically, these specialized tools integrate specific therapeutic frameworks, adhere to clinical guidelines, and continuously undergo rigorous validation and refinement by individuals with content expertise to ensure the tools appropriately address sensitive clinical content, manage nuanced psychological interactions, and follow cultural and ethical considerations. It is crucial to systematically evaluate both general‐purpose AI tools and purpose‐built, clinically informed AI solutions to better understand their respective strengths and limitations.

We showcase five emerging use cases for clinical AI tools in the context of PTSD treatment, some of which were presented as part of a symposium at the 40th Annual Meeting of the International Society for Traumatic Stress Studies. We briefly present each use case, discuss the potential benefits and limitations of each, and conclude with broader considerations for integrating AI into trauma‐focused interventions. The first two use cases involve AI‐assisted training tools for clinicians learning WET and Socratic dialogue. The third use case focuses on an AI‐assisted automated fidelity rating system aimed at improving adherence to evidence‐based PTSD protocols. The last two use cases feature AI‐assisted therapy tools, such as an AI Worksheet Helper to assist with CPT homework completion and Socrates 2.0, a multiagent AI tool that aims to help engage patients in Socratic dialogue. Although these examples are far from exhaustive, they illustrate how generative AI could help clinicians learn key elements of evidence‐based therapy, improve treatment fidelity, and enable more flexible patient support outside of therapy sessions. The tools presented here are all based on foundation LLMs and are prompt engineered for specific use cases. They predominantly use text‐based and speech‐to‐text interactions. It is important to note that other clinical AI tools not presented here extend beyond LLM‐based applications alone; emerging solutions increasingly combine AI with other innovative treatment modalities, such as augmented reality exposure therapy, to enhance trauma‐focused treatment engagement, effectiveness, and accessibility (Javanbakht et al., 2024).

AI‐ASSISTED TRAINING TOOLS

Generative AI can play an important role in clinician training by bridging the gap between limited peer‐based role‐plays and the real‐world demands of patient care. Studies have shown that therapists’ first cases, on average, have lower levels of symptom reduction compared to subsequent cases (Johnson et al., 2019; Mallard Swanson et al., 2021). Additionally, many therapists are hesitant to begin providing trauma‐focused treatments to their patients, and without additional support, they may be less likely to offer these treatments to patients after the initial training period (Monson et al., 2018). There is some evidence that practice and support, such as consultation, may increase provider confidence and lead to better clinical outcomes, particularly for therapists who endorse lower self‐efficacy in providing a trauma‐focused treatment (Pace et al., 2021).

By simulating a wide range of clinical interactions on demand, AI‐based tools can reduce training costs and time investments, particularly for underresourced clinics or busy practitioners (Stade, Stirman, Ungar, et al., 2024). Moreover, some clinicians lack protected time to engage in standard consultation even though research has demonstrated the importance of postworkshop support for both the adoption of evidence‐based treatments and improved outcomes (Foa et al., 2020; Monson et al., 2018). AI‐assisted training tools may enable clinicians who are not able to attend consultations to get additional practice and automated feedback. These platforms can offer consistent, standardized practice scenarios that can be flexibly tailored to individual learning needs while providing clinicians with repeated exposure to difficult conversations without risking patient well‐being. In the context of trauma‐focused treatments, enhancing provider competencies is crucial, and generative AI‐driven training can help ensure clinicians gain both foundational skills and confidence before working directly with individuals experiencing PTSD and related mental health concerns. Simulation with immediate AI‐generated feedback has the potential to foster skill development and self‐efficacy while minimizing the impact of missteps and lack of practice (Stade, Eichstaedt, et al., 2024).

One such AI‐assisted training tool is Socrates Coach, which was designed to help clinicians develop and refine Socratic dialogue skills. Socratic dialogue is closely linked to clinical outcomes (Farmer et al., 2017) but can be a challenging skill to master and can lead to awkward clinical interactions until skill and comfort increase (Waltman et al., 2017). Building on advances from patient‐facing tools like Socrates 2.0, Socrates Coach focuses on training clinicians in a digital environment where they can repeatedly practice Socratic dialogue with AI‐simulated patients. These simulated patients, whose responses are generated by LLMs, can be tailored to varying difficulties and clinical presentations; clinicians also have the option to create custom AI patients akin to character design in role‐playing games to meet individual training needs. In addition, three distinct AI “supervisors” review each response, sections of the conversation, and the entire dialogue, respectively. The AI supervisors provide nuanced real‐time feedback on alternative questions clinicians may ask, possible directions for the conversation, and the clinicians’ overall strengths and areas for growth to improve clinicians’ Socratic dialogue abilities.

Whereas Socrates Coach focuses on a specific therapeutic skill, there is also a benefit to allowing practice and feedback on a full range of skills between a provider's initial training and the initiation of a new therapy modality with patients. One such effort, TherapyTrainer (Stade, Eichstaedt, et al., 2024), is a multiagent LLM that is prompt engineered by individuals with expertise in WET and includes input from WET consultants and clinicians. TherapyTrainer provides opportunities to practice full WET protocols, as well as specific elements, with two simulated patients with different backgrounds and traumatic experiences. Clinicians receive real‐time, AI‐generated feedback on each skill and can ask the LLM “consultant” questions when they need guidance. Subsequent versions will also include additional “challenge” scenarios that allow therapists to practice addressing high‐risk or challenging situations. Clinicians can practice at their own pace, but given the relative brevity of the WET protocol and the amount of session time that the patient spends writing the narrative relative to the time spent in conversation, it is possible to simulate full protocols in less time than would be required for consultation on full training cases with real patients. Planned research will compare post‐WET workshop TherapyTrainer to workshop alone and workshop with consultation conditions to determine whether simulation and feedback can yield similar or better results than these more standard forms of training.

Despite the benefits of AI‐assisted training tools like Socrates Coach and TherapyTrainer, these technologies currently face notable challenges. Although LLM‐generated dialogues can offer diverse training scenarios, they may not fully capture the nuanced emotional and relational dynamics of genuine patient encounters, particularly in high‐stakes situations. For example, situations in which patients express suicidal ideation or disclose trauma or abuse require therapists to demonstrate empathic attunement, immediate and appropriate emotional responsiveness, ethical decision‐making, and adaptive relational skills. AI tools imitating human patients may lack the ability to fully replicate the emotionality and high‐stakes nature of such situations accurately.

In addition, ongoing audits of training data and dialogue outputs are essential to ensure cultural relevance and reduce potential biases. By systematically reviewing training data to ensure the representation of diversity, including the inclusion of dialogues reflecting varied cultural backgrounds, language nuances, and communication styles, among others, culturally diverse groups of clinicians and patient advocates can help identify and address unintended stereotypes or biases.

Finally, AI alone cannot determine clinicians’ readiness to apply these skills with real patients. Thus, these tools should not be used for this purpose; experienced human supervisors, informed by objective assessments of skill, must play a pivotal role in assessing competence. The supervisory process could involve structured observation sessions, such as role‐play scenarios with standardized patients or supervised interactions with real patients in clinical practicum settings. Supervisors could systematically use current best practices (e.g., structured observation checklists, fidelity rating scales, or competency‐based rubrics) to evaluate critical therapist behaviors, interpersonal skills, clinical judgment, and therapeutic responsiveness. The time investment required could be managed by initially leveraging AI training to bring trainees to a baseline level of competence and flagging areas of strength and growth opportunities, thereby allowing supervisors to focus targeted, shorter observations on specific clinical skills or challenging therapeutic scenarios. Rather than replacing observation, this hybrid approach would use AI tools to build trainee confidence, highlight specific strengths and weaknesses through performance dashboards, and streamline readiness determinations, reserving supervisor time for the most meaningful, skill‐relevant assessments. Through such a hybrid approach, supervisors could efficiently determine clinician readiness and assess quality without significantly offsetting the time‐saving benefits offered by AI‐assisted initial training. With thoughtful integration and attention to these concerns, tools like Socrates Coach, TherapyTrainer, and others have the potential to fill an important gap in therapist training. Ultimately, such tools may help enhance the overall quality of care provided to patients with PTSD and other mental health conditions.

AI‐assisted training tools will need to be rigorously evaluated to determine whether they can actually increase trainees’ or therapists’ confidence and skills, as well as whether their implementation can reduce training, supervision, or consultation costs. It will also be imperative to ensure that such tools are designed to teach therapists how to think about clinical situations rather than simply providing them with answers. Therapists will need to be equipped to handle challenging scenarios even when AI tools are not available to them. Although concerns about overreliance are valid, it may be argued that real‐time feedback may not substantially differ from traditional consultation. Future research should examine whether real‐time AI feedback can support skill development through ongoing reflection and deliberate practice.

AI‐ASSISTED TREATMENT FIDELITY RATING TOOLS

Another use case of AI‐based tools involves their application in the rating of providers’ fidelity to different treatments, such as CPT. Therapist fidelity (i.e., the degree to which providers adhere to the structure, techniques, and elements of a manualized therapy) is widely recognized as a key determinant of treatment outcomes for PTSD (Farmer et al., 2017; Stirman et al., 2021). In CPT, fidelity typically involves delivering core components, such as psychoeducation; facilitating cognitive restructuring and emotional processing; and assigning relevant homework (Resick et al., 2024; Stirman et al., 2021). Despite its importance, fidelity monitoring can be resource‐intensive, as it requires trained coders to review session transcripts or recordings (Stirman et al., 2021).

An AI fidelity rating system prototype was designed to review text‐based therapy sessions and classify each therapist message according to fidelity‐relevant categories, such as Socratic dialogue or assigning homework. In contrast to earlier‐generation machine learning–based programs, which demonstrate high reliability but require training with very large sets of human‐coded data, LLM‐based models can be trained using manuals and codebooks, and then refined through prompt engineering (Lenton‐Brym et al., 2025). Early efforts to use this prototype highlighted the importance of examining larger exchanges to enable the LLM to detect context and transitions from one therapeutic activity to another. For instance, a therapist might shift from explaining a concept (i.e., psychoeducation) to reviewing related homework within the same session; without contextual information for cases in which each statement is reviewed independently, these two actions could be mislabeled as the same category.

Once validated at scale, automated fidelity rating tools, such as the AI Fidelity Rating System prototype, could provide near–real‐time feedback to clinicians (Lenton‐Brym et al., 2025). In the future, such tools could alert clinicians to potential deviations from the treatment manual. Alerts could also be combined with feedback about how to increase fidelity using the same LLM or another, connected AI agent. It is easy to envision how such tools could also be expanded to include audio or video recordings of sessions, potentially providing even more nuanced fidelity ratings given the availability of multimodal data that can be evaluated. This approach has the potential to substantially reduce both the time and cost associated with coding therapy sessions for fidelity and providing actionable feedback, which can ultimately improve outcomes. Despite the potential promises of AI fidelity rating systems, it will be critical to ensure the accuracy of both the fidelity ratings and potential corrective feedback.

AI‐ASSISTED THERAPY TOOLS

In addition to supporting clinician training and assessing treatment fidelity, AI tools may also be used to enhance existing treatment approaches by either facilitating specific skills or supporting important treatment elements. One such example, Socrates 2.0, a multiagent AI tool, was designed to engage patients in Socratic dialogue outside of session and reinforce cognitive reappraisal outside of therapy (Held et al., 2024). Socratic dialogue is a hallmark component of evidence‐based psychotherapies for PTSD and other conditions (Braun et al., 2015; Freeman, 2005; Resick et al., 2024). In Socratic dialogue, clinicians ask strategic questions to facilitate cognitive reappraisal by encouraging patients to examine the evidence for and against their beliefs. Despite its effectiveness in facilitating symptom change (Braun et al., 2015; Vittorio et al., 2022), it can be challenging to replicate this process independently between sessions. Socrates 2.0 addresses this issue through the use of multiple AI “agents” that work collaboratively in the background to provide the patient with high‐quality Socratic dialogue. The primary AI agent functions like a virtual therapist, posing questions in a conversational style. A second, hidden AI agent supervises the AI therapist by providing guidance for how to improve the dialogue, similar to a human supervisor with bug‐in‐ear supervisory capabilities. A third, also hidden AI agent evaluates the user's belief strength over time and suggests to the AI therapist when to change the line of questions in cases where no change is detected or to stop the interaction once belief change is observed. Research has shown that the combination of different collaborative AI agents can help improve the quality of responses (Talebirad & Nadiri, 2023), which may be especially important in sensitive therapeutic contexts in which high quality responses are paramount. During development and early feasibility work, the combination of multiple AI agents in Socrates 2.0 was perceived to have improved response quality and improved safety, such as an increased ability to stay on track (Held et al., 2024).

Feasibility data from 55 users and eight clinicians who interacted with Socrates 2.0 suggest high satisfaction and feasibility. Users appreciated the tool's constant availability and generally found the tool helpful in nonjudgmentally examining their beliefs. Moreover, although it was not the main focus of the feasibility study, users reported moderate reductions in PTSD and depressive symptoms over the course of the 4‐week study period. Participants also shared improvement‐oriented feedback and remarked on the tool's limited functionality of only asking Socratic questions. These early results highlight both the potential and limitations of AI‐driven Socratic dialogue and how tools like Socrates 2.0 could be used to facilitate specific interventions. Going forward, it will be important to determine how to optimally integrate tools like Socrates 2.0 with first‐line PTSD treatments and to conduct large randomized controlled trials to evaluate the effectiveness of these tools. It will also be important to continue to evaluate these tools’ ability to provide culturally and contextually appropriate responses for a diverse range of users.

Another core component of CPT and other cognitive behavioral therapies is the assignment and completion of homework assignments, which encourage patients to practice skills learned in‐session on their own time (Cooper et al., 2017; Resick et al., 2024). However, many individuals struggle to complete these assignments effectively for reasons including confusion about instructions, an insufficient grasp of important therapy concepts, or a lack of immediate feedback (Helbig & Fehm, 2004; Kazantzis et al., 2016; Kazantzis & Shinkfield, 2007). These challenges can result in suboptimal practice and can delay skill acquisition until the next therapy session when patients can receive feedback. Difficulties with homework completion have been shown to be associated with diminished treatment outcomes (Kazantzis et al., 2016; Stirman et al., 2018).

The AI‐assisted CPT Worksheet Helper is intended to address some of the aforementioned challenges. The tool was designed to review or guide patients through the completion of CPT worksheets, such as the ABC (i.e., “activating event, belief, consequence”) worksheet. An underlying LLM automatically analyzes user input to identify common mistakes, suggests clarification, and offers improvement‐oriented feedback that mirrors what a therapist might provide during an in‐session review of the therapy worksheets. For example, if a patient mistakenly enters beliefs in the “feelings” column, the CPT Worksheet Helper encourages them to differentiate emotions from cognitions. Such immediate, personalized guidance can bolster patient motivation, reduce confusion, and potentially accelerate the development of more adaptive thinking patterns, ultimately accelerating and improving treatment outcomes.

Despite the promise of real‐time support, it will be essential to build assistive tools like the CPT Worksheet Helper to ensure that patients do not become overreliant on AI, such as by simply following AI suggestions rather than actively acquiring key therapeutic skills. Future research should, therefore, evaluate the type and timing of AI‐generated feedback to determine how to optimize skill acquisition and retention. A personalized approach may be most effective, as feedback preferences are likely to differ. Whereas some patients might desire frequent guidance, others who have an advanced understanding of CPT may prefer more occasional support or review of completed worksheets rather than feedback during the exercise.

DISCUSSION

The five use cases presented here illustrate different ways in which generative AI might enhance the reach, consistency, and impact of PTSD treatments in the future. It will be imperative to integrate input from both patients and clinicians in the design process. User‐centered design processes, such as iteration based on regular feedback and codesign sessions, can help make AI tools clinically robust and user‐friendly and ensure the tools align with the established principles of evidence‐based interventions. Moreover, involving different stakeholders in usability testing throughout the development cycle can help identify potential pitfalls early and ensure that the clinical AI tools are not only technically sound but also responsive to real‐world needs.

Despite the advancements in AI and AI's potential to improve the treatment of PTSD and other mental health concerns, including initial research and development of direct‐to‐consumer LLMs that emulate psychotherapy, it is critical to take a cautious and phased approach to the use of AI in clinical care (Stade, Eichstaedt, et al., 2024). There are a number of issues that need to be top‐of‐mind for individuals developing, evaluating, and using such tools. First, clinical safety must be a top priority. AI systems intended for direct patient use require robust protocols for detecting concerning levels of distress and must have the ability to involve a skilled human provider as needed (Stade, Stirman, Ungar, et al., 2024; Stade et al., 2025). This may be challenging at times, as one of the stated benefits of using AI tools is their constant availability, whereas clinician availability after hours is limited and may require the use of crisis hotlines. Users should be explicitly made aware of the intended use and the known limitations of the available tools (e.g., no real‐time monitoring by human clinicians, inability to perform crisis intervention). Second, ethical and legal ambiguities persist. For example, liability‐related questions when using AI‐generated suggestions may influence clinical decisions. Third, purpose‐built therapeutic AI tools must specifically address the tendency of general‐purpose LLMs to produce responses primarily aimed at satisfying user preferences or expectations rather than prioritizing factual accuracy or clinical appropriateness. Purpose‐built AI tools used in mental health contexts should, thus, be designed to consistently prioritize clinical accuracy, safety, and appropriateness over user satisfaction alone and should be protected against user modification or undue influence. Additionally, transparent documentation of the development and validation processes of these AI tools is crucial. Such documentation should include the initial model training (e.g., prompt engineering and/or training therapeutic materials, such as treatment manuals, clinical vignettes, or annotated psychotherapy transcripts) and validation processes (e.g., assessing accuracy and appropriateness through expert clinician testing of prototypes and iterative feedback loops to improve tools). The aforementioned issues need to be resolved, and users of therapeutic AI tools (both clinicians and patients) need to be educated on any limitations and liabilities and explicitly consent to the use of such tools. Lastly, protecting patient confidentiality through encryption, secure data storage, and compliance with privacy regulations is paramount (Torous & Blease, 2024).

Another challenge is the potential for AI models to perpetuate biases contained in training datasets (Torous & Blease, 2024). Developers and researchers must actively scrutinize, or “red team,” the quality of these tools to ensure that different users are not subjected to disproportionately negative, insensitive, or low‐quality responses and that users receive culturally and contextually appropriate responses. Finally, even in light of the availability of increasingly sophisticated clinical AI tools, skilled trauma clinicians are still critically needed. Just as a pilot may rely on autopilot software for most of the flight but needs to be skilled enough to take over as needed, clinicians must be able to interpret AI outputs, monitor for potential errors, and play a key role in developing and maintaining the therapeutic alliance essential to effective treatment. Patient education surrounding the capabilities, limitations, and intended role of these tools is also critical, and it will be important to help individuals understand that AI tools are generally intended to augment, but not replace, existing human‐delivered PTSD treatments (Stade, Stirman, Held, et al., 2024).

It will be imperative for clinicians, researchers, ethicists, policymakers, and technology developers to collaborate closely during the development and evaluation of clinical generative AI tools that are intended to be used in the treatment of PTSD and other mental health concerns (Stade et al., 2025). When designed thoughtfully, these tools may address long‐standing barriers to mental health care and offer renewed hope for individuals with PTSD. Eventually, clinics might designate “AI champions” who have in‐depth knowledge of different AI tools’ features, limitations, and data security protocols; these champions can train colleagues and troubleshoot technical issues. Real‐time feedback from both clinicians and patients can guide iterative refinements, ensuring the system remains user‐friendly, clinically relevant, and compliant with privacy standards. Over time, insights from these pilot projects can form the basis of evidence‐based guidelines for wider implementation.

AUTHOR NOTE

Philip Held receives grant support from the Department of Defense (W81XWH‐22‐1‐0739; HT9425‐24‐1‐0666; HT9425‐24‐1‐0637), Wounded Warrior Project, United Services Automobile Association (USAA)/Face the Fight, and the Crown Family Foundation. Shannon Wiltsey Stirman and Katherine Dondanville receive grant support from the National Institute of Mental Health (NIMH; RF1 MH 128785), which supported some of the work discussed herein. Elizabeth Stade has performed advising work for mental health companies using artificial intelligence (AI): Jimini Health, Sonar Mental Health, and Sonia Health.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the Department of Defense, Wounded Warrior Project, the U.S. Department of Veterans Affairs, USAA, NIMH, or any other funding agency. All authors declare that they have no competing interests.

Held, P. , Stade, E. C. , Dondanville, K. , & Wiltsey Stirman, S. (2025). Generative artificial intelligence in posttraumatic stress disorder treatment: Exploring five different use cases. Journal of Traumatic Stress, 38, 813–820. 10.1002/jts.23188

REFERENCES

  1. Braun, J. D. , Strunk, D. R. , Sasso, K. E. , & Cooper, A. A. (2015). Therapist use of Socratic questioning predicts session‐to‐session symptom change in cognitive therapy for depression. Behaviour Research and Therapy, 70, 32–37. 10.1016/j.brat.2015.05.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Coombs, N. C. , Meriwether, W. E. , Caringi, J. , & Newcomer, S. R. (2021). Barriers to healthcare access among U.S. adults with mental health challenges: A population‐based study. SSM ‐ Population Health, 15, Article 100847. 10.1016/j.ssmph.2021.100847 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Cooper, A. A. , Kline, A. C. , Graham, B. , Bedard‐Gilligan, M. , Mello, P. G. , Feeny, N. C. , & Zoellner, L. A. (2017). Homework “dose,” type, and helpfulness as predictors of clinical outcomes in prolonged exposure for PTSD. Behavior Therapy, 48(2), 182–194. 10.1016/j.beth.2016.02.013 [DOI] [PubMed] [Google Scholar]
  4. Farmer, C. C. , Mitchell, K. S. , Parker‐Guilbert, K. , & Galovski, T. E. (2017). Fidelity to the cognitive processing therapy protocol: Evaluation of critical elements. Behavior Therapy, 48(2), 195–206. 10.1016/j.beth.2016.02.009 [DOI] [PubMed] [Google Scholar]
  5. Foa, E. B. , Hembree, E. A. , Rothbaum, B. O. , & Rauch, S. A. M. (2019). Prolonged exposure therapy for PTSD: Emotional processing of traumatic experiences: Therapist guide (2nd ed). Oxford University Press. [Google Scholar]
  6. Foa, E. B. , McLean, C. P. , Brown, L. A. , Zang, Y. , Rosenfield, D. , Zandberg, L. J. , Ealey, W. , Hanson, B. S. , Hunter, L. R. , Lillard, I. J. , Patterson, T. J. , Rosado, J. , Scott, V. , Weber, C. , Wise, J. E. , Zamora, C. D. , Mintz, J. , Young‐McCaughan, S. , Peterson, A. L. , & for the STRONG STAR. Consortium . (2020). The effects of a prolonged exposure workshop with and without consultation on provider and patient outcomes: A randomized implementation trial. Implementation Science, 15(1), Article 59. 10.1186/s13012-020-01014-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Freeman, A. (2005). Socratic dialogue. In Freeman A., Felgoise S. H., Nezu C. M., Nezu A. M., & Reinecke M. A. (Eds.), Encyclopedia of cognitive behavior therapy (pp. 380–384). Springer. 10.1007/0-306-48581-8_106 [DOI] [Google Scholar]
  8. Helbig, S. , & Fehm, L. (2004). Problems with homework in CBT: Rare exception or rather frequent? Behavioural and Cognitive Psychotherapy, 32(3), 291–301. 10.1017/S1352465804001365 [DOI] [Google Scholar]
  9. Held, P. , Pridgen, S. A. , Chen, Y. , Akhtar, Z. , Amin, D. , & Pohorence, S. (2024). Mixed method feasibility study protocol for Socrates 2.0: A novel cognitive behavioral therapy‐based generative ai tool to facilitate Socratic dialogue. JMIR Research Protocols, 13, Article e58195. 10.2196/58195 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Javanbakht, A. , Hinchey, L. , Gorski, K. , Ballard, A. , Ritchie, L. , & Amirsadri, A. (2024). Unreal that feels real: Artificial intelligence‐enhanced augmented reality for treating social and occupational dysfunction in post‐traumatic stress disorder and anxiety disorders. European Journal of Psychotraumatology, 15(1), Article 2418248. 10.1080/20008066.2024.2418248 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Johnson, J. E. , Stout, R. L. , Miller, T. R. , Zlotnick, C. , Cerbo, L. A. , Andrade, J. T. , Nargiso, J. , Bonner, J. , & Wiltsey‐Stirman, S. (2019). Randomized cost‐effectiveness trial of group interpersonal psychotherapy (IPT) for prisoners with major depression. Journal of Consulting and Clinical Psychology, 87(4), 392–406. 10.1037/ccp0000379 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Kantor, V. , Knefel, M. , & Lueger‐Schuster, B. (2017). Perceived barriers and facilitators of mental health service utilization in adult trauma survivors: A systematic review. Clinical Psychology Review, 52, 52–68. 10.1016/j.cpr.2016.12.001 [DOI] [PubMed] [Google Scholar]
  13. Kazantzis, N. , & Shinkfield, G. (2007). Conceptualizing patient barriers to nonadherence with homework assignments. Cognitive and Behavioral Practice, 14(3), 317–324. 10.1016/j.cbpra.2006.08.003 [DOI] [Google Scholar]
  14. Kazantzis, N. , Whittington, C. , Zelencich, L. , Kyrios, M. , Norton, P. J. , & Hofmann, S. G. (2016). Quantity and quality of homework compliance: A meta‐analysis of relations with outcome in cognitive behavior therapy. Behavior Therapy, 47(5), 755–772. 10.1016/j.beth.2016.05.002 [DOI] [PubMed] [Google Scholar]
  15. Lenton‐Brym, A. P. , Collins, A. , Lane, J. , Busso, C. , Ouyang, J. , Fitzpatrick, S. , Kuo, J. R. , & Monson, C. M. (2025). Using machine learning to increase access to and engagement with trauma‐focused interventions for posttraumatic stress disorder. British Journal of Clinical Psychology, 64(1), 125–136. 10.1111/bjc.12468 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Li, H. , Zhang, R. , Lee, Y.‐C. , Kraut, R. E. , & Mohr, D. C. (2023). Systematic review and meta‐analysis of AI‐based conversational agents for promoting mental health and well‐being. NPJ Digital Medicine, 6(1), Article 236. 10.1038/s41746-023-00979-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Mallard Swanson, K. , Song, J. , Beristianos, M. , Aajmain, S. , Lane, J. E. M. , Landy, M. S. H. , Suvak, M. K. , Shields, N. , Monson, C. M. , & Stirman, S. W. (2021). A glimpse into the “black box”: Which elements of consultation in an ebp are associated with client symptom change and therapist fidelity? Implementation Research and Practice, 2, Article 263348952110517. 10.1177/26334895211051791 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Monson, C. M. , Shields, N. , Suvak, M. K. , Lane, J. E. M. , Shnaider, P. , Landy, M. S. H. , Wagner, A. C. , Sijercic, I. , Masina, T. , Wanklyn, S. G. , & Stirman, S. W. (2018). A randomized controlled effectiveness trial of training strategies in cognitive processing therapy for posttraumatic stress disorder: Impact on patient outcomes. Behaviour Research and Therapy, 110, 31–40. 10.1016/j.brat.2018.08.007 [DOI] [PubMed] [Google Scholar]
  19. Pace, B. T. , Song, J. , Suvak, M. K. , Shields, N. , Monson, C. M. , & Stirman, S. W. (2021). Therapist self‐efficacy in delivering cognitive processing therapy in a randomized controlled implementation trial. Cognitive and Behavioral Practice, 28(3), 327–335. 10.1016/j.cbpra.2020.08.002 [DOI] [Google Scholar]
  20. Resick, P. A. , Monson, C. M. , & Chard, K. M. (2024). Cognitive processing therapy for PTSD: A comprehensive therapist manual (2nd ed.). Guilford Press. [Google Scholar]
  21. Sloan, D. M. , & Marx, B. P. (2025). Written exposure therapy for PTSD: A brief treatment approach for mental health professionals (2nd ed.). American Psychological Association. 10.1037/0000441-000 [DOI] [Google Scholar]
  22. Stade, E. C. , Eichstaedt, J. C. , Kaysen, D. , Salecha, A. , Greenberger, A. , Singhvi, S. , & Stirman, S. W. (2024). TherapyTrainer: Using AI to train therapists in written exposure therapy. Open Science Framework. 10.31219/osf.io/wx93m [DOI] [Google Scholar]
  23. Stade, E. C. , Eichstaedt, J. C. , Kim, J. P. , & Stirman, S. W. (2025). Readiness evaluation for AI‐Mental Health Deployment and Implementation (READI): A review and proposed framework . PsyArXiv. 10.31234/osf.io/8zqhw_v2 [DOI] [PMC free article] [PubMed]
  24. Stade, E. C. , Stirman, S. W. , Held, P. , Schwartz, H. A. , & Eichstaedt, J. C. (2024). Designing clinical psychological AI that reduces suffering: Challenges and technical considerations. Open Science Framework. 10.31234/osf.io/tjv8p [DOI] [Google Scholar]
  25. 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). Large language models could change the future of behavioral healthcare: A proposal for responsible development and evaluation. NPJ Mental Health Research, 3(1), Article 12. 10.1038/s44184-024-00056-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Stirman, S. W. , Gutner, C. A. , Gamarra, J. , Suvak, M. K. , Vogt, D. , Johnson, C. , Wachen, J. S. , Dondanville, K. A. , Yarvis, J. S. , Mintz, J. , Peterson, A. L. , Young‐McCaughan, S. , & Resick, P. A. (2021). A novel approach to the assessment of fidelity to a cognitive behavioral therapy for PTSD using clinical worksheets: A proof of concept with cognitive processing therapy. Behavior Therapy, 52(3), 656–672. 10.1016/j.beth.2020.08.005 [DOI] [PubMed] [Google Scholar]
  27. Stirman, S. W. , Gutner, C. A. , Suvak, M. K. , Adler, A. , Calloway, A. , & Resick, P. A. (2018). Homework completion, patient characteristics, and symptom change in cognitive processing therapy for PTSD. Behavior Therapy, 49(5), 741–755. 10.1016/j.beth.2017.12.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Talebirad, Y. , & Nadiri, A. (2023). Multi‐agent collaboration: Harnessing the power of intelligent LLM agents . ArXiv. 10.48550/ARXIV.2306.03314 [DOI]
  29. Torous, J. , & Blease, C. (2024). Generative artificial intelligence in mental health care: Potential benefits and current challenges. World Psychiatry, 23(1), 1–2. 10.1002/wps.21148 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. U.S. Department of Veterans Affairs/U.S. Department of Defense . (2023). VA/DoD clinical practice guideline for management of posttraumatic stress disorder and acute stress disorder . https://www.healthquality.va.gov/guidelines/MH/ptsd/VA‐DoD‐CPG‐PTSD‐Full‐CPGAug242023.pdf∖
  31. Vittorio, L. N. , Murphy, S. T. , Braun, J. D. , & Strunk, D. R. (2022). Using Socratic questioning to promote cognitive change and achieve depressive symptom reduction: Evidence of cognitive change as a mediator. Behaviour Research and Therapy, 150, Article 104035. 10.1016/j.brat.2022.104035 [DOI] [PubMed] [Google Scholar]
  32. Waltman, S. , Hall, B. C. , McFarr, L. M. , Beck, A. T. , & Creed, T. A. (2017). In‐session stuck points and pitfalls of community clinicians learning CBT: Qualitative investigation. Cognitive and Behavioral Practice, 24(2), 256–267. 10.1016/j.cbpra.2016.04.002 [DOI] [Google Scholar]

Articles from Journal of Traumatic Stress are provided here courtesy of Wiley

RESOURCES