Beyond Compliance: A Resistance-Informed Motivation Reasoning Framework for Challenging Psychological Client Simulation

Danni Liu; Bo Liu; Yuxin Hu; Hantao Zhao; Yan Liu; Ding Ding; Jiahui Jin; Jiuxin Cao

cs.AIApr 12, 2026

Beyond Compliance: A Resistance-Informed Motivation Reasoning Framework for Challenging Psychological Client Simulation

Authors: Danni Liu, Bo Liu, Yuxin Hu, Hantao Zhao, Yan Liu, Ding Ding, Jiahui Jin, Jiuxin Cao

Organizations: Southeast University, Nanjing, China · Purple Mountain Laboratories, Nanjing, China · The Nanjing Derong Wisdom Information Technology Co., Ltd., Nanjing, China

Abstract

Psychological client simulators have emerged as a scalable solution for training and evaluating counselor trainees and psychological LLMs. Yet existing simulators exhibit unrealistic over-compliance, leaving counselors underprepared for the challenging behaviors common in real-world practice. To bridge this gap, we present ResistClient, which systematically models challenging client behaviors grounded in Client Resistance Theory by integrating external behaviors with underlying motivational mechanisms. To this end, we propose Resistance-Informed Motivation Reasoning (RIMR), a two-stage training framework. First, RIMR mitigates compliance bias via supervised fine-tuning on RPC, a large-scale resistance-oriented psychological conversation dataset covering diverse client profiles. Second, beyond surface-level response imitation, RIMR models psychologically coherent motivation reasoning before response generation, jointly optimizing motivation authenticity and response consistency via process-supervised reinforcement learning. Extensive automatic and expert evaluations show that ResistClient substantially outperforms existing simulators in challenge fidelity, behavioral plausibility, and reasoning coherence. Moreover, ResistClient facilities evaluation of psychological LLMs under challenging conditions, offering new optimization directions for mental health dialogue systems.

Figures & tables

Figure 1: Existing LLM-based client simulators exhibit overly-compliance. ResistClient generates authentic resistant behaviors through motivation reasoning steps.

Figure 2: Overview of our proposed ResistClient with Resistance-informed Motivation Reasoning framework.

Figure 3: Application of ResistClient in interactive counseling training.

Models	Automated Metrics(%)				Manual Metrics
Models	Precision	Recall	F1 Score	RTF	Fid. ( $\uparrow$ )	Rat. ( $\uparrow$ )	Qua. ( $\uparrow$ )
GPT-5.1	59.31	62.88	61.04	35.94	1.42	1.35	2.52
DeepSeek-V3.2	52.87	57.56	55.12	36.91	1.29	1.21	2.40
Kimi-K2-thinking	47.08	51.19	49.05	36.86	1.24	1.19	2.18
GLM-4.6	49.10	54.72	51.76	37.78	1.27	1.23	2.34
Qwen3-8B	36.52	48.54	41.68	45.06	1.10	1.04	1.88

Table 1: Resistance simulation performance of different models in psychological conversations.

Figure 4: Confusion matrices comparing reaction-type generation across variants. Rows represent ground-truth reaction types; columns represent generated types.

Client	PATIENT	Anna	Yang et al.	Resist
Simulator	- $\Psi$	Agent	( 2025b )	Client
Automated Metrics
CCR	87.94	78.62	62.33	60.84
Turns	11.24	12.65	16.67	17.88
Coh. ( $\uparrow$ )	0.51	0.62	0.68	0.73
Manual Metrics

Table 2: Evaluation results of client simulators.

Models	Automated		Manual
Models	RTF	Turns	Eff. ( $\uparrow$ )	CDD ( $\downarrow$ )	CPD ( $\uparrow$ )
GPT-5.1	41.14	15.24	2.04	1.58	1.92
Gemini-3-flash	43.77	21.52	1.82	2.02	1.87
DeepSeek-V3.2	40.38	18.40	1.93	1.67	2.05
GLM-4.6	44.54	24.36	1.87	1.75	1.83
MeChat	51.93	43.22	1.61	2.08	1.68

Table 3: Performance of psychological LLMs when interacting with ResistClient.

Appendix figures & tables12 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 5: Distribution of different topics in client profiles

Reaction Category	Label Type	Behavioral Description	Motivational Interpretation
Resistant Reactions	Controlling Resistance	- Dominates the conversation /interrupts - Rejects guidance on conversation direction - Clings to own views, disregards input	Seeks to maintain control to avoid being influenced; insists on own perspective and resists change.

	Emotional Resistance	- Expresses anger or aggressive speech - Displays sadness/emotional breakdown - Shows despair or overgeneralized negative emotions	Externalizes emotions or projects blame to avoid confronting deeper psychological pain.

	Defensive Resistance	- Questions the practitioner’s expertise or competence - Challenges the consultation method or process - Responds with sarcasm or exaggeration	Projects anxiety onto the practitioner to prevent internal conflicts from being triggered.

Appendix

Table 4: Classification and Interpretation of Dataset Labels

Resistance Type	Typical Trigger Situations	Psychological Problems	High-Risk 5P Profile Features
Controlling Resistance	Direct advice, reframing attempts, or agenda-setting by the counselor that challenges the client’s autonomy or preferred narrative.	Autonomy, agency, and perceived self-coherence.	Predisposing : High need for control; rigid belief systems. Perpetuating : Interpersonal dominance patterns.
Emotional Resistance	Emotion-focused prompts or interpretations that surface intense affect without sufficient stabilization or safety cues.	Affect regulation, emotional tolerance, and vulnerability defenses.	Precipitating : Acute relational loss or trauma exposure. Perpetuating : Poor emotion regulation strategies.
Defensive Resistance	Meta-level questioning of the counselor’s methods, competence, or therapeutic intent.	Threatened self-image and externalized anxiety.	Predisposing : Prior negative counseling experiences. Perpetuating : Mistrust of authority figures.
Avoidant Resistance	Exploratory questions targeting core conflicts, personal responsibility, or emotionally salient themes.	Cognitive avoidance and attentional disengagement.	Predisposing : Habitual avoidance coping styles. Perpetuating : Reinforcement of distraction-based regulation.
Compliant Resistance	Open-ended or reflective prompts that invite deeper exploration beyond surface-level agreement.	Superficial compliance masking emotional disengagement.	Predisposing : Fear of interpersonal conflict or rejection. Perpetuating : Over-adaptation to perceived expectations.

Appendix

Table 5: Resistance Trigger Timing Specification

Figure 6: Distribution of different resistance types in RPC dataset

Models	Acc.(%)	Quality	Cost ($)
GPT-5.1	74.4	2.31	10.00
DeepSeek-V3.2	72.3	2.47	0.43
Gemini-3-Flash	68.6	2.05	3.00

Appendix

Table 6: Comparison of annotation accuracy(Acc.), rewriting quality(Quality), and API cost(Cost). Rewriting quality is scored by counselors on a 0–3 scale, with higher scores indicating more human-like resistance expression. API Cost is the output price per 1 million tokens.

\mathbf{R}_{i}=\{r_{i,1}^{\mathrm{idx}(1)},\,r_{i,2}^{\mathrm{idx}(2)},\,(r_{i,3}+r_{i,5})^{\mathrm{idx}(3)},\,(r_{i,4}+r_{i,5})^{\mathrm{idx}(4)}\}

Appendix

Algorithm 1 Offline Token-level GRPO with Consistency-aware Reward Reweighting

Figure 7: Example of 5p profile annotation

Figure 8: Example dataset with resistant behavior annotations

Figure 9: Examples of conversation incorporating reasoning processes

Figure 10: Informed Consent Form for Counsellor Participation

Figure 11: ResistClient and SoulChat2.0 Conversation Example (Chinese Version)

Figure 12: ResistClient and SoulChat2.0 Conversation Example (English Version)

Explore similar work

PatientAct: Theory-Grounded Mental Health Client Simulation
Aug 13, 2026Sahand Sabour, TszYam NG, Yaqian Chen +3Mental Health Counseling User Simulation
The CoAI Group, DCST, Institute for Artificial Intelligence, Tsinghua University, Beijing, China · Department of Psychology, Beijing Normal University, Beijing, China · Counseling and Psychological Development Guidance Center, Tsinghua University, Beijing, China
When Clients Stop Following: A Cognitive Conceptualization Diagram-driven Framework for Strategic Counseling
Jun 3, 2026Yihao Qin, Junyi Zhao, Changsheng Ma +4Mental Health Counseling Cognitive Modeling
School of Information Science and Engineering, Lanzhou University
MASC: A Multi-Agent Self-Calibration Framework with Latent Construct Alignment for Consistent Client Role-Playing in Psychological Counseling
Oct 6, 2026Shixin Peng, Kun Jiang, Jiaxing Zheng +2Large Language Model-Based Role-Play Simulation Multi-Agent LLM Systems
Central China Normal University, Wuhan, China · Shanghai Jiao Tong University, Shanghai, China

Aug 13, 2026J/K move · Enter open · S save

PatientAct: Theory-Grounded Mental Health Client Simulation

Sahand Sabour, TszYam NG, Yaqian Chen, Guanqun Bi+2

The CoAI Group, DCST, Institute for Artificial Intelligence, Tsinghua University, Beijing, China · Department of Psychology, Beijing Normal University, Beijing, China · Counseling and Psychological Development Guidance Center, Tsinghua University, Beijing, China

Mental Health Counseling User Simulation

Open paper AskPDF arXiv

LLM-based simulated clients are increasingly used to train novice counselors, evaluate LLM therapists, and generate synthetic data. However, current simulators produce overly cooperative clients that disclose too readily, accept therapeutic reframes without resistance, and resolve core issues within a single session. We trace these issues to profiles that lack causal depth and behavioral mechanisms that treat all content as equally accessible. We present PatientAct, a framework for client simulation grounded in established clinical theories. Our profiles integrate the 5Ps clinical case formulation, providing causal depth without tying the design to any single therapeutic modality. During simulation, profiles include a dynamic memory layer in which items carry trust thresholds (e.g., symptoms are available early, whereas formative memories require a sustained therapeutic alliance). At each turn, the client's emotional reaction and behavior are modeled before generating a response. If the therapist approaches gated content, PatientAct expresses resistance in terms of quantity, content, and style rather than defaulting to cooperation or a single resistance pattern. We evaluate our framework on 40 clinical situations and demonstrate that it generates diverse profiles with high clinical plausibility. Moreover, PatientAct significantly outperforms the baselines, yielding substantial gains in resistance quality and behavioral realism. Our code and data are publicly available via github.com/Sahandfer/PatientHub.

Abstract

Explore similar work

PatientAct: Theory-Grounded Mental Health Client Simulation

When Clients Stop Following: A Cognitive Conceptualization Diagram-driven Framework for Strategic Counseling

MASC: A Multi-Agent Self-Calibration Framework with Latent Construct Alignment for Consistent Client Role-Playing in Psychological Counseling