Generative artificial intelligence in plastic surgery: emerging applications in preoperative psychological screening and patient empowerment

INTRODUCTION

An important but inconsistently implemented component of aesthetic and reconstructive plastic surgery, preoperative psychological screening helps clinicians identify patients at elevated risk for postoperative dissatisfaction, emotional maladjustment, or decisional regret^[1]. Psychological factors - including body image distortion, perfectionistic aesthetic ideals, subclinical body dysmorphic traits, and unrealistic expectations have been linked to lower satisfaction, reduced adherence during recovery, and increased postoperative conflicts^[1]. In routine practice, however, structured psychological interviews and standardized psychometric scales are not consistently employed. Challenges include limited consultation time, restricted access to trained mental health professionals, and patients’ reluctance to disclose distress during brief, symptom-focused encounter^[1]. At the same time, artificial intelligence (AI) is increasingly integrated into plastic and reconstructive surgery, aiding in tasks such as risk prediction, preoperative planning, and outcome assessment^[2].

These workflow constraints necessitate scalable, individualized, and low-barrier tools to foster psychological readiness and align expectations before surgery. Generative AI - particularly LLM-enabled conversational systems - has exhibited good potential to sustain naturalistic, context-aware dialogue and may help elicit motivation-, expectation-, and affect-related cues from patient narratives. In the preoperative plastic surgery setting, these systems facilitate a structured workflow that integrates patient-reported concerns and questionnaire responses, free-text narratives, screening-like dialogue, patient-facing psychoeducation, clinician-facing summarization, and safety governance [Figure 1] - has demonstrated the ability to sustain naturalistic, context-aware dialogue, and may help elicit motivation-, expectation-, and affect-related cues from patient narratives^[3]. Compared to static questionnaires, LLM-powered conversational interfaces can adapt follow-up questions in real time, effectively mimicking semi-structured interviews. This dynamic capability suggests a potential role for generative AI in augmenting preoperative counseling and supporting the early identification of patients who could benefit from more in-depth clinician review or specialized referral pathways^[4]. Importantly, AI deployment in plastic surgery is no longer purely conceptual; instead, AI methods are actively being explored across subspecialties like breast reconstruction for applications such as preoperative planning, complication prediction, aesthetic outcome assessment, and patient education, thereby underscoring its broader trajectory toward patient-centered care^[5].

Figure 1. A conceptual overview of generative AI for preoperative psychological screening and patient empowerment in plastic surgery. AI: Artificial intelligence; LLM: large language model.

This schematic illustrates how an LLM-enabled conversational interface processes patient-reported information, including aesthetic concerns, motivations, expectations, questionnaire responses, and free-text narratives. The system supports screening-like intake, psychoeducation, expectation clarification, and clinician-facing summarization. It is designed to augment, rather than replace, clinician judgment, and operates within a governed framework that includes human oversight, non-diagnostic use, privacy protection, bias awareness, and escalation pathways.

Initial studies are evaluating LLM performance in patient-facing communication. In simulated consultations (e.g., rhinoplasty, abdominoplasty), LLM-generated responses have been rated as coherent and readable, potentially incorporating elements of motivational interviewing (MI)-style reflection and cognitive-behavioral psychoeducation^[4,6]. However, the evidence base is nascent, with current evaluations often limited to vignette-based studies or focused on feasibility. Major challenges include the reliability of LLMs in recognizing nuanced psychological risks, the limited interpretability of their outputs, and safety concerns such as hallucinated content or demographic and cultural bias. Furthermore, few publications describe comprehensive, end-to-end architectures that integrate conversational screening, structured reporting for surgeons, and patient-facing educational or supportive modules within clinically governed workflows^[7].

This narrative review synthesizes emerging concepts and early evidence on the application of generative AI for preoperative psychological screening and patient empowerment in plastic surgery. We first outline the psychological risks pertinent to aesthetic surgery candidates and summarize unmet needs in current practice^[3]. We then examine generative AI’s potential to support conversational screening, expectation management, and scalable psychoeducation, while underscoring the critical need for human oversight and clear boundaries of use. Finally, we discuss its limitations and future directions - including multimodal sensing, federated learning, and ethical governance - to propose a pragmatic roadmap for safe and clinically meaningful translation.

To obtain the relevant literature for this narrative review, an extensive retrieval process was conducted in PubMed, Scopus, and Google Scholar, covering the period from January 2015 to March 2025. Search keywords included: generative AI, large language models (LLMs), ChatGPT, plastic surgery, aesthetic surgery, preoperative psychological screening, patient empowerment, and natural language processing (NLP). We prioritized peer-reviewed primary literature, including systematic reviews and clinical feasibility studies, that examined the application of AI technology in surgical decision-making, psychological risk evaluation, and patient education. Articles were screened for relevance to plastic and reconstructive surgery clinical situations, with a particular focus on novel conversational AI tools and their ethical or practical implications.

PSYCHOLOGICAL RISK LANDSCAPE IN PLASTIC SURGERY CANDIDATES

The psychological profile of aesthetic plastic surgery candidates is often more complex than their stated motivations imply. Although these procedures are elective and highly personal, requests for cosmetic change may stem less from objective deformity or functional impairment and more from perceived shortcomings in self-image, social standing, or emotional well-being^[8]. With beauty ideals increasingly influenced by digitally curated content and algorithmic aesthetics, a patient’s preoperative psychological readiness and their ability to maintain stable expectations have become crucial for postoperative satisfaction and psychosocial resilience.

Body image-related psychopathology, particularly body dysmorphic disorder (BDD) and subclinical dysmorphic traits, is a significant concern within this population. Patients exhibiting dysmorphic preoccupation, cognitive rigidity, or externally contingent self-worth often report dissatisfaction even with objectively acceptable surgical outcomes. Clinically, this manifests as persistent fixation on minor imperfections, repeated reassurance-seeking, emotional volatility, and, at times, escalated conflict with providers. Consequently, studies have explored structured or technology-assisted screening approaches to facilitate earlier identification and referral of these higher-risk individuals. However, the strength and generalizability of the available evidence vary across different settings and instruments^[9].

A complementary line of work examines language use as a proxy for psychological distress. NLP models, applied to patient-generated text - such as open-ended intake responses, clinical narratives, or interview transcripts - have shown promise in identifying linguistic markers associated with anxiety, depressive symptoms, rumination, and crisis-related content^[10]. In preoperative plastic surgery, language signals should not be interpreted as standalone diagnostic outputs. Instead, their assessment requires privacy-preserving preprocessing, contextual review, and clinician-supervised thresholds [Figure 2]. However, applying such approaches to plastic surgery demands careful contextual consideration. This is because appearance-related narratives can be emotionally charged without necessarily reflecting a clinical disorder, and signals derived from social media or non-clinical text may not directly map onto surgical decision-making risk. These challenges underscore the need for domain-specific constructs and clinically governed thresholds when implementing language-based screening in aesthetic practice.

Figure 2. NLP-assisted workflow for preoperative psychological risk review in plastic surgery. NLP: Natural language processing.

This schematic illustrates how patient-generated text - including open-ended intake responses, aesthetic concerns, motivation-related statements, and consultation narratives - is processed using privacy-preserving preprocessing and NLP-assisted language analysis. The potential linguistic signals identified, such as affective tone, rumination-related language, catastrophic wording, social comparison cues, and crisis-related language cues, should be interpreted within a clinical context. These signals are meant to support descriptive summaries, follow-up questions, and clinician review, not automated diagnosis or definitive risk classification.

Despite broad agreement on the importance of psychological readiness, routine clinical workflows often lack systematic tools to capture relevant risk profiles. While common screening instruments like the Symptom Checklist-90 (SCL-90) or the Beck Depression Inventory (BDI) offer general measures of psychopathology, they often overlook constructs specific to surgery, such as aesthetic perfectionism, appearance-based social comparison, media-driven body dissatisfaction, and unstable or externally imposed expectations^[1,11]. Moreover, plastic surgery consultations are typically brief, high-volume, and primarily focused on anatomical planning, which inherently limits the feasibility of in-depth psychological interviewing. Surgeons often lack the time and formal training for structured assessments, while patients may underreport distress due to stigma, minimization, or fear of being denied surgery.

Even when psychological risk is suspected, scalable and accessible support options are often limited. Consequently, many patients experiencing appearance-related anxiety or strong social pressures do not receive targeted, evidence-informed guidance prior to surgery. This leaves emotional vulnerabilities unaddressed and increases the likelihood of postoperative regret or maladjustment^[12]. While educational innovations, such as AI-assisted simulation and individualized training in graduate medical education, can help bridge some clinician communication gaps, they do not fully address patients’ need for continuous, accessible preparation and expectation management^[13].

These challenges underscore a two-fold unmet need: (1) practical mechanisms for early, structured identification of psychological risk signals in routine preoperative workflows; and (2) scalable support promoting readiness and realistic expectations without replacing clinical judgment. This justifies exploring generative AI as a patient-facing interface to facilitate structured dialogue, support psychoeducation, and triage higher-risk presentations for clinician review within a governed care pathway.

GENERATIVE AI AS A TOOL FOR PREOPERATIVE PSYCHOLOGICAL SCREENING

Psychological vulnerabilities in aesthetic surgery candidates are often subtle, context-dependent, and underreported. Consequently, traditional approaches - brief clinician-led interviews and fixed questionnaires - frequently overlook relevant cognitive-emotional patterns, including perfectionistic appearance ideals, externally driven motivation, and self-evaluation shaped by social media. Such assessment gaps have spurred interest in generative AI, particularly LLMs like ChatGPT, as a conversational interface capable of supporting structured preoperative assessment through adaptive questioning and patient-friendly psychoeducation. In contrast to static instruments, LLM-enabled dialogue offers a flexible way to explore motivations, expectations, and emotional cues. This approach can potentially reduce barriers to disclosure and help identify cases warranting clinician review, rather than functioning as a standalone diagnostic system. Evidence from adjacent surgical contexts further suggests that AI-enabled decision aids can improve shared decision-making and decision quality by providing individualized risk-benefit information and outcome estimates^[14]. Leveraging these capabilities, generative AI applications in preoperative plastic surgery care encompass several clinically relevant domains: psychological screening, decision support, psychoeducation and consent support, expectation management, and self-guided supportive functions [Figure 3].

Figure 3. Major application domains of generative AI in preoperative plastic surgery care. AI: Artificial intelligence; FAQs: frequently asked questions.

This schematic summarizes the major application domains of generative AI in preoperative plastic surgery care: psychological screening, decision support, psychoeducation and consent support, expectation management, and self-guided supportive functions. These applications are designed to be adjunctive and human-supervised , facilitating patient empowerment and supporting clinicians in assessment, rather than serving as autonomous diagnostic or decision-making tools.

LANGUAGE-BASED PSYCHOLOGICAL ASSESSMENT POTENTIAL

Trained on vast corpora of human communication, LLMs can discern affective tone, semantic emphasis, and discourse patterns within patient narratives. Unlike fixed-response psychometric instruments, LLM-enabled interfaces can interpret unstructured free-text input and generate context-aware follow-up questions, thereby approximating elements of semi-structured interviewing. In preoperative aesthetic consultations, this capability could help identify risk-relevant cues regarding patient expectations, motivation, and distress.

For example, language that is absolutist (e.g., “I can’t live with this nose”), ruminative (e.g., “I think about it all day”), or comparison-focused (e.g., “everyone else looks better”) may signal heightened preoccupation or fragile self-evaluation and can be used to prompt clarification rather than to establish a diagnosis^[15].

NLP research, from a broader perspective, suggests that unstructured patient-generated text can contain linguistic markers of psychological distress, while aesthetic surgery literature emphasizes assessing body image concerns and symptoms of BDD using validated screening tools^[9,10]. Within this context, open-ended aesthetic narratives can offer valuable insights into motivation, expectations, and distress; however, such linguistic cues should serve as prompts for clinician review, not as standalone diagnostic evidence. These observations support the feasibility of integrating language-based screening workflows as an adjunct to traditional assessment Conversational systems could thereby (i) elicit more comprehensive contextual detail than brief checklists; (ii) identify patterns necessitating clinician review; and (iii) document expectation-related themes directly pertinent to surgical counseling. Nevertheless, given that emotionally intense appearance narratives can manifest in the absence of a clinical disorder, linguistic cues must be interpreted within a structured framework - ideally combined with validated instruments and clear escalation thresholds - rather than being utilized for standalone risk stratification.

SIMULATED MI-INFORMED DIALOGUE AND TRIAGE-ORIENTED SUMMARIZATION

Generative AI can also support preoperative counseling by incorporating elements of MI. MI is a validated communication approach designed to elicit patients’ values, strengthen autonomy, and resolve ambivalence in health-related decisions. Through techniques like reflective listening, affect labeling, and values clarification, AI-powered systems can prompt patients to express their motivations and expectations in greater detail. These systems can also gently challenge beliefs that are rigid or externally driven. For example, if a patient expresses a desire to “look like someone on Instagram”, an MI-informed prompt - such as “How do you think achieving that look would affect your day-to-day life?” - can encourage deeper elaboration. This process helps determine if their goals are stable, internally endorsed, and realistically linked to their well-being^[16]. Importantly, in this setting, such dialogue should be framed as supportive communication and expectation clarification, distinct from psychotherapy.

This capability is especially pertinent in plastic surgery, where motivations can fluctuate between intrinsic drivers (e.g., self-confidence, identity congruence) and extrinsic pressures (e.g., peer approval, relationship dynamics, social media norms)^[17]. Structured conversational prompts can help surgeons identify and document patient expectations during consultations. These prompts surface actionable themes - such as perceived social contingencies, appearance-based self-worth, or idealized outcome comparisons - which then support clinician review and enable targeted counseling. Furthermore, adjacent evidence indicates that LLM-based tools can improve patient-facing materials, including informed consent documents and educational content. Similarly, perioperative chatbots in other surgical specialties have shown high patient satisfaction and improved knowledge acquisition^[17-19]. Together, these findings support the integration of MI-informed conversational modules into a structured, hybrid preoperative workflow that prioritizes education, expectation alignment, and clear escalation pathways for emerging concerns.

INTEGRATION WITH STANDARDIZED METRICS AND HYBRID APPROACHES

Importantly, LLM-enabled screening support does not need to function in isolation. A hybrid framework - combining validated psychometric screening instruments and patient-reported outcome measures with open-ended dialogue - can increase contextual depth while maintaining measurement structure. When scale results are ambiguous or borderline, the system can generate targeted follow-up questions to clarify motivation, expectations, and functional impairment. This supports clinician review, rather than enabling automated diagnosis^[9]. For example, a patient might score low on a general anxiety measure but reveal persistent rumination or perfectionistic language in free-text responses. This discrepancy signals the need for an in-person discussion.

To operationalize these capabilities for perioperative use, we categorized current and emerging use-cases into distinct application domains, specifying patient-facing outputs, clinician-facing deliverables, and recommended evaluation metrics [Table 1].

Moreover, text from AI-guided interactions can be analyzed using NLP pipelines to extract features like affective polarity, cognitive rigidity markers, and semantic themes (e.g., social comparison, externalized self-worth). These features could complement scale-based results, supporting descriptive risk characterization or triage summaries. However, their clinical validity, calibration, and fairness across populations require prospective evaluation.

CLINICAL FEASIBILITY AND INITIAL EVIDENCE

Early studies have investigated the potential of using ChatGPT and related LLMs for patient communication in surgical settings. In simulated consultations for procedures like rhinoplasty and abdominoplasty, blinded reviewers rated LLM-generated responses as comparable to, and in some aspects even more readable and comprehensive than, those written by humans^[20]. Most available studies on LLMs in healthcare focus on general patient education and question-answering, rather than validated psychological assessment or outcome prediction. Nonetheless, these initial findings suggest LLMs can generate coherent, patient-friendly explanations and maintain an empathic tone. These foundational capabilities are crucial for more structured applications, such as clarifying preoperative expectations and creating triage-oriented summaries in aesthetic practice. While broader reviews of AI and machine learning in facial aesthetic surgery highlight growth in AI-supported diagnosis, prediction, and counseling, the specific contribution and clinical effectiveness of generative models remain inconsistently studied across procedures and endpoints^[21].

Generative AI can be integrated into preoperative digital workflows as standalone chat interfaces or as modules within electronic intake systems, enabling patient engagement outside the time constraints of in-person consultations^[22,23]. Asynchronous, conversational formats can make it easier for individuals to share information and reflect at their own pace on their motivations and expectations.

PATIENT EMPOWERMENT THROUGH AI-DRIVEN PSYCHOEDUCATION AND EXPECTATION MANAGEMENT

Preoperative psychological screening aims not only to identify elevated risk but also to promote emotional readiness and informed, preference-sensitive decision-making. In this context, generative AI can support patient empowerment by delivering tailored psychoeducation, structuring expectation clarification, and reinforcing shared decision-making in a scalable format. This is particularly relevant in aesthetic surgery, where outcomes are inherently subjective and heavily influenced by preoperative expectations, attribution styles, and perceived social factors. Patient-facing conversational systems can present educational content, prompt reflection on goals and trade-offs, and help patients articulate realistic success criteria - functions that complement, rather than replace, clinician counseling.

Expectation management is also facilitated by advances in outcome visualization. Specifically, in rhinoplasty, image-based simulation and deep learning approaches have been investigated for generating individualized postoperative appearance predictions and standardizing patient-surgeon communication regarding plausible outcomes^[23]. Such tools could enhance preoperative understanding by reorienting discussions from idealized targets to anatomically and procedurally feasible ranges; however, their accuracy, generalizability, and impact on satisfaction necessitate rigorous validation. Concurrently, generative adversarial networks (GANs) and related generative imaging methods have been employed for synthetic data generation and the modeling of postoperative appearance, yet standardized development and evaluation frameworks are essential to ensure clinical validity, mitigate bias, and prevent the amplification of unrealistic expectations^[24].

COGNITIVE REFRAMING AND STRUCTURED SELF-HELP SUPPORT

Beyond outcome visualization, generative AI can support preoperative preparation by providing structured psychoeducation and skills-focused self-help prompts, both grounded in cognitive-behavioral principles. In aesthetic surgery candidates, common maladaptive cognitions include all-or-nothing thinking (e.g., “If my nose isn’t perfect, the surgery was a failure”), catastrophizing (e.g., “Everyone will notice the scar”), and external validation dependency (e.g., “I need this surgery to be accepted”). Conversational systems, through guided reflection, Socratic-style questioning, and alternative appraisal prompts, can help patients articulate assumptions, consider more balanced interpretations, and define realistic success criteria. These capabilities support expectation alignment and emotional readiness during decision-making. Crucially, such modules must be framed as supportive educational and self-reflection tools, not psychotherapy, and must incorporate clear boundaries and escalation pathways for concerning symptoms or crisis signals.

This strategy aligns with patient-centered care, promoting informed participation and psychological preparedness alongside surgical planning. In practice, generative AI should function as an adjunct, extending access to standardized educational content and structured reflection beyond the clinic, while preserving clinician oversight for interpretation, risk discussion, and referral decisions. To minimize harm, implementation must prioritize safety-by-design measures, including conservative language, avoidance of diagnostic labeling, transparent disclosures, and clinician-facing summaries that highlight red-flag content needing follow-up.

DESIGN FRAMEWORK FOR AN INTEGRATED AI-POWERED PREOPERATIVE SYSTEM

While components of AI-assisted screening and education have been studied in broader contexts like mental health and general surgery, there is currently insufficient data to determine if these systems can be directly applied to aesthetic surgery patients with clearly defined efficacy or implementation boundaries. Therefore, the architecture presented here should be considered a proposed conceptual framework, not a clinically validated model.

As summarized in Figure 4, an effective future AI-enhanced preoperative system can be organized into three sequential and interlinked stages: (a) initial engagement and structured intake; (b) personalized psychoeducation and expectation clarification; and (c) clinician-facing summary and consultation support. Crucially, such a system should not function as an autonomous diagnostic or decision-making tool. Instead, it must operate within a governed clinical pathway, characterized by human oversight, non-diagnostic use, privacy protection, bias awareness, transparent clinical boundaries, and clearly defined escalation triggers for concerning signals [Figure 4].

Figure 4. Conceptual framework of an integrated AI-powered preoperative system in plastic surgery. AI: Artificial intelligence.

This figure presents a proposed three-stage framework for integrating generative AI into preoperative plastic surgery care. Stage 1 involves hybrid intake, utilizing validated questionnaires and open-ended conversational input to capture patient motivations, expectations, functional impact, and body image-related concerns; Stage 2 delivers targeted psychoeducation and expectation clarification through standardized, evidence-informed support modules; Stage 3 provides clinician-facing summaries that highlight patient goals and expectations, motivational drivers, external pressures, discordance between structured scale scores and narrative cues, and prompts for follow-up discussion or referral when appropriate. The framework is conceptualized as a clinician-supervised support system, distinct from a clinically validated autonomous diagnostic or recommendation model. Its implementation must adhere to stringent safety and governance protocols, encompassing human oversight, non-diagnostic application, robust privacy protection, proactive bias awareness, clear escalation pathways, and transparent clinical boundaries.

A pragmatic architecture can be organized into three stages:
1. Initial engagement and structured intake (screening-like support): a conversational agent guides patients through a hybrid intake combining validated psychometric instruments (e.g., BDD and body image scales, as appropriate) with open-ended questions. The goal is to capture motivations, expectations, and functional impact while enabling patients to disclose sensitive concerns in a self-paced format. Outputs at this stage should be descriptive (e.g., “themes” or “flags”) rather than diagnostic labels. Predefined thresholds should trigger clinician review or referral pathways.

2. Personalized psychoeducation and expectation clarification: the system delivers targeted modules, informed by intake themes and questionnaire responses. These modules cover areas such as procedure-specific education, expectation alignment prompts, media literacy content, and structured self-help exercises grounded in cognitive-behavioral principles. For example, patients with elevated appearance-related anxiety may receive guided reflection prompts and coping skills education, while those with unrealistic expectations may be offered scenario-based trade-off explanations and visualization-guided counseling prompts. All module content must be standardized, evidence-informed, and carefully worded to avoid therapeutic claims or deterministic predictions.

3. Clinician-facing summary and consultation support: the platform generates a concise surgeon report from patient inputs, highlighting: (a) stated goals and expectations; (b) motivational drivers and external pressures; (c) discordance between scale scores and narrative cues; and (d) prompts for consultation discussion. For detected risks, the report recommends next steps (e.g., additional in-person assessment, shared decision-making aids, or mental health referral), supporting clinician judgment over automated decision-making.

PRACTICAL CLINICAL INTEGRATION AND MEDICO-LEGAL CONSIDERATIONS

To realistically incorporate such conceptual systems into everyday clinical practice, they must seamlessly align with existing preoperative workflows. Patients could interact with the AI screening tool via a secure mobile application at home or on a tablet in the waiting room before their initial consultation. This asynchronous approach saves valuable clinic time and allows patients to reflect on their motivations privately.

However, strict clinician oversight remains non-negotiable. The AI should function purely as a “triage and summary” assistant, flagging potential psychological vulnerabilities (e.g., body dysmorphic traits or unrealistic expectations) for the surgeon to explore during the face-to-face encounter. From a medico-legal perspective, integrating AI-generated outputs into patient counseling introduces novel liability risks. For instance, if an AI system fails to detect severe psychological distress, or if a surgeon overlooks an AI-generated red flag leading to postoperative psychological harm, the allocation of liability among the software developer, healthcare institution, and clinician becomes ambiguous. Therefore, clinical integration must include clear patient disclaimers: the AI is not a diagnostic tool, and ultimate responsibility for surgical clearance and psychological assessment always rests with the attending plastic surgeon.

LIMITATIONS AND ETHICAL CONSIDERATIONS

Notwithstanding its potential, the integration of generative AI into routine preoperative care presents substantial technical, ethical, and regulatory challenges, particularly in contexts involving psychologically sensitive and high-stakes decision-making. A principal concern is algorithmic bias: should training and validation data lack representativeness across demographic, cultural, and aesthetic-preference groups, model outputs may exhibit suboptimal performance for specific populations or inadvertently perpetuate existing disparities^[25,26]. Both privacy and security are paramount, given that these systems may process highly sensitive narratives pertaining to body image, mental health symptoms, and social stressors. In addition, LLMs are prone to generating inaccurate or fabricated content (referred to as “hallucinations”), which is impermissible in clinical counseling contexts absent robust safeguards, rigorous validation, and comprehensive oversight^[27,28]. Finally, the evolving regulatory landscape for AI-enabled tools - especially those used for screening, triage, or decision support - along with unclear accountability structures, may impede safe deployment^[28].

Given that preoperative psychological assessment is a high-stakes application, it is essential to move beyond general concerns and concretely specify failure modes, potential harms, and mitigation pathways that encompass both technical controls and clinical governance [Table 2]. Crucially, these mitigations must be implemented as “safety-by-design” requirements rather than optional features.

Implicit bias warrants particular attention in aesthetic surgery, where concepts of “normality”, attractiveness, and ideal outcomes are shaped by cultural and societal norms^[29]. When AI systems are trained predominantly on data reflecting a narrow demographic or aesthetic standard, they may embed biases, such as privileging certain facial proportions, skin tones, or beauty ideals. Consequently, their counseling or expectation-setting outputs might not align with diverse identities and preferences.

Addressing this challenge requires not only diverse datasets but also continuous auditing, including subgroup performance reporting and model recalibration. Furthermore, robust governance structures are essential to ensure an equity-focused review of both AI outputs and subsequent clinical actions.

In practice, fairness assessment should incorporate conservative scope boundaries, explicitly avoiding prescriptive beauty judgments. This, combined with essential clinician oversight, will help ensure that AI-mediated communication remains supportive, patient-centered, and culturally responsive.

FUTURE DIRECTIONS

To advance clinical translation, future work must transition from feasibility demonstrations to standardized validation and implementation science. We summarize key evidence gaps and propose corresponding study designs, comparators, and endpoints to support clinical validation, safety assessment, and regulatory readiness [Table 3].

To achieve multimodal and context-aware assessment, future preoperative systems must move beyond text-only interaction, which often misses clinically relevant affective and behavioral signals. Such systems could therefore explore multimodal approaches, combining language-based input with additional patient-derived cues like voice or prosodic features, facial expression dynamics, response latency, and engagement patterns. As illustrated in Figure 5, integrating these signals into an AI-assisted workflow would enable exploratory multimodal fusion, analysis of affective and behavioral patterns, detection of expectation-instability cues, integration of distress-related signals, risk flagging, and clinician-supportive interpretation. These approaches would support a more contextual understanding of distress, expectation misalignment, and decisional conflict within psychologically sensitive preoperative settings. It is crucial to note, however, that these applications represent future research directions rather than clinically validated assessment systems. Their translation into plastic surgery practice will require careful calibration, privacy-preserving data capture, transparent thresholds, thorough bias assessment, robust human oversight, and clearly governed escalation pathways^[30].

Figure 5. Conceptual framework for AI-supported multimodal preoperative assessment in plastic surgery. AI: Artificial intelligence.

This schematic illustrates a conceptual framework for future multimodal preoperative assessment support in plastic surgery. It proposes integrating various patient-derived signals - such as text narratives, voice/prosodic cues, facial expression dynamics, response latency, and engagement patterns - through exploratory AI-assisted multimodal analysis. This analysis aims to support risk flagging, expectation assessment, clinician-facing summarization, and escalation. Crucially, this framework is for conceptual illustration only; it does not represent a standalone diagnostic system, a validated psychological assessment tool, or a clinically validated autonomous platform. Any future implementation would require human oversight, privacy-preserving data capture, transparent thresholds, bias assessment, and clearly governed escalation pathways. The authors created this schematic using professional figure-preparation software based on their conceptual designs, without incorporating any third-party copyrighted material.

CONCLUSION

AI is increasingly transforming plastic and reconstructive surgery, impacting not only technical planning and prediction but also communication and decision-making in elective aesthetic care. Generative AI, particularly LLMs, shows promise as a patient-facing interface, supporting preoperative conversations about motivation, expectations, and psychosocial readiness.

Despite the common occurrence of psychological risk factors linked to postoperative satisfaction and regret, routine practice consistently shows a gap: structured assessment and preparatory support are often inconsistently delivered. This inconsistency stems from time constraints, limited specialist availability, and barriers to patient disclosure. Existing tools, while valuable, may not adequately account for surgery-specific factors such as appearance-based perfectionism, social comparison, externally driven motivation, and unstable expectations.

Emerging evidence suggests that LLM-enabled systems can facilitate screening-like intake interactions, deliver tailored psychoeducation, and structure expectation clarification in an accessible, low-stigma format. Combining validated psychometric instruments with open-ended dialogue in hybrid designs may help contextualize patient concerns and generate clinician-facing summaries, guiding targeted discussion during consultation. These capabilities are especially relevant for high-volume settings, where asynchronous engagement can extend support beyond the clinic visit.

However, psychologically sensitive applications necessitate conservative scope boundaries and safety-by-design governance. Risks such as hallucinated content, bias (including aesthetic normativity), privacy exposure, and overreliance demand human oversight, clear escalation pathways, transparent disclosure, and ongoing auditing. Crucially, generative AI should serve as an adjunct that augments - rather than replaces - clinical judgment, empathy, and referral decision-making.

Future progress hinges on multidisciplinary collaboration to establish standardized validation endpoints, implementation protocols, and governance frameworks that align with evolving regulatory expectations. Through rigorous prospective evaluation and responsible deployment, generative AI can become a clinically meaningful component of preoperative care, supporting psychological preparedness and patient agency as integral contributors to positive aesthetic surgical outcomes^[31].