Abstract
Background/Objectives: Large language models (LLMs) such as ChatGPT, Gemini and Claude are increasingly used by patients to research new diagnoses, including testicular cancer. The aim of this study was to evaluate and compare the quality of responses generated by these three platforms to patient-facing frequently asked questions on testicular cancer, including a focused evaluation of fertility and sexual health questions. Methods: Fifteen questions were developed from established patient education resources and refined by senior urology consultant review. Each question was posed to ChatGPT, Gemini and Claude using a standardised patient-perspective prompt. The 45 responses were de-identified, randomised and assessed in a single consensus meeting by a urology panel using the Quality Analysis of Medical Artificial Intelligence (QAMAI) tool. Comparative analysis used the Friedman test with Bonferroni-corrected pairwise Wilcoxon signed-rank testing. Results: All three models produced moderate quality responses, with a mean total QAMAI score of 21.8 for ChatGPT, 22.4 for Claude, and 22.1 for Gemini (Friedman p = 0.246). Clarity differed significantly between models (adjusted p = 0.022), with Claude and Gemini outperforming ChatGPT. In the fertility and sexual health domain, Claude (24.0) and Gemini (23.0) outperformed ChatGPT (19.7). No response from any model spontaneously cited references or directed the reader to external resources. Conclusions: ChatGPT, Gemini and Claude deliver broadly equivalent overall quality on patient-facing testicular cancer questions, with inter-model differences in clarity and in the fertility and sexual health domain. The absence of source citation is a clinically relevant limitation. These tools should be regarded as a supplement to, rather than a substitute for, specialist counselling.
1. Introduction
Testicular cancer is the most common solid malignancy in men aged 15 to 44 years and accounts for approximately 1% of all male cancers globally [1]. The incidence has risen steadily over recent decades and is projected to continue rising worldwide [2]. Broader treatment availability and earlier diagnosis have kept mortality low [3]; nevertheless, a new diagnosis remains a profoundly distressing event for a young patient cohort at a formative stage of their personal, occupational, and reproductive lives [4]. From the point of diagnosis, fertility concerns, body image, sexual function, and long-term survivorship assume central importance [5].
The demographic affected by testicular cancer is also among those most likely to seek health information online [6]. Young adults consistently report higher rates of digital health-information seeking than older patient groups [7,8], and the sensitive nature of genitourinary symptoms may further encourage private online research in preference to clinical contact [9]. An increasingly popular source of such information is artificial intelligence (AI), particularly large language models (LLMs) such as ChatGPT (OpenAI), Claude (Anthropic), and Gemini (Google) [10]. Trained on extensive text corpora, LLMs are capable of generating coherent, conversational responses to medical queries, summarising clinical information, and offering diagnostic and treatment guidance [11].
However, LLMs function by predicting the most probable subsequent word on the basis of statistical patterns within their training data, meaning that although outputs appear coherent and contextually appropriate, they may be inaccurate or fabricated [12]. Such inaccuracies may carry serious consequences in medicine, particularly when LLMs are increasingly the first source patients consult after a new diagnosis, before any clinical encounter takes place [13]. The quality of LLM-generated information is therefore clinically consequential, because the information shapes how patients understand their disease, the questions they bring to consultation, and the concerns they have already begun to form before being seen. This is particularly pertinent in testicular cancer, where counselling is dominated by considerations of fertility preservation and the time-critical need for sperm cryopreservation [14], body image concerns following loss of a testicle [15], and the potential for long-term impacts on sexual and hormonal function [16]. These domains demand individualised, psychosocially attuned, and time-critical guidance, qualities for which the statistical nature of LLMs may be poorly suited.
A growing body of evidence has begun to evaluate LLMs in urological oncology, with comparative studies in prostate cancer and broader urological topics showing meaningful inter-model performance variation despite superficially similar outputs [17,18]. Question provenance appears to influence measured LLM performance. Beyatli et al. [19] evaluated ChatGPT-4 against both guideline-derived and patient-focused questions in upper tract urothelial carcinoma and reported higher accuracy on the patient-derived question set than on guideline-based queries, underscoring the importance of evaluating LLMs against the questions that patients actually ask. Within testicular cancer specifically, two prior studies have evaluated AI-generated information. Uysal et al. [20] previously compared two versions of ChatGPT using guideline-derived questions, but the question set was clinician-derived rather than reflecting what patients actually ask, and the comparison was limited to a single LLM platform. More recently, Lucas et al. [21] evaluated four AI chatbots (ChatGPT 4.0, Bing AI, ChatSonic and Perplexity) for testicular cancer patient information using the DISCERN, Patient Education Materials Assessment Tool (PEMAT) and Flesch–Kincaid instruments. However, these instruments were developed and validated for written consumer health information and printable patient education materials, rather than for AI-generated content, and they did not examine performance in fertility, hormonal and sexual health domains central to counselling this cohort. To our knowledge, no prior study has evaluated AI-generated testicular cancer information using a tool specifically validated for the assessment of medical artificial intelligence, nor directly compared ChatGPT, Gemini and Claude in this disease setting.
The aim of the present study was therefore to evaluate and compare the quality of responses generated by ChatGPT, Gemini and Claude to patient-facing frequently asked questions on testicular cancer, including a focused evaluation of fertility, hormonal and sexual health questions.
2. Methods
2.1. Study Design
We performed a cross-sectional, blinded comparative evaluation of responses generated by three publicly available LLMs, ChatGPT, Gemini and Claude, to standardised questions simulating a clinical consultation for men newly diagnosed with testicular cancer. Responses were assessed in a single consensus meeting by a panel of urology clinicians using the Quality Analysis of Medical Artificial Intelligence (QAMAI) tool [22], a six-item instrument developed and validated specifically for the evaluation of health information generated by AI platforms, with high reported internal consistency and inter-rater reliability, supporting its use in this setting. The study was exempt from ethics approval as it involved no human participants or identifiable patient data.
2.2. Selection of LLMs
ChatGPT, Gemini and Claude were selected on the basis of public availability, free-tier access and widespread consumer use. Free-access versions were used to reflect the platforms most likely to be accessed by patients. All queries were performed on 14 May 2026 using the free-tier web interfaces of ChatGPT (v5.5, OpenAI, San Francisco, CA, USA), Gemini (v3.1, Google, Mountain View, CA, USA) and Claude (Sonnet 4.6, Anthropic, San Francisco, CA, USA), accessed from Sydney, Australia. All queries were performed using a single account for each model to minimise account-level variation, and chat history, memory and personalisation features were disabled to prevent carry-over between queries.
2.3. Question Development
Questions were developed through a structured process designed to ensure that question sets reflected what patients actually ask. First, frequently asked questions were extracted from patient education resources by two authors (YB, JY), including the American Cancer Society, Cancer Council Australia, Moffitt Cancer Centre and Henry Ford Health testicular cancer information pages. Recurring themes were identified across sources and consolidated into 12 core questions reflecting the topics most frequently raised by patients on these online resources. Questions were derived from published patient education resources rather than from direct patient consultation. Additional questions were added to address clinically important issues that may be underrepresented in generic patient information, including sperm cryopreservation, testosterone function, sexual health and testicular prosthesis. The final question list was reviewed by a senior consultant urologist (PR) to ensure questions reflected real questions that patients posed during counselling and refined before data collection.
The final question set comprised 15 questions grouped thematically into five domains. These were basics and risk (3 questions), symptoms and diagnosis (3 questions), treatment (4 questions), prognosis and survivorship (2 questions), and fertility and sexual health (3 questions). The full question set is presented in Table 1.
Table 1.
Full list of final questions grouped by thematic domain.
2.4. Data Collection
To simulate the perspective of a newly diagnosed patient and ensure consistency across models and questions, each query was preceded by a standardised prompt: “I am a patient newly diagnosed with testicular cancer. In 200 words or less please tell me, [insert question]”. All queries were performed on a single day, with a new conversation initiated for each question to avoid contextual carry-over. The standardised prompt and question were entered verbatim, and the first response generated by each model was saved without regeneration, editing, follow-up prompts or clarification.
All 45 responses were compiled into a single rating document. Reponses were de-identified, and model-specific phrases identifying the source platform were redacted without alteration of substantive clinical content, and the order of model responses was randomised within each question, with responses labelled A, B, and C in a randomised sequence specific to each question. A master key linking response labels to source models was retained by the lead author (YB) and was not shared with the rating panel until rating was complete.
2.5. Rating Procedure
Responses were assessed using the QAMAI tool, which comprises six items, each scored on a 5-point Likert scale from 1 (strongly disagree) to 5 (strongly agree). The six items assess accuracy, clarity, relevance, completeness, sources and usefulness of the response. Item scores are summed to yield a total QAMAI score ranging from 6 to 30, with higher scores indicating higher quality. Total scores correspond to five quality grades as defined in the original validation, rated as very poor (6–11), poor (12–17), moderate (18–23), good (24–29) and excellent (30) [22].
Rating was undertaken in a single-session quality consensus meeting by a urology panel blinded to the source model of each response, following the consensus methodology described by Pompili et al. [17]. The panel comprised one consultant urologist (PR) and three urology registrars (JY, JS, SL) at varying stages of training. For each response, panel members provided an initial individual score across all six QAMAI items in writing before discussion, to limit the influence of senior voices on initial scoring. The response was then opened for discussion among the panel, with final scores assigned after agreement among raters, with individual scores of raters not recorded. Discussion focused on clinical accuracy, omissions, misleading statements, appropriateness for a newly diagnosed patient, safety-netting and psychosocial relevance. Any principal qualitative concerns raised by the consensus panel for any particular answer were recorded.
A focused descriptive analysis was undertaken for each of the three questions in the fertility and sexual health domain, given their clinical importance to this patient population and anticipated challenges LLMs may face in this area. For Question 13 (sperm cryopreservation), responses were assessed against four pre-specified clinical elements: whether the response (a) recommended sperm banking, (b) conveyed urgency, (c) recommended banking before surgery or further treatment, and (d) acknowledged the risk of infertility with subsequent treatment. For Question 14 (testosterone and testosterone replacement therapy [TRT]) and Question 15 (sexual function and testicular prosthesis) the consensus panel’s qualitative observations during scoring were captured and recorded descriptively, including any clinically significant inclusions or omissions.
2.6. Statistical Analysis
Descriptive statistics were used to summarise QAMAI scores. Continuous scores are reported as mean, and median with interquartile range (IQR). Categorical quality grades are reported as frequencies and percentages. Mean scores for each of the six QAMAI items were reported separately, along with proportion of responses falling into each quality grade per model. Mean total QAMAI scores were calculated per domain per model.
Comparative analysis across the three models was performed using the Friedman test for related samples and for the QAMAI domains, treating each question as a matched observation across models. Where significant differences were identified, post hoc pairwise Wilcoxon signed-rank tests with Bonferroni correction were planned. A two-sided p-value of less than 0.05 was considered statistically significant. A pre-specified domain-level analysis was performed by grouping questions into the five domains described above.
Statistical analysis was performed using RStudio (v2024.12.1+563; Posit Software, PBC, Boston, MA, USA, 2025).
3. Results
3.1. Overall Performance
ChatGPT, Claude and Gemini all successfully generated a response to each of the 15 questions (Supplementary Table S1). Mean total QAMAI scores across the question set were 21.8 for ChatGPT (median 22, range 18–25), 22.40 for Claude (median 23, range 18–25), and 22.13 for Gemini (median 22, range 16–24). All three models fell within the moderate quality grade on average, with no model reaching the good quality threshold (≥24) overall. Comparative analysis across all 15 questions showed no statistically significant difference in total QAMAI score between models (Friedman χ2 = 2.81, p = 0.25).
When responses were categorised by quality grade, ChatGPT produced 3 good and 12 moderate responses, Gemini produced 4 good, 10 moderate and 1 poor response, and Claude produced 5 good and 10 moderate responses (Table 2). No response from any model reached excellent or very poor quality.
Table 2.
QAMAI quality grades by model.
3.2. Per-Item Analysis and Comparative Testing
Mean per-item scores were broadly similar across models, although a notable pattern emerged in the Sources item. All 45 responses scored 0 on the Sources item, as none of the three models spontaneously cited references or directed the reader to authoritative external resources. Mean per-item scores excluding sources are shown in Table 3. Accuracy was equivalent across ChatGPT and Gemini (4.40) and marginally lower in Claude (4.33). Clarity was higher in Claude and Gemini (both 4.67) than in ChatGPT (4.07). Completeness was highest in ChatGPT (4.60), and usefulness was highest in Claude (4.73).
Table 3.
Mean QAMAI per-item scores excluding sources.
No statistically significant difference was detected between models for accuracy, relevance, completeness and usefulness. The only QAMAI item showing a statistically significant difference between models was clarity (Friedman χ2 = 10.84, raw p = 0.004, adjusted p = 0.022). Post hoc Wilcoxon signed-rank testing with Bonferroni correction showed a significant difference between ChatGPT and Gemini for clarity (adjusted p = 0.038), but not between ChatGPT and Claude (adjusted p = 0.13) or Claude and Gemini (adjusted p = 1.00).
3.3. Thematic Sub-Analysis
Mean total QAMAI scores per thematic domain are summarised in Table 4. Performance was broadly comparable across the basics and risks, symptoms and diagnosis, and treatment domains, with the mean total scores ranging from 21.33 to 23.00 across models within these domains. The prognosis and survivorship domain showed marginally higher scores for Gemini (22.5) and Claude (23) than ChatGPT (21.00). The most pronounced inter-model difference occurred in the fertility and sexual health domain. Within this domain, ChatGPT achieved a mean total score of 19.67, which fell within moderate quality, while Gemini scored 23.00 (moderate quality), and Claude scored 24.00 (good quality). The consensus panel attributed ChatGPT’s lower performance to its less explicit handling of the timing and urgency of fertility preservation and to the late framing of testicular prosthesis pre-operative considerations.
Table 4.
Mean total QAMAI score by thematic domain.
3.4. Fertility and Sexual Health Questions
Q13 was pre-specified as a focused outcome of clinical interest given its time-critical nature and is presented in detail in Table 5. All three models recommended sperm banking, conveyed urgency, recommended banking before surgery or further treatment, and acknowledged risk of permanent infertility from subsequent chemotherapy or radiotherapy. QAMAI scores were 21 for ChatGPT, 24 for Gemini, and 25 for Claude. The consensus panel attributed ChatGPT’s lower score to its opening qualification (“if you have time before surgery”), which framed banking conditionally rather than as a default recommendation. Gemini’s response was more explicit regarding urgency but did not clearly state that a single sample may still be worthwhile when time is limited. Claude received the highest score but its categorical language (“Don’t let anyone tell you there’s no time”) does not sufficiently recognise that clinical urgency can in some cases override fertility preservation.
Table 5.
Focused sub-analysis of sperm cryopreservation responses.
For Q14 and Q15, the consensus panel scores favoured Claude (24 and 23 respectively) and Gemini (23 and 22) over ChatGPT (20 and 18). The panel attributed this gap to two specific features. Claude was the only model to mention retrograde ejaculation following retroperitoneal lymph node dissection in response to Q15 and explicitly frame both prosthesis and fertility preservation as decisions to be discussed with urologists before surgery, on the grounds that timing matters.
4. Discussion
This study evaluated three publicly available large language models, ChatGPT, Gemini and Claude, in their capacity to respond to patient-facing frequently asked questions on testicular cancer, with an additional focus on fertility and sexual health domains central to counselling this cohort. All three models produced coherent and broadly accurate responses, each achieving a moderate quality grade on the QAMAI tool, but with no model reaching good quality on average. While no statistically significant difference was identified in overall total QAMAI score between the three models, meaningful inter-model differences emerged with regard to clarity and within the fertility and sexual health domain.
Generative AI continues to attract interest in medical and scientific communities, with rapidly expanding adoption across medical research, diagnostic support, clinical decision-making and patient-facing applications [23]. Within urology, ChatGPT has been previously evaluated against board examinations [24] and in clinical decision-making scenarios in uro-oncology [25] and urolithiasis [26]. Application of LLMs to patient education is gaining increasing interest, with Guo et al. [27] evaluating ChatGPT in a simulated bladder cancer consultation, reporting that responses were largely accurate but superficial and lacking quantitative evidence-based content [27]. Moreover, Thia et al. [28] found that ChatGPT-produced information was a useful adjunct for increasing health literacy, but clinician input was still vital to help patients understand the nuances of their disease. A comparable pattern has been reported beyond bladder cancer, with Beyatli et al. [19] evaluating ChatGPT-4 across guideline-derived and patient-focused questions in upper tract urothelial carcinoma, finding accuracy to be high overall and higher for patient-facing question sets, but with the same limitations in depth observed across the literature. The present study mirrored these patterns, with the LLMs producing fluent and broadly accurate output that was of moderate to good quality, but that lacked the depth and personalisation necessary to be a standalone patient resource.
Inter-model evaluation within urology has been previously performed, with Alasker et al. [18] comparing ChatGPT-3.5, ChatGPT-4 and Google Bard across 52 prostate cancer questions, reporting differences in comprehensiveness, accuracy and readability between models. Pompili et al. [17] compared three LLMs in generating patient information leaflets across four urology topics, finding that meaningful inter-model variation existed despite superficially similar outputs. Within testicular cancer specifically, Uysal et al. [20] compared ChatGPT-3.5 and ChatGPT-4 using guideline-derived questions, finding that the new model outperformed its predecessor. Most recently, Lucas et al. [21] evaluated four AI chatbots for testicular cancer patient information using DISCERN, PEMAT and Flesch–Kincaid scores, reporting moderate understandability but uniformly poor quality and actionability, with responses pitched above recommended health literacy reading levels. While informative, these instruments were developed for written consumer health information (DISCERN) and printable or audiovisual patient education (PEMAT) and were not designed to capture dimensions pertinent to AI-generated medical content, such as factual accuracy, clinical relevance and overall usefulness for decision-making [22]. The present study advances this body of work, first by applying the QAMAI tool, which to date is the only instrument specifically developed and validated for the assessment of medical AI-generated information [22]; secondly, by comparing three contemporary LLMs (ChatGPT, Gemini and Claude) in testicular cancer; and third, by including a focused evaluation of fertility and sexual health questions central to counselling this patient population. The finding that overall quality did not differ significantly between models suggests that regardless of the version of these tools accessed by patients, they deliver broadly equivalent information at the level of aggregate quality; however, this may reflect a ceiling effect within the QAMAI instrument rather than true equivalence. Inter-model differences were observed with regard to clarity and fertility and sexual health domains.
The fertility and sexual health domain revealed the largest inter-model gap in the thematic sub-analysis, with ChatGPT scoring four points below Claude and three points below Gemini. Whilst all three models recommended sperm banking, conveyed urgency and acknowledged permanent infertility risk in Q13, the consensus panel distinguished between responses on the basis of framing. ChatGPT’s condition opening (“if you have time before surgery”) risks undermining the urgency the decision requires. Claude was the only model to address retrograde ejaculation following retroperitoneal lymph node dissection and to explicitly frame prosthesis and fertility preservation as decisions to be discussed before surgery; however, its categorical language (“Don’t let anyone tell you there’s no time”) does not factor in individual patient circumstances or their need for nuanced advice rather than absolute reassurance. For a young man researching his disease before specialist consultation, these differences are non-trivial. Unmet supportive care needs have been consistently reported in testicular cancer survivors, particularly in relation to counselling on fertility preservation, sexual function and prosthesis decision-making [4], and the inter-model variability identified in our study suggests that LLMs in their current form are unlikely to bridge this counselling gap.
A consistent observation across the LLM evaluation literature is that these tools ideally play a complementary role rather than substitute for patient counselling, with Carl et al. [29] reporting in a prospective study of urology patients that urologists were overwhelmingly preferred over LLMs for treatment recommendations and anxiety reduction. A further limitation identified in our study was the universal absence of source citation across all 45 responses, meaning that patients consulting these tools would receive information without verifiable provenance and are not directed towards resources for further reading. The provision of sources within LLM-generated content has been shown in varying degrees across studies applying the QAMAI tool. Furukawa et al. [30] reported low rates of source provision across the LLMs they evaluated, with ChatGPT producing no references at all, consistent with the present findings. In contrast, Bozhöyük et al. [31] observed inter-model variation in source provision, with sources generated in most LLM responses overall. This variability likely reflects both the probabilistic nature of LLM output and the influence of prompt structure on response content. The standardised prompt used in our study, which request a patient-perspective response of under 200 words, is likely to have constrained the inclusion of references, a pattern also observed in outputs reported by Guo et al. [27], whose similar patient-perspective prompt likewise yielded no references across any of the LLM outputs generated.
Of note, while not formally analysed, recalculation of total QAMAI scores excluding the Sources item would have yielded scores of 87%, 89%, and 90% of the available total for ChatGPT, Gemini and Claude respectively. Regardless, the absence of sources remains a clinically important finding. Clinicians counselling newly diagnosed men should therefore anticipate that their patients have already consulted these tools, and that any clinical information encountered may carry no traceable references or direction to further resources. For this reason, routine direction towards patient resources, including national society guidelines and survivorship organisations, remains an important component of the consultation. Whilst the versions evaluated will be superseded by newer generations, the absence of spontaneous citation and the concentration of inter-model divergence within fertility and sexual health are structural observations that may not resolve with incremental model updates.
This study has several limitations. First, LLMs employ probabilistic generation, and the responses captured represent a single sample datum on the date of data collection, and as the models are continuously updated these findings represent a snapshot rather than a stable assessment. Each question was only posted once, without regeneration, rephrasing or follow-up prompting, whereas patients typically interact conversationally, and future research should focus on LLM performance across real-world patient interactions. Only the free-tier version was evaluated, limiting extrapolation to paid or reasoning-enabled tiers. Questions were extract from established patient education resources rather than generated directly by patients and may therefore reflect what clinicians anticipate patients will ask rather than the full range encountered in practice. Moreover, individual scores prior to consensus were not recorded, meaning that consensus scoring may mask variability between raters, and responses were rated by clinicians rather than patients, which cannot capture how reassuring or actionable a newly diagnosed man finds them, with evaluation of LLM-generated information by patients themselves an important direction for future work. Additionally, mean per-item scores clustered between 4.07 and 4.73 on a five-point scale, and the closely similar total score may therefore reflect a ceiling effect within the instrument rather than established equivalence. Furthermore, models were not prompted to provide sources, and the absence of citations reflected unprompted behaviour rather than capability. Finally, only English-language responses were evaluated.
5. Conclusions
ChatGPT, Gemini and Claude deliver broadly equivalent overall quality on patient-facing testicular cancer questions; however, inter-model differences exist in clarity and in questions relating to fertility and sexual health. There was a universal absence of spontaneous source citation across all three platforms, which is an important consideration for clinicians. These tools serve as a useful supplement to, rather than a replacement for, specialist counselling. This distinction is particularly important in testicular cancer, where fertility preservation and treatment timing are individualised and time-critical, and where the inter-model differences observed in the framing of urgency could meaningfully influence a newly diagnosed man’s decisions before he is seen.
Supplementary Materials
The following supporting information can be downloaded at https://www.mdpi.com/article/10.3390/siuj7050073/s1. Table S1: Verbatim model responses, QAMAI scoring and consensus comments.
Author Contributions
Conceptualization, Y.B. and J.Y.; methodology, Y.B., J.Y. and J.S.; software, Y.B.; validation, Y.B., J.Y., S.L. and J.S.; formal analysis, Y.B. and J.Y.; investigation, J.Y., S.L., J.S. and P.R.; data curation, Y.B.; writing—original draft, Y.B. and J.Y.; writing—review and editing, S.L., J.S. and P.R.; visualisation, S.L.; supervision, P.R.; project administration, Y.B., J.Y. and P.R. All authors have read and agreed to the published version of the manuscript.
Funding
This research received no external funding.
Institutional Review Board Statement
Not applicable.
Informed Consent Statement
Not applicable.
Data Availability Statement
The original contributions presented in this study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.
Conflicts of Interest
The authors declare no conflicts of interest.
References
- Chavarriaga, J.; Nappi, L.; Papachristofilou, A.; Conduit, C.; Hamilton, R.J. Testicular cancer. Lancet 2025, 406, 76–90. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Mao, C.; Tao, C.; Yang, C.; Shen, J.; Li, G. Trends and projections of global testicular cancer burden from 1990 to 2035. Front. Oncol. 2026, 16, 1634710. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Huang, J.; Chan, S.C.; Tin, M.S.; Liu, X.; Lok, V.T.-T.; Ngai, C.H.; Zhang, L.; Lucero-Prisno, D.E., III; Xu, W.; Zheng, Z.-J.; et al. Worldwide Distribution, Risk Factors, and Temporal Trends of Testicular Cancer Incidence and Mortality: A Global Analysis. Eur. Urol. Oncol. 2022, 5, 566–576. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Doyle, R.; Craft, P.; Turner, M.; Paterson, C. Identifying the unmet supportive care needs of individuals affected by testicular cancer: A systematic review. J. Cancer Surviv. 2024, 18, 263–287. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Bultz, B.D.; Kelly, B.; Rosberger, Z.; Forbes, C.; Railton, C.; Tavener, M.; Lloyd, C.; Stukalin, I.; White, B.; Schulte, F.S.M. “Uncovering the unspoken”: A narrative analysis of patients’ experiences with testicular cancer. Support. Care Cancer 2024, 32, 584. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Lim, M.S.C.; Molenaar, A.; Brennan, L.; Reid, M.; McCaffrey, T. Young Adults’ Use of Different Social Media Platforms for Health Information: Insights from Web-Based Conversations. J. Med. Internet Res. 2022, 24, e23656. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Maon, S.N.; Hassan, N.M.; Seman, S.A.A. Online Health Information Seeking Behavior Pattern. Adv. Sci. Lett. 2017, 23, 10582–10585. [Google Scholar] [CrossRef] [Scilit]
- Demirci, Ş.; Uğurluoğlu, Ö.; Konca, M.; Çakmak, C. Socio-demographic characteristics affect health information seeking on the Internet in Turkey. Health Inf. Libr. J. 2021, 38, 304–312. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Cieślikowski, W.A.; Kasperczak, M.; Milecki, T.; Antczak, A. Reasons behind the Delayed Diagnosis of Testicular Cancer: A Retrospective Analysis. Int. J. Environ. Res. Public Health 2023, 20, 4752. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Van Booven, D.; Chen, C.-B. Chapter 6—ChatGPT and healthcare—Current and future prospects. In Artificial Intelligence in Urologic Malignancies; Arora, H., Ed.; Academic Press: Cambridge, MA, USA, 2025; pp. 173–193. [Google Scholar]
- Chow, J.C.L.; Wong, V.; Li, K. Generative Pre-Trained Transformer-Empowered Healthcare Conversations: Current Trends, Challenges, and Future Directions in Large Language Model-Enabled Medical Chatbots. BioMedInformatics 2024, 4, 837–852. [Google Scholar] [CrossRef] [Scilit]
- Bélisle-Pipon, J.-C. Why we need to be careful with LLMs in medicine. Front. Med. 2024, 11, 1495582. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Sauerbrei, A.; Kerasidou, A.; Lucivero, F.; Hallowell, N. The impact of artificial intelligence on the person-centred, doctor-patient relationship: Some problems and solutions. BMC Med. Inform. Decis. Mak. 2023, 23, 73. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Menzel, V.; Richter, E.; Helke, C.; Bürk, B.T.; Erb, H.H.H.; Leike, S.; Borkowetz, A.; Thomas, C.; Baunacke, M. Utilization of sperm cryopreservation in patients with testicular cancer. J. Cancer Res. Clin. Oncol. 2024, 150, 201. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Dax, V.; Ftanou, M.; Tran, B.; Lewin, J.; Wallace, R.; Seidler, Z.; Wiley, J.F. The impact of testicular cancer and its treatment on masculinity: A systematic review. Psycho-Oncology 2022, 31, 1459–1473. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Yumura, Y.; Takeshima, T.; Komeya, M.; Kuroda, S.; Saito, T.; Karibe, J. Fertility and sexual dysfunction in young male cancer survivors. Reprod. Med. Biol. 2022, 21, e12481. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Pompili, D.; Richa, Y.; Collins, P.; Richards, H.; Hennessey, D.B. Using artificial intelligence to generate medical literature for urology patients: A comparison of three different large language models. World J. Urol. 2024, 42, 455. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Alasker, A.; Alsalamah, S.; Alshathri, N.; Almansour, N.; Alsalamah, F.; Alghafees, M.; AlKhamees, M.; Alsaikhan, B. Performance of large language models (LLMs) in providing prostate cancer information. BMC Urol. 2024, 24, 177. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Beyatlı, M.; Güngör, H.S.; İnkaya, A.; Sobay, R.; Tahra, A.; Küçük, E.V. Expert Evaluation of ChatGPT-4 Responses to Upper Tract Urothelial Carcinoma Questions: A Prospective Comparative Study with Guideline-Based and Patient-Focused Queries. J. Clin. Med. 2025, 14, 6353. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Uysal, Ü.; Uçar, M.; Sağır, S. ChatGPT-3.5 and ChatGPT-4 Performance in Testicular Cancer: A Comparative Study. Bull. Urooncol. 2025, 24, 40–46. [Google Scholar] [CrossRef] [Scilit]
- Lucas, H.; Dittmer, B.; Stapleton, P.; Tran, B.; Corcoran, N.M.; Sathianathen, N. Artificial Intelligence Chatbots as Information Sources on Testicular Cancer: Quality, Readability and Actionability. Soc. Int. Urol. J. 2026, 7, 27. [Google Scholar] [CrossRef] [Scilit]
- Vaira, L.A.; Lechien, J.R.; Abbate, V.; Allevi, F.; Audino, G.; Beltramini, G.A.; Bergonzani, M.; Boscolo-Rizzo, P.; Califano, G.; Cammaroto, G.; et al. Validation of the Quality Analysis of Medical Artificial Intelligence (QAMAI) tool: A new tool to assess the quality of health information provided by AI platforms. Eur. Arch. Otorhinolaryngol. 2024, 281, 6123–6131. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Teo, Z.L.; Thirunavukarasu, A.J.; Elangovan, K.; Cheng, H.; Moova, P.; Soetikno, B.; Nielsen, C.; Pollreisz, A.; Ting, D.S.J.; Morris, R.J.T.; et al. Generative artificial intelligence in medicine. Nat. Med. 2025, 31, 3270–3282. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Tsai, C.-Y.; Hsieh, S.-J.; Huang, H.-H.; Deng, J.-H.; Huang, Y.-Y.; Cheng, P.-Y. Performance of ChatGPT on the Taiwan urology board examination: Insights into current strengths and shortcomings. World J. Urol. 2024, 42, 250. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Toal, M.; Hill, C.; Quinn, M.; O’Neill, C.; Maxwell, A.P. Large language Models’ clinical decision-making on when to perform a kidney biopsy: Comparative study. J. Med. Internet Res. 2025, 27, e73603. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Talyshinskii, A.; Juliebø-Jones, P.; Zeeshan Hameed, B.M.; Naik, N.; Adhikari, K.; Zhanbyrbekuly, U.; Tzelves, L.; Somani, B.K. ChatGPT as a clinical decision maker for urolithiasis: Compliance with the current European Association of Urology Guidelines. Eur. Urol. Open Sci. 2024, 69, 51–62. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Guo, A.A.; Razi, B.; Kim, P.; Canagasingham, A.; Vass, J.; Chalasani, V.; Rasiah, K.; Chung, A. The role of artificial intelligence in patient education: A bladder cancer consultation with ChatGPT. Soc. Int. Urol. J. 2024, 5, 214–224. [Google Scholar] [CrossRef] [Scilit]
- Thia, I.; Saluja, M. ChatGPT: Is This Patient Education Tool for Urological Malignancies Readable for the General Population? Res. Rep. Urol. 2024, 16, 31–37. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Carl, N.; Nguyen, L.; Haggenmüller, S.; Joachim Hetz, M.; Theres Winterstein, J.; Otto Hartung, F.; Gruene, B.; Nikolas Kather, J.; Holland-Letz, T.; Stephan Michel, M.; et al. Comparing Patient’s Confidence in Clinical Capabilities in Urology: Large Language Models Versus Urologists. Eur. Urol. Open Sci. 2024, 70, 91–98. [Google Scholar] [CrossRef] [Scilit] [PubMed]
- Furukawa, E.; Okada, H.; Nishiie, Y.; Okuhara, T. WCN26-1655 Reliability of AI-Generated Responses on Frequently-Posed Questions by Patients with Chronic Kidney Disease. Kidney Int. Rep. 2026, 11, 104046. [Google Scholar] [CrossRef] [Scilit]
- Bozhöyük, M.S.; Yücel, L. A comprehensive evaluation of artificial intelligence–provided information on common ENT surgical procedures using the QAMAI tool. J. Laryngol. Otol. 2025, 139, 1109–1114. [Google Scholar] [CrossRef] [Scilit] [PubMed]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the authors. Published by MDPI on behalf of the Société Internationale d’Urologie. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license.