ChatGPT
ChatGPT is a generative artificial intelligence chatbot and large language model studied in biomedical and health-care contexts as a source of information, an educational simulator, a clinical-reasoning aid, and a potential component of clinical decision support.
ChatGPT is a generative artificial intelligence chatbot and large language model studied in biomedical and health-care contexts as a source of information, an educational simulator, a clinical-reasoning aid, and a potential component of clinical decision support. The supplied literature includes evaluations of text-based ChatGPT, multimodal models for image interpretation, and ChatGPT Voice Mode. These applications place ChatGPT at the intersection of artificial intelligence, medical education, professional consultation, and patient-facing health information.
In the reported studies, ChatGPT was assessed using examination questions, dental images and diagnostic comparisons, clinical prediction tasks, readability and trust measures, and user surveys. Its medical relevance is therefore evaluative rather than pharmacological or biological: the research examines accuracy, sensitivity, specificity, concordance, clarity, clinical validity, and user perceptions when ChatGPT is applied to health-related tasks. The literature represented here primarily extends the themes of medical decision support, dental diagnostic accuracy, and urology clinical reasoning, while also addressing mental-health support and educational simulation.
Rebuilt from PubMed 18 Sept 2026 · no new papers today
Where the papers sit
13 papers study chatgpt directly. The themes below are drawn from those 13. 1 paradigm shift follows.
-
AI Trust in Healthcare : Trust and reliance on AI are being measured among patients, clinicians, and trainees. Findings emphasize educational value alongside risks of problematic dependence and variable examination performance. 5 papers · 38.5%
-
Clinical LLM Comparison : Clinical language models often provide safe answers but differ in accuracy, completeness, and treatment recommendations across specialties. Comparative benchmarking and human oversight remain necessary for counseling, examinations, and patient-facing use. 5 papers · 38.5%
-
AI Diagnostic Accuracy : AI tools are being benchmarked against dentists, clinicians, and standard tests for caries, dental images, and postoperative complications. Sensitivity, specificity, predictive values, and external validation remain central. 3 papers · 23.1%
In anxious young adults and people with bipolar disorder, ChatGPT can become a psychologically consequential relational aid rather than merely an informational or cognitive-support tool
The anxious young adult with generalized anxiety disorder and major depressive disorder was assumed to be using ChatGPT for routine cognitive assistance, but developed functional dependence for composing messages, interpreting social interactions, predicting the future, and making decisions, with increasing discomfort when acting independently; people with bipolar disorder likewise moved from information seeking to personal, emotional, and relational use, including emotionally intimate interactions and perceived therapeutic benefits. Together, these reports show that ChatGPT use can reinforce reassurance seeking and form attachment-like patterns, so its effects cannot be assessed only through accuracy, accessibility, or task utility; psychological dependence and relational involvement become clinically relevant outcomes 42748377Sep 42657654Aug.
Recent Findings on ChatGPT
Medical AI Trust and Use: ChatGPT now supports health information, emotional support, professional writing, examination preparation, and routine cognitive decisions, but trust and acceptance remain uneven 42664144Aug42657654Aug42716991Sep42573794Aug42748377Sep. Experts rated its myocardial infarction responses highly for accuracy and safety, whereas an external reviewer identified poor completeness 42664144Aug. Users nevertheless reported following medical recommendations without consultation, including during urgent situations, and some psychiatric users described ChatGPT as more accessible than professionals 42664144Aug42657654Aug. Younger orthopedic professionals reported more use, while older respondents more often supported disclosure in manuscripts; adjusted age effects were not significant 42716991Sep. An anxiety case linked reassurance seeking to cognitive offloading and reduced autonomy, while examination results supported educational use but not clinical competence 42748377Sep42573794Aug. These findings support adjunctive use with ethical frameworks, clearer instructions, safety mechanisms, and clinician awareness of overreliance 42664144Aug42657654Aug42716991Sep42748377Sep.
Clinical Language Model Evaluation: ChatGPT and other large language models performed well on specialized questions, but accuracy, completeness, reasoning, and recommendations varied by model and task 42747656Sep42748159Sep42720754Sep42686188Sep42583889Aug. On unruptured intracranial aneurysms, ChatGPT and Gemini showed near-identical, internally reproducible recommendations, while Claude diverged conservatively; 19.4% of cases lacked unanimity 42747656Sep. ChatGPT scored better than Gemini for breast cancer answers and scored highly for reproductive counselling, although completeness and relevance varied and inter-rater agreement was poor to slight 42748159Sep42720754Sep. ChatGPT also led Claude and AMBOSS in urology accuracy, concordance, and EAU guideline alignment, whereas pharmacy students found voice-mode realism and communication accuracy inconsistent 42686188Sep42583889Aug. Future evaluations are testing prompting, parameterization, expert supervision, educator-rated performance, and objective competency rather than relying on answer scores alone 42747656Sep42720754Sep42748159Sep42583889Aug.
AI Diagnostic Accuracy: Multimodal ChatGPT models can interpret dental chart images, but performance falls on synthesis and remains below residents for several retrieval and interpretation tasks 42696739Sep. GPT-5 Thinking matched dental students on most question-type contrasts, while residents outperformed it on several Type A-C comparisons 42696739Sep. Against clinical examination for caries, teledentistry exceeded ChatGPT in sensitivity (89.5% vs 78.9%), specificity (85.7% vs 66.7%), and AUC (0.876 vs 0.728), although neither differed significantly from the reference standard 42632883Aug. After proximal femoral nailing, ChatGPT showed limited predictive accuracy for cut-out, a high false-positive burden, and inverted calibration 42541590Aug. Further work is targeting image quality, larger datasets, task-specific training, external validation, and defined clinical boundaries before diagnostic deployment 42632883Aug42541590Aug42696739Sep.
Written from 13 PubMed abstracts, each one cited by PMID above. Published: 2026-09-06. Last written: 2026-09-18 by GPT. Drafted by language models from published abstracts; not medical advice.