Generative AI as a Relational Bridge in Couples Counseling: A Supplemental Approach

Natural Language Processing

Natural language processing (NLP) refers to methods that allow computers to analyze, summarize, and generate human language. In mental health, NLP tools may help support language-focused tasks such as organizing text, summarizing written material, and identifying recurring linguistic patterns. However, their outputs require careful clinical interpretation and oversight (Lawrence et al., 2024). When used in therapist-facing tools, NLP can organize client-provided text, such as journal entries or check-ins, and highlight themes or escalation markers (e.g., absolutist language such as “always” or “never”). In couples counseling, NLP outputs are descriptive indicators—not conclusions. Therapists interpret patterns within systemic formulations, including attachment, culture, stress, power, and context.

Large Language Models

LLMs are generative systems trained on large text datasets that can produce coherent dialog, summaries, and structured prompts. Emerging research suggests that, under therapist guidance, LLMs may support mental health practice in bounded ways, including psychoeducation, structured skills rehearsal, and drafting language for clinical or educational purposes, while remaining limited by risks or bias, inaccuracy, and overgeneralization (Lawrence et al., 2024). In couples counseling, LLMs may assist with skills rehearsal by generating examples of reflective listening, soft startup language, repair attempts, and needs-based statements. They may also assist with drafting psychoeducational handouts or between-session exercises for therapist review. However, LLMs should not be relied upon to assess risk, diagnose problems, or arbitrate relational disputes, and their use should remain subordinate to therapist judgment and ethical oversight (ACA, 2024a, 2024b; Lawrence et al., 2024).

Conversational Agents and Chatbots

Conversational agents, or chatbots, are interfaces that combine generative language with rules, boundaries, and sometimes escalation pathways. Research on conversational agents in health and mental health settings suggests that they may provide accessible, low-intensity support for structured interactions, but their usefulness depends on the context, scope, and level of clinical oversight (Lawrence et al., 2024). In couples counseling, chatbots are most appropriate when restricted to low-acuity skills practice, such as guiding turn-taking or prompting validation statements. Couples should understand that the tool is not a crisis resource. Earlier studies found inconsistent responses to disclosures involving mental health crisis and interpersonal violence, accentuating the need for caution in safety-relevant domains (Miner et al., 2016).

Communication Practice and Skills Rehearsal

Generative AI can support communication practice by offering structured templates and role-based rehearsal. For example, couples may use AI to generate an apology framework (i.e., acknowledgment of impact, responsibility, and repair plan) or to rephrase escalatory statements into needs-based language. When used collaboratively by both partners, these tools may support clearer expression of concerns and, in some cases, reduce perceived defensiveness, consistent with findings that users experience AI-generated relationship guidance as helpful in structured contexts (Vowels et al., 2024). Clinically, AI is best positioned as a workbook-like supplement that supports the practice of therapist-selected skills rather than providing independent interpretations of relationship meaning.

Therapists can track communication patterns over time to assess the effectiveness of AI. For example, a decrease in harsh startups and an increase in partners noticing and responding more positively to each other's bids for connection may suggest that the couple is interacting in less adversarial and more supportive ways. These interactional processes have been repeatedly associated with relationship satisfaction and stability (Gottman, 1994; Gottman & Silver, 1999).

Initial Exploration and Therapy Readiness

AI tools may offer a low-cost entry point for couples who are ambivalent about therapy or who face access barriers. In early exploration, generative AI may help partners clarify emotions, organize concerns, and identify potential next steps for therapy. This aligns with Luo et al. (2025), who found that individuals frequently use ChatGPT to interpret emotional experiences and guide subsequent actions. As part of this exploratory process, AI may also support partners in articulating shared goals for therapy (e.g., reducing escalation, rebuilding trust, strengthening friendship) and in generating questions or topics they wish to bring into intake, without replacing therapist-led assessment or goal setting. This role may be most appropriate when relational distress is mild to moderate and when there are no indicators of coercion, violence, or acute safety concerns. Counselors can frame this use as preparation that increases readiness for professional services rather than as a substitute for therapy.

Resource Generation and Relationship Maintenance

Generative AI may rapidly generate psychoeducational summaries, communication exercises, and ideas for relationship maintenance (e.g., weekly check-in agendas, appreciation prompts, or shared meaning activities). When counselors review and tailor these resources, AI may reduce administrative burden and improve between-session follow-through. Professional guidance remains essential because generic advice may not fit a couple's cultural values, relational history, or clinical context (Lawrence et al., 2024).

Therapist-Aided Reflection on Patterns

When used by therapists with explicit consent and appropriate privacy safeguards, generative AI may assist with organizing themes from structured client materials (e.g., journaling or session preparation exercises) and identifying repeated language patterns that reflect escalation or withdrawal (e.g., blame, contempt, defensiveness, or stonewalling). In practice, this use may resemble analytic tools that support case formulation; the therapist reviews AI-generated summaries as provisional hypotheses and integrates them into systemic assessment and treatment planning rather than treating them as independent clinical interpretations. Consistent with counseling guidance, AI-assisted analytics should be continuously evaluated for accuracy, appropriateness, and potential bias, with professional judgment remaining central to interpretation and intervention (ACA, 2024a; Lawrence et al., 2024). To embed cultural humility in pattern analytics, therapists may consider prompts that question whether language “red flags” signal cultural style rather than pathology. These reflections can safeguard against biased interpretations of AI-flagged patterns and prompt therapists to consider whether language differences reflect cultural communication styles rather than pathology (Hardy & Laszloffy, 1995; Sue et al., 2019).

Collectively, these applications highlight a persistent gap in the marriage and family therapy literature concerning how clinicians translate ethical guidance on AI into concrete, therapist-in-the-loop practices. These applications illustrate the need for practical tools that support communication rehearsal and therapy preparation without substituting for clinical judgment. Addressing this gap requires examples that are intentionally bounded, skill-focused, and embedded within ongoing therapeutic oversight. This concern is also consistent with therapist-perspective research. In a qualitative preprint study of 25 experienced couple and family therapists, interviewees emphasized privacy concerns, depersonalization risks, and the potential impact of AI on the therapeutic relationship (Aviram, 2025).

Depersonalization and Triangulation

AI use can depersonalize relational work when partners outsource difficult conversations to the tool or seek the AI's validation rather than engaging directly. A common relational risk associated with the use of generative AI in couple work is triangulation. Within family systems frameworks, triangulation occurs when a third party is introduced to regulate relational anxiety, often at the expense of direct engagement between partners (McCauley et al., 2021). Generative AI may function in this role when one partner consults AI privately, receives perceived validation, and then invokes AI output as an external authority in conflict, thereby intensifying polarization and avoidance. This concern aligns with ethical MFT scholarship, warning that technology-mediated interventions can inadvertently disrupt relational boundaries and alliances if systemic implications are not carefully considered (Lawrence et al., 2024). Hertlein and Springer (2026) suggest that a common manifestation of this process occurs when one partner privately consults AI, receives one-sided validation, and then uses AI output as evidence against the other partner. This dynamic may increase polarization and undermine therapy goals by shifting the focus from mutual understanding to adjudication. Therapists should proactively assess whether AI use is strengthening or bypassing direct relational engagement.

Inability to Address Deep Issues, Power, and Safety

Generative AI is not designed to reliably assess coercion, intimate partner violence, suicidality, or acute crisis risk. Research on conversational agents has shown that they produce inconsistent or incomplete responses to disclosures involving interpersonal violence (Miner et al., 2016). As a result, AI should not be positioned as a mediator for high-conflict relationships, trauma-related dynamics, or safety-relevant situations. Therapists must treat these contexts as contraindications for AI-guided relationship coaching and prioritize appropriate risk assessment, safety planning, and referrals.

Decision Points

The model includes five decision points to guide session planning and supervision. First, the therapist clarifies the purpose of AI use, such as skills rehearsal, therapy preparation, or homework support. Second, the therapist conducts a risk screen, giving particular attention to coercion, violence, acute safety risk, and severe instability. Third, the therapist establishes a transparency agreement to reduce secrecy and triangulation, including shared use when possible and disclosure of individual use. Fourth, the therapist addresses data and platform safeguards, including data minimization and explicit discussion of privacy risks. Fifth, the therapist ensures therapist oversight: AI outputs are reviewed and contextualized, and AI is not used as an independent decision-maker.

Role-play scenarios can allow therapists to practice articulating the purpose of AI use, conducting risk screenings, establishing transparency agreements, addressing privacy concerns, and reviewing AI outputs collaboratively. This experiential practice transforms abstract steps into embodied skills and may enhance readiness to apply the framework in real-world sessions.

Contraindications

The model identifies common exit ramps. AI use is contraindicated when it is used to prove a partner wrong, as this can heighten defensiveness and undermine therapy goals. It is also contraindicated when AI is used to provide diagnostic labels, because AI lacks clinical judgment and can contribute to misinterpretation. Managing crises through AI is not advisable because AI cannot assess real-time risk or provide empathic support. Additionally, AI should not replace vulnerable engagement, as this can prevent direct communication between partners. AI is also contraindicated when there are indicators of coercion, violence, or acute safety risk, as these situations require personalized intervention and professional judgment. Finally, when privacy protections are unclear, AI should not be used, because confidentiality and data security are paramount in therapeutic contexts. In such situations, the therapist's ethical responsibilities require prioritizing safety, appropriate services, and clinically indicated referrals.

Practice Framework: Therapist-in-the-Loop Use of Generative AI

Building on the Relational Bridge Model, the following practice framework offers concrete steps for clinicians who wish to ethically integrate generative AI as a supplemental tool. These steps are designed to preserve the therapeutic alliance and direct relational engagement while remaining consistent with Standard VI's emphasis on appropriateness, risk-benefit discussion, security, and clinician competence (AAMFT, 2015).

Step 1: Assess AI Use, Nonuse, and Preferences at Intake. During intake and early sessions, therapists assess whether either partner uses AI for relationship support and, if not, whether the couple is interested in using AI-based tools as part of structured, therapist-guided homework. Therapists can also assess access and digital comfort.

Step 2: Screen for Risk and Contraindications. Therapists screen for coercion, violence, stalking, and acute safety concerns. If risk indicators are present, AI-guided relationship coaching should not be recommended. Therapists prioritize evidence-based safety procedures, mandated reporting obligations as applicable, and appropriate referrals.

Step 3: Establish a Transparency Agreement. Therapists help couples create agreements about AI use. Agreements may include using AI together for selected exercises, disclosing individual AI use relevant to the relationship, avoiding weaponizing AI outputs, and bringing AI-generated exercises to therapy for review rather than implementing them uncritically.

Step 4: Obtain Informed Consent for AI-Integrated Homework. When therapists recommend or integrate AI tools (e.g., therapist-approved prompts for between-session practice), they discuss risks and benefits, privacy limitations, and role boundaries. Informed consent emphasizes that AI is not a therapist or crisis resource, and may generate inaccurate or biased content. Clients are encouraged to limit the use of identifying information and to report concerns or adverse experiences in session.

Step 5: Use AI for Bounded, Skill-Focused Tasks. Therapists select tasks that align with the couple's stage of therapy and their capacity for self-regulation. Examples include reflective listening rehearsal, needs-based statement drafting, repair attempts, and structured check-in agendas. Therapists avoid tasks that require trauma processing, crisis intervention, or adjudication of right and wrong.

Step 6: Debrief AI Use in Session. Therapists routinely debrief AI use, asking what felt helpful, what felt invalidating, and how AI use affected the couple's direct communication. An effective debrief opener might be, “What surprised or unsettled you about using the tool?” This approach signals that ambivalence is expected and welcome. Debriefing also allows the therapist to correct inaccurate information, address relational meaning, and strengthen the couple's capacity for vulnerability and repair.

Case of Maya

Maya (age 38) and Chris (age 40) presented for couples counseling after several years of escalating conflict around parenting, household labor, and emotional connection. Maya described feeling chronically alone and overburdened; Chris described feeling criticized and inadequate. During conflict, Maya sought problem-solving and reassurance, while Chris withdrew and grew quiet, which intensified Maya's distress. Both partners reported decreased intimacy and increased resentment over the past year.

Implications for Marriage and Family Counselors

In practice, clinicians can address the increasing presence of generative AI by adopting an ethically grounded, therapist-in-the-loop approach tailored to each couple's context. Some couples may already use AI for relationship support, while others may be unfamiliar, skeptical, or disinterested. Therapists can respond effectively in both situations by assessing AI exposure and preferences, clarifying roles and boundaries, and recommending AI-supported exercises only when clinically appropriate.

Clinicians should assess AI use and technology preferences. AI can be assessed alongside other self-help behaviors without presuming that it is always present. For couples already using AI, therapists can ask what tools they use, what problems they bring to AI, whether use is shared or kept secret, and whether AI use supports or undermines direct engagement. For couples who have not used AI, therapists can assess openness, access, and potential fit before making any recommendation. Some couples may prefer nondigital homework or may have privacy concerns that outweigh potential benefits. When a therapist recommends AI-assisted practice, the recommendation can be framed as optional, time-limited, and skill-specific, with the expectation that the output will be reviewed and contextualized in session to reduce triangulation and role confusion.

AAMFT (2015) ethics language should be utilized to set boundaries. When AI is incorporated into therapy, clinicians can use the logic of Standard VI to guide decision-making: Determine appropriateness, discuss risks and benefits, attend to security and confidentiality, and practice within competence (AAMFT, 2015). In couples work, therapists can also draw from Standard II's emphasis on protecting confidentiality across multiple clients by developing clear agreements about what information is shared, how AI outputs are handled, and how partners’ privacy is respected.

It is important to frame AI as a tool and not as a therapist. Clinicians can reduce unrealistic expectations by framing AI as comparable to a workbook or a communication coach that supports skills practice rather than a clinician. This framing can preserve the couple's reliance on the therapeutic relationship for deeper emotional processing and systemic change.

It is imperative for clinicians to identify contraindications and prioritize safety. Clinicians should avoid recommending AI-guided relationship coaching when there are indicators of coercion, violence, stalking, acute crisis, or severe instability. In these contexts, AI's limitations in risk assessment and crisis response make it an unsafe substitute for professional services. Therapists prioritize safety planning, appropriate referrals, and evidence-based interventions.

Implications for Counselor Education and Supervision

Counselor educators and supervisors can prepare trainees for AI-augmented practice by integrating technology ethics, critical evaluation skills, and applied decision-making. The ACA's recommendations for practicing counselors emphasize competence, transparency, informed consent, and ongoing evaluation of AI outputs (ACA 2024a). Training can include supervised exercises in which students evaluate AI-generated communication scripts for bias, cultural fit, and potential relational harm, followed by debriefing on alliance and ethics.

Competence and supervision for AI-integrated practice can be strengthened by incorporating structured discussion in supervision about how AI tools influence the therapeutic relationship and couple dynamics. Supervision can incorporate discussion of how AI tools affect the therapeutic relationship and couple dynamics. Supervisors can model responsible use (e.g., using AI for drafting psychoeducation that is then edited for cultural responsiveness) while emphasizing the counselor's accountability for clinical decisions. Given that AI platforms and policies evolve rapidly, supervision should include ongoing attention to privacy, documentation, and professional boundaries.

Strengths and Limitations

The Relational Bridge Model's primary strength is its clinical pragmatism. It translates ethical principles into concrete decision points that therapists can apply in intake, treatment planning, and supervision. The model also foregrounds couple-specific risks such as triangulation and confidentiality complexities. However, the framework is conceptual rather than an empirically tested intervention. Research on generative AI in couples counseling is emerging, and current findings largely derive from experiential prompts, simulation studies, and broader mental health AI literature (Lawrence et al., 2024; Vowels et al., 2024). Future research is needed to evaluate outcomes, equity impacts, and best practices for therapist-audited AI integration in couples therapy. Randomized trials, single-case designs, and qualitative studies could together help bridge the current research-practice gap and clarify where AI-supported interventions are most and least clinically useful.