Large Language Model‐Assisted Shared Decision‐Making (SDM) in Rhinology: A Randomized Controlled Pilot Study Authors: Omer Baker, Clifford J He, Andrew Yousef, Vivienne M Li, Adam S DeConde, Carol H Yan Published: 2026 ABSTRACT ABSTRACTObjectiveShared decision‐making (SDM) integrates patient preferences into evidence‐based care. Generative AI tools may offer a novel approach to facilitate SDM in rhinology, but their clinical utility has not been assessed.MethodsIn this prospective randomized pilot study conducted at a single tertiary academic rhinology clinic, English‐speaking adult patients with medically refractory chronic rhinosinusitis (CRS) and/or nasal obstruction were randomized to use either ChatGPT‐4 or Google to ask treatment‐related questions. The primary outcome was change in Decisional Conflict Scale (DCS) score. Secondary outcomes were changes in a treatment‐related knowledge questionnaire (MCQ), system usability scale (SUS) scores, and accuracy ratings of ChatGPT responses.ResultsA total of 57 participants completed the study (ChatGPT arm = 29; Google arm = 28). Both groups showed significant reductions in DCS scores with a mean difference of −4.3 in the ChatGPT arm (95% CI: −7.1 to −1.5, p = 0.004) and −2.7 in the Google arm (95% CI: −4.3 to −1.1, p = 0.002), with no significant difference between groups. Participants uncertain about treatment at baseline had higher decisional conflict both pre‐and post‐intervention, and demonstrated less improvement than those who preferred surgery. Treatment preferences and MCQ performance remained unchanged post‐intervention in both groups. SUS scores were comparable between platforms, and ChatGPT's responses were rated as moderately accurate (7.2/10, SD = 2.0).ConclusionSDM using ChatGPT and Google may reduce decisional conflict in patients considering sinus surgery, though patients with baseline uncertainty may require additional support. Larger, multi‐center trials are needed to validate these tools and explore their integration across diverse settings.Level of EvidenceLevel II.Keywords: ChatGPT sinusitis outcomes, chronic rhinosinusitis, GenAI, patient satisfaction, rhinitis, shared decision making Introduction The decision to pursue surgical intervention for medically refractory chronic rhinosinusitis (CRS) and/or nasal obstruction is highly complex and sensitive to individual patient preferences. Recent studies found that meeting preoperative expectations is the most important predictor of postoperative patient satisfaction after sinus surgery, emphasizing the need for effective and comprehensive patient counseling methods [1]. Shared decision‐making (SDM) is a well‐established and increasingly popular model for educating patients and assisting with decisions related to undergoing surgery. The collaborative model can be distilled into three physician‐guided steps: introduce choice, describe options, and help patients explore preferences to make a decision [2]. SDM has been demonstrated to improve postoperative patient satisfaction, and reduces decisional conflict and regret [3, 4, 5]. However, several practical barriers hinder its widespread adoption, including time constraints, provider motivation, and discomfort handling disagreements during SDM [6]. Given the inherent complexity in balancing surgical versus medical management in rhinology, especially with variable patient preferences and uncertainty regarding CRS treatment outcomes, there is a particular need to enhance patient counseling and decision‐making clarity. First launched to the public in November 2022, ChatGPT (Open AI, San Francisco, CA) is a free AI‐driven chatbot that utilizes machine learning algorithms to engage in humanlike conversations with its users. ChatGPT's role in surgical counseling is being explored across nearly all disciplines, including head and neck surgery, where it has been reported to provide patient education comparable to standard‐of‐care [7, 8]. While it has not been approved as a stand‐alone clinical resource for patients, it may serve as a useful adjunct to physician counseling for patient education. No studies to date have investigated the efficacy of SDM with ChatGPT in the field of rhinology. While traditional online search engines (e.g., Google) offer immediate information, their completeness and ability to address clinical questions may be highly variable and more difficult to navigate. Thus, this prospective, randomized controlled trial aimed to investigate whether SDM facilitated by ChatGPT could more effectively decrease decisional conflict compared to traditional internet searches. We hypothesized that the integration of ChatGPT in SDM would decrease decisional conflict in patients with medically recalcitrant CRS and nasal obstruction. Materials and Methods Study Design and Setting In a randomized controlled trial (UC San Diego Institutional Review Board #810119), English‐speaking adult patients with medically refractory chronic rhinosinusitis (CRS) and/or nasal obstruction were recruited from a single outpatient academic rhinology clinic, between June 2024 and December 2024. All participants met eligibility criteria for surgical intervention, had undergone one‐on‐one counseling with one of two faculty rhinologists, and were actively weighing the decision whether to pursue surgery. The study intervention was limited to a single clinic encounter and occurred immediately after routine preoperative counseling, without any alterations to the typical clinical workflow. Data Collection Prior to intervention, all consented participants completed a validated 16‐item Decisional Conflict Scale (DCS) using the Statement Format (Table S1) [9]. Participants with a diagnosis of CRS also completed a 4‐item multiple‐choice questionnaire (MCQ), assessing their understanding of their CRS diagnosis, surgical and medical treatment options, and associated risks and benefits [10]. The questions evaluated the appropriate timeframe for surgery after failed medical therapy, the likelihood of symptom improvement after surgery, the most common risks associated with sinus surgery, and which steps (e.g., nasal irrigation) optimize post‐operative outcomes. Self‐reported demographic data and medical history were also collected via survey from all participants. Participants were randomized 1:1 using the NIH Clinical Trial Randomization Tool to either ChatGPT (GPT‐4) or the traditional internet search engine (Google) arm (with the generative AI features such as Gemini disabled) [11]. All participants were allowed to freely interact with the tools and were given 15 min to anonymously ask at least three treatment‐related questions of their choosing about their proposed surgical treatment, without standardized prompts, to reflect naturalistic engagement. Post‐intervention, participants repeated the DCS and MCQ surveys in addition to completing the validated 10‐item System Usability Scale (SUS) that assesses patients' perceived usability of their assigned intervention as a platform for education; SUS scores range from 0 to 100, with higher scores indicating higher usability among study participants [12]. Additionally, both before and after intervention, participants were asked to classify their treatment preference as either a preference for surgical therapy, medical therapy, no therapy, or unsure. Outcomes and Statistical Analysis The primary outcome measure was change in decisional conflict scale (DCS) score, which ranges from 0 (no decisional conflict) to 100 (extremely high decisional conflict). Secondary outcome measures included changes in treatment‐related knowledge (4‐item questionnaire, MCQ), system usability scale (SUS) score, and rhinologist‐assessed accuracy of ChatGPT's responses using a 0–10 Likert scale. Primary analysis aimed to compare the percent change in pre‐and post‐intervention DCS score, both within and between the ChatGPT and Google arms. Due to no demographic or clinical factors being identified as significant confounders, univariate t‐tests were used. Cohen's d was calculated to estimate effect sizes. Secondary analyses included Fisher's exact tests to assess changes in treatment‐related knowledge (MCQ) and post‐intervention treatment preferences, in addition to independent t‐tests to compare SUS scores between the ChatGPT and Google arms. Additionally, similar subgroup analyses were performed comparing participants who were unsure about their preferred intervention to those who preferred surgery. All statistical analyses were conducted in STATA (StataCorp LLC, Texas, USA), with statistical significance set at p < 0.05. Deidentified source data may be made available upon reasonable request. Results Of 95 patients who were screened for trial eligibility, 60 participants were recruited for enrollment and 57 were included in the analysis with 29 in the ChatGPT group and 28 in the Google group as depicted in the CONSORT diagram (Figure 1). Three participants withdrew from the study due to a reported lack of time for study completion. At baseline, there were no significant demographic differences observed between the two groups in terms of age, gender, race, household income, or highest level of education (Table 1). Additionally, clinical characteristics, such as SNOT‐22 scores, Lund‐Mackay scores, and Charlson‐Comorbidity Index (CCI), were comparable between the two groups. CONSORT diagram of participants. TABLE 1Demographic and clinical characteristics of study participants.ChatGPT group (n = 29)Google group (n = 28) p Age in years, mean (SD)50.1 (20.6)51.9 (18.5)0.727Gender, n (%)0.232Male17 (58.6%)11 (39.3%)Female12 (41.4%)16 (57.1%)Non‐binary0 (0.0%)1 (3.6%)Race, n (%)0.663White14 (48.3%)18 (64.3%)Hispanic/Latinx6 (20.7%)5 (17.9%)Asian4 (13.8%)2 (7.1%)Black/African American3 (10.3%)3 (10.7%)Other2 (6.9%)0 (0.0%)Household income, n (%)< $50,0002 (6.9%)8 (28.6%)$50,000–$99,9996 (20.7%)3 (10.7%)$100,000–$199,9998 (27.6%)7 (25.0%)> $200,0004 (13.8%)5 (17.9%)Highest level of education, n (%)Did not graduate high school1 (3.4%)1 (3.6%)Completed high school3 (10.3%)3 (10.7%)Some college5 (17.2%)4 (14.3%)Completed college degree11 (37.9%)8 (28.6%)Advanced degree7 (24.1%)10 (35.7%)Indication for ESS, n (%)0.575Chronic sinusitis5 (17.2%)8 (28.6%)Nasal obstruction8 (27.6%)7 (25.0%)Both16 (55.2%)13 (46.4%)SNOT‐22, mean (SD)41.9 (22.2)34.2 (24.7)0.284History of allergies, n (%)19 (65.6%)13 (46.4%)0.186History of asthma, n (%)7 (24.1%)6 (21.4%)0.529Charlson‐comorbidity index (CCI) score, mean (SD)2.1 (2.4)1.9 (1.7)0.609Lund‐Mackay (LM) score, mean (SD)6.8 (6.2)6.7 (3.7)0.931Number of previous ENT providers seen for condition, mean (SD)1.6 (2.0)1.4 (1.5)0.737Number of antibiotic courses in the past year, mean (SD)1.1 (1.4)1.4 (1.7)0.598Unsure about treatment preference, n (%)8 (27.6%)5 (17.9%)0.530 Demographic and clinical characteristics of study participants. Following intervention, participants in both groups demonstrated a significant reduction in total DCS scores (Table 2). A mean difference of −4.31 (95% CI: −7.12 to −1.50, p = 0.004) was observed in the ChatGPT group, and −2.68 (95% CI: −4.25 to −1.11, p = 0.002) in the Google group. Analysis of the informed subscore revealed significant improvement for both groups, with a mean difference of −4.31 (95% CI: −8.34 to −0.28, p = 0.037) in the ChatGPT group and −3.57 in the Google group (95% CI: −6.41 to −0.73, p = 0.016). Similar trends were observed in the values clarity subscore, with a mean difference of −6.61 (95% CI: −11.50 to −1.71, p = 0.010) in the ChatGPT group and −3.87 (95% CI: −7.22 to −0.52, p = 0.025) in the Google group. The uncertainty subscore significantly improved with a mean difference of −7.18 (95% CI: −12.59 to −1.78, p = 0.011) in the ChatGPT group, whereas the Google group showed no significant change (95% CI: −6.17 to 0.22, p = 0.067). TABLE 2Decisional conflict scale (DCS) scores pre‐and‐post intervention, ChatGPT and Google arms (within groups).Pre‐interventionPost‐interventionMean difference (95% CI) p ChatGPT group (n = 29)Total DCS score17.5 (14.7)13.2 (13.7)−4.31 (−7.12, −1.50) 0.004 Uncertainty subscore24.1 (21.6)17.0 (17.5)−7.18 (−12.59, −1.78) 0.011 Informed subscore16.4 (13.8)12.1 (14.0)−4.31 (−8.34, −0.28) 0.037 Values clarity subscore17.8 (16.5)11.2 (14.3) −6.61 (−11.50, −1.71) 0.010 Support subscore12.1 (12.7)9.5 (12.3)−2.59 (−5.41, 0.24)0.071Effective decision subscore17.2 (16.7)15.5 (15.9)−1.72 (−4.09, 0.64)0.147Google group (n = 28)Total DCS score14.1 (14.4)11.4 (14.7)−2.68 (−4.25, −1.11) 0.001 Uncertainty subscore18.8 (23.6)15.8 (21.8)−2.98 (−6.17, 0.22)0.067Informed subscore12.8 (12.1)9.2 (12.3)−3.57 (−6.41, −0.73) 0.016 Values clarity subscore12.5 (14.8)8.6 (13.9)−3.87 (−7.22, −0.52) 0.025 Support subscore10.1 (13.5)8.9 (13.8)−1.19 (−4.33, 1.95)0.443Effective decision subscore15.6 (18.0)13.6 (17.6)−2.01 (−4.73, 0.71)0.142 Note: p values reflect within‐group changes pre versus post‐intervention, using student t‐tests. Higher scores reflect higher levels of decisional conflict. Bold values indicate statistical significance (p < 0.05). Decisional conflict scale (DCS) scores pre‐and‐post intervention, ChatGPT and Google arms (within groups). Note: p values reflect within‐group changes pre versus post‐intervention, using student t‐tests. Higher scores reflect higher levels of decisional conflict. Bold values indicate statistical significance (p < 0.05). When comparing the percent change from baseline, the ChatGPT group showed a 35.8% decrease in total DCS scores compared to 19.9% in the Google group (mean difference −15.9%, 95% CI: −35.6% to 3.8%). However, the degree of improvement in total DCS scores and in the subscores was not statistically different between the ChatGPT and Google groups (Table 3), suggesting that both interventions were similarly effective in reducing decisional conflict. TABLE 3Change in decisional conflict scale (DCS) scores following ChatGPT and Google intervention.Change in DCS score (%) post‐interventionChatGPTArm (n = 29)Google arm (n = 28)Mean difference (95% CI) p Cohen's d Total DCS score−35.8%−19.9%−15.9% (−35.6%, 3.8%)0.1110.43Uncertainty subscore−23.4%−11.6%−11.8% (−30.0%, 6.4%)0.2000.34Informed subscore−18.7%−22.9%4.2% (−25.6%, 34.1%)0.777−0.08Values clarity subscore−21.8%−16.3%−5.5% (−27.9%, 16.9%)0.6260.13Support subscore−11.2%−5.0%−6.2% (−32.7%, 20.3%)0.6390.12Effective decision subscore−8.0%−7.6%−0.4% (−13.5%, 12.8%)0.9550.01 Note: Values represent mean percent change in total and subscale DCS scores following intervention; p‐values reflect between‐group differences assessed via independent t‐tests. Positive values in Cohen's d indicate greater improvement in decisional conflict in the ChatGPT group compared to Google. Change in decisional conflict scale (DCS) scores following ChatGPT and Google intervention. Note: Values represent mean percent change in total and subscale DCS scores following intervention; p‐values reflect between‐group differences assessed via independent t‐tests. Positive values in Cohen's d indicate greater improvement in decisional conflict in the ChatGPT group compared to Google. Prior to intervention, participants classified their treatment preference as either a preference for surgical therapy, medical therapy, no therapy, or unsure. Neither intervention (ChatGPT or Google) resulted in a change of patient treatment preference, as shown in Table S2. To assess the impact of baseline treatment preference on decisional conflict, we compared participants who were unsure about their preferred intervention (n = 13) to those who preferred surgery (n = 36). Participants in the “unsure” group started with significantly higher total DCS scores of 30.8 compared to 9.8 in the pro‐surgery group, with a mean difference of −21.0 points (95% CI: −28.2 to −13.8, p < 0.001). This difference persisted despite intervention, with post‐intervention scores of 29.3 versus 7.0 and a mean difference of −22.3 points (95% CI: −29.4 to −15.3, p < 0.001). When examining percent change from baseline, participants who initially preferred surgery experienced significantly greater reductions in overall decisional conflict (−30.7% vs. −4.3%, p = 0.003), particularly in the informed (−32.9% vs. 22.6%, p = 0.006) and support subscores (−10.9% vs. 22.6%, p = 0.027; Table 4, Figure 2). Interestingly, those who were undecided at baseline demonstrated an increase in decisional conflict in the informed and support domains, suggesting that shared decision‐making using either ChatGPT or Google was less effective at reducing conflict for participants who remain uncertain about their treatment path. TABLE 4Change in decisional conflict scale (DCS) scores pre‐and‐post intervention, unsure versus pro‐surgery group.Percent (%) difference in outcome post‐interventionPro‐surgery group (n = 36)Unsure group (n = 13)Mean difference (95% CI) p Cohen's d Total DCS score−30.7%−4.3%−26.3% (−48.1%, −3.87%) 0.023 0.76Uncertainty subscore−12.5%−13.2%0.7% (−19.0% to 20.4%)0.943−0.02Informed subscore−32.9%17.3%−50.2% (−85.1%, −15.3%) 0.006 0.94Values clarity subscore−17.1%−10.1%−7.0% (−33.1%, 19.1%)0.5920.17Support subscore−10.9%22.6%−33.5% (−62.9%, −4.1%) 0.027 0.74Effective decision subscore−9.7%0.3%−10.0% (−25.9%, 6.0%)0.2150.41 Note: p values were calculated using Student's t‐tests. Positive values in Cohen's d indicate greater improvement in decisional conflict in the pro‐surgery group compared to those who were unsure. Bold values indicate statistical significance (p < 0.05). Change in decisional conflict scale (DCS) scores pre‐and‐post intervention, unsure versus pro‐surgery group. Note: p values were calculated using Student's t‐tests. Positive values in Cohen's d indicate greater improvement in decisional conflict in the pro‐surgery group compared to those who were unsure. Bold values indicate statistical significance (p < 0.05). Percent change in decisional conflict scores (DCS) pre‐and‐post intervention, based on baseline treatment preference. System Usability Scale (SUS) scores were determined for each group to assess participant comfort level and acceptance of each platform to help with SDM. We found that SUS scores were comparable between the two groups (ChatGPT: 67.3 ± 16.9, Google: 66.4 ± 13.5, p = 0.826), as were their subdomains: ease of use (ChatGPT: 31.6 ± 11.3, Google: 32.3 ± 10.3, p = 0.789) and learnability (ChatGPT: 35.8 ± 8.8 vs. Google: 34.1 ± 7.9, p = 0.457). Overall system usability scores for ChatGPT are similar to previously published data, with a raw score equivalent to the 50th percentile and considered at high marginal acceptability or just below the threshold for good usability [13, 14]. The SUS score to assess the usability of a traditional search engine (Google) was very similar. No significant change in treatment‐related MCQ performance was observed in either group after intervention (p = 0.640). Additionally, using a 10‐point Likert scale, fellowship trained rhinologists independently and blindly evaluated ChatGPT answer accuracy. Across a total of 56 responses, the average accuracy rating was 7.2 (SD: 2.0). Discussion In this pilot study, we evaluated the feasibility of using a generative AI program (ChatGPT) in shared decision‐making (SDM) among rhinology patients by comparing its efficacy to traditional search engines such as Google. Both ChatGPT and Google interventions were associated with a reduction in decisional conflict, suggesting that patient‐facing digital tools may offer value in enhancing informed and values‐based treatment decisions. No significant differences were found between groups, suggesting that artificial intelligence (AI) conversational tools may offer a comparable degree of decisional support to traditional internet searches in the context of surgical counseling. Interestingly, participants who were unsure about their treatment preference at baseline had substantially higher decisional conflict across all domains, both before and after the intervention. Moreover, while both the unsure and pro‐surgery groups demonstrated reductions in decisional conflict following intervention, the degree of improvement was significantly greater among those who initially preferred surgery. Paradoxically, participants in the unsure group exhibited increased decisional conflict in the informed and support subdomains, suggesting that exposure to information without sufficient guidance may heighten confusion or reinforce ambivalence. Despite these changes in decisional conflict scores, neither group demonstrated significant shifts in treatment preference post‐intervention, indicating that informational tools alone may be insufficient to alter clinical decision‐making in the absence of personalized counseling. This subgroup may represent a key population for targeted interventions, potentially incorporating structured clinician guidance or hybrid decision‐making approaches to better support patients with high baseline uncertainty. Previous research on various SDM interventions, such as decision aids (DAs) and question prompt lists (QPLs), has demonstrated similar achievements in reducing decisional conflict [15, 16]. To our knowledge, no previous studies have examined the real‐world integration of ChatGPT in SDM, making this one of the first to explore its use in rhinology and surgical counseling. However, our findings are supported by studies in other fields that have demonstrated a positive impact of integrating AI‐based tools into surgical consultations. In a randomized controlled trial of patients undergoing total knee replacement, an AI‐enabled, personalized decision aid significantly improved decision quality, level of SDM, and satisfaction among patients [17]. Similarly, ChatGPT‐assisted informed consent has been shown to reduce perioperative anxiety and improve satisfaction among patients undergoing total knee arthroplasty, as well as enhance the efficiency of consultations in pre‐vasectomy counseling [18, 19]. This study builds on our prior work demonstrating the utility of structured SDM interventions in surgical counseling and suggests that conversational AI may be a similarly useful tool, especially in settings where personalized advice may address patient concerns more effectively than fixed‐format tools (e.g., decision aids) [10]. Traditional search engines such as Google have been used for nearly two decades by both patients and physicians to access medical information, with early studies on Google debating as a diagnostic aid dating as far back as 2006 [20, 21, 22]. Notably, recent research has revealed that search engines only answered 50%–70% of health questions correctly, while large language models (LLMs) correctly answered about 80% of questions [23]. While many studies have demonstrated ChatGPT fares better than Google in patient education, the accuracy of ChatGPT in rhinology remains disputed. In our study, ChatGPT's accuracy in answering health‐related queries was rated as 7.2/10 (moderately accurate), faring somewhat similarly to other studies that have reported accuracy over 80% [24]. In contrast, a recent study found that 80% of ChatGPT's responses were at least partially incorrect, misleading, or incomplete with a mean reliability score of 1.65/3 [25]. While these discrepancies may partly stem from the rapid evolution of ChatGPT's technology (e.g., GPT3.5 vs. GPT4), further investigation is warranted to examine ChatGPT's reliability and safety in real‐world use, particularly given evidence surrounding its propagation of harmful biases and race‐based misconceptions [26, 27]. Additionally, study participants assigned similar System Usability Scale (SUS) scores to both ChatGPT and Google, with a 67.3 and 66.4 respectively. Given a SUS score of 68 is considered by many as the benchmark for usability of digital health tools, this suggests potential barriers in independently accessing these technologies among patients seeking additional counseling [28, 29]. Potential contributing factors include limited familiarity in navigating technology, skepticism toward their use for health‐related questions, or broader challenges in accessing technology. Future improvements in ChatGPT's usability may be expected with its integration of new features, such as image and voice interaction [30]. While this study demonstrates the feasibility of using ChatGPT in SDM, several limitations exist. Our study was a single‐center pilot study with a relatively small sample size, which limits generalizability and our power to detect subtle effects between ChatGPT and Google. However, several findings with moderate to large effect sizes (Cohen's d) provide preliminary evidence demonstrating potential benefits of generative AI tools in SDM. Although participants were provided instruction on how to use each platform, several key metrics (especially within the Google arm) were not captured, including the content of their queries, the number of links accessed, and how they engaged with the information. There may have been individual differences in how participants search for and process information online that contributed to the observed outcomes, alongside the inherent effects of the actual platforms. Additionally, the participants completed the intervention and follow‐up assessments in a single clinical encounter, limiting the time available to process information, reflect on values, or discuss options with family or others. The intervention was administered immediately following one‐on‐one counseling with a rhinologist, which likely lowered baseline decisional conflict and may have reduced our ability to detect between‐group differences. However, this design reflected standard clinical practice and ensured that the interventions were evaluated within a real‐world decision‐making context, allowing assessment of whether large language models provide additional decisional support beyond routine physician counseling. Conducting the intervention within the medical office, rather than at home or over an extended period, may have constrained the depth of engagement with the decision‐making tools and influenced decisional conflict scores and treatment preference. Lastly, we did not assess participants' perception of clarity or trustworthiness of the information provided beyond SUS scores, which would have helped provide important context for a patient‐centered interpretation of these interventions. Conclusion Overall, the decision to undergo surgery for medically refractory CRS and/or nasal obstruction is often complex and highly sensitive to patient preferences. Our study provides early evidence that an AI‐driven conversational agent such as ChatGPT may support SDM among rhinology patients by reducing decisional conflict and improving patient confidence about their treatment options. However, the reliability of utilizing ChatGPT in clinical practice, as well as its effectiveness in direct comparison to standalone counseling or other tools such as DAs, remains uncertain. Further research across diverse patient populations is warranted to refine its integration into clinical workflow, compare its effectiveness to standalone physician‐led counseling, improve our understanding of the patient perspective, and address the safety and ethical concerns surrounding its implementation. Funding Baker received the UC San Diego School of Medicine Summer Research Fellowship to help support the research, authorship, and/or publication of this article. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Table S1: Decisional conflict scale (DCS) survey completed by study participants. Table S2: Preferred treatment pre‐and‐post intervention, ChatGPT and Google arms. Baker O., He C. J., Yousef A., Li V. M., DeConde A. S., and Yan C. H., “Large Language Model‐Assisted Shared Decision‐Making (SDM) in Rhinology: A Randomized Controlled Pilot Study,” Laryngoscope Investigative Otolaryngology 11, no. 3 (2026): e70440, 10.1002/lio2.70440. This manuscript was presented as an oral presentation at the American Rhinologic Society COSM meeting on May 17th, 2025, in New Orleans, LA. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request. References Associated Data Supplementary Materials Table S1: Decisional conflict scale (DCS) survey completed by study participants. Table S2: Preferred treatment pre‐and‐post intervention, ChatGPT and Google arms. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.