other · Journal of medical systems · la publicación, 5 oct 2026 · gratis
Diez chatbots respondieron 6,291 preguntas de pacientes reumáticos; nueve de cada diez respuestas calificadas recibieron un visto bueno
El estudio alemán no midió si los pacientes mejoraron, y el aparato no tenía ninguna vigilancia de seguridad en marcha. Los elogios vienen de quienes decidieron calificar.
Versión breve · la versión detallada sigue, unos 7 min
Pregunte al chat de weeklyAI
Pregúnteme sobre la IA en su vida: le señalo las notas que weeklyAI ha publicado sobre el tema, esta semana y antes.
Las conversaciones se guardan mientras exista weeklyAI, para mejorar la publicación. Se responde en el idioma en que usted escribe.
- El estudio, de un vistazo
- Quiénes
- Pacientes reumáticos adultos que usaron diez chatbots en alemán
- Cuántos
- 6,291 preguntas registradas
- Dónde
- Alemania
- Cuándo
- Entre septiembre de 2025 y enero de 2026
- Tipo de estudio
- analysis of what people did
- Quién lo hizo
- Universidades y centros de reumatología alemanes, con la Liga Reumática alemana
- El límite que importa
- No se midió si los pacientes aprendieron o mejoraron, y no había vigilancia de seguridad en marcha.
Son porcentajes de las 602 personas que eligieron contestar el cuestionario, no de todos los usuarios; la confianza quedó por debajo del gusto.
Una respuesta segura no es prueba de que sea correcta.
Diez chatbots con nombre de enfermedad, cada uno armado con una sola guía médica alemana. Los pusieron en línea trece centros de reumatología y seis organizaciones de pacientes, y estuvieron abiertos del 2 de septiembre de 2025 al 31 de enero de 2026.
Los pacientes escribieron 6,291 preguntas en 1,534 conversaciones. Preguntaron por su enfermedad, por sus medicamentos y por cómo se vigilan, y por cómo se hace el diagnóstico.
De las 2,671 respuestas que la gente calificó, 2,481 recibieron un visto bueno. Son 92.9 de cada cien. La queja más repetida fue que la respuesta se quedaba corta: 125 de las 190 respuestas rechazadas, casi dos tercios.
Sólo el 42.5% de las respuestas recibió alguna calificación, y el cuestionario era opcional. Los números que siguen vienen de las personas que eligieron hablar.
Un cuestionario lo contestaron 602 usuarios. El 84.6% dijo que el chatbot era fácil de usar, el 84.1% que las respuestas se entendían, y el 80.2% que era un buen complemento a los materiales de educación para pacientes. Un 67.3% lo encontró confiable. La confianza, ahí, no llegó a siete de cada diez.
Cada chatbot sabía una guía y nada más. En 4.2% de las preguntas contestó que no podía responder. Y de las 6,291 respuestas, sólo el 45.0% se juzgó del todo apegada a la guía. Ese juicio lo hizo un programa de inteligencia artificial, y coincidió poco con el de los médicos.
El programa marcó once respuestas como posiblemente peligrosas. Dos reumatólogos las revisaron y los dos dijeron que una de esas once podía hacer daño. El sistema no tenía ninguna vigilancia de seguridad funcionando en tiempo real. Tampoco hay evidencia de que los pacientes aprendieran más ni de que les fuera mejor: eso no se midió.
El estudio es alemán y sus guías son alemanas. Un tema que la guía no cubre, o un medicamento aprobado después, puede quedar fuera aunque la respuesta suene segura. En otro país, con otras guías, no se probó.
Cuando se tope con un chatbot de salud, busque uno que diga de qué documento saca cada respuesta y que muestre la fuente al lado. Si contesta sin fuente, o contesta de más, ya sabe qué preguntarle.
Qué significa para usted
Ese visto bueno lo dieron pacientes adultos en Alemania, sobre diez chatbots armados cada uno con una guía alemana de reumatología. No prueba que usted aprendiera más ni que su tratamiento mejorara, porque eso no se midió. Cuando use uno, pida ver de qué documento sale la respuesta y qué fecha tiene; si el tema no está en esa guía, la respuesta puede sonar segura y aun así quedar corta.
Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6
Quién pagó: El estudio contó con una subvención de la Deutsche Rheumastiftung otorgada a J.K., y dos autores recibieron apoyo parcial del proyecto Horizon Europe SPIDERR (Acuerdo de subvención n.º 101080711); el artículo no indica si los financiadores tuvieron algún papel, y uno de los autores trabaja para el Zentrum für Telemedizin Bad Kissingen, cuya plataforma Buddy alojó los chatbots.
No tome esto como consejo médico profesional.
Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1490 palabras · unos 7 minLeerla →Cerrar
Diez chatbots en alemán respondieron 6,291 preguntas de pacientes reumáticos. Casi todas las que fueron calificadas gustaron.
Lo que el estudio midió fue el gusto de quien respondió, no si alguien aprendió algo.
Hay un momento en la consulta en que el reumatólogo ya se levantó y usted todavía tiene la duda. ¿Cuánto tarda en hacer efecto el medicamento? ¿Puedo embarazarme con este tratamiento? ¿Qué señales indican que la enfermedad se está reactivando? El artículo lo dice sin rodeos: las enfermedades reumáticas duran toda la vida y obligan a manejar síntomas, tratamientos, efectos secundarios, cambios de rutina y vigilancia de la enfermedad1. Muchas de esas preguntas quedan mal atendidas en la consulta, que suele estar apretada de tiempo2. Y no se trata solo del paciente recién diagnosticado: con los años surgen dudas nuevas, desde planear un embarazo hasta el riesgo de infecciones3.
Para responderlas, un grupo alemán construyó diez chatbots, uno por enfermedad, apoyados en las guías clínicas alemanas, y los puso a funcionar en 13 centros de reumatología y seis organizaciones de pacientes4. Entre septiembre de 2025 y enero de 2026 registraron 6,291 preguntas5. De las 2,671 respuestas que los usuarios calificaron, 2,481, el 92.9 por ciento, recibieron una calificación positiva6. Entre los 602 usuarios que contestaron el cuestionario, el 84.6 por ciento dijo que el chatbot era fácil de usar, el 84.1 por ciento que las respuestas se entendían fácil y el 80.2 por ciento que era un buen complemento a los materiales de educación para pacientes7.
Conviene ver cómo funciona el aparato. La técnica se llama recuperación aumentada: el sistema no contesta solo con lo que el modelo "sabe", sino que primero busca pasajes en documentos definidos de antemano y arma la respuesta a partir de ellos8. A eso se suma una instrucción fija: responder en lenguaje llano, sin términos médicos complicados ni abreviaturas sin explicar, y ceñirse estrictamente a la base de conocimiento de la guía9. Es la diferencia entre un programa que improvisa y uno que contesta con el libro abierto, aunque el libro sea uno solo.
El lado menos vistoso está en los números que midió un modelo de lenguaje, no los pacientes. Ese evaluador automático calificó el 95.3 por ciento de las respuestas como completamente seguras y el 79.1 por ciento como completamente correctas10. La adherencia a la guía salió mucho peor: solo el 45.0 por ciento se consideró plenamente apegada, y la concordancia con la evaluación médica fue débil11. La causa de algunos errores tiene nombre: el sistema se apegaba tanto a la guía cargada que repetía información desactualizada, por ejemplo cuando la guía todavía no incluía medicamentos recién aprobados12.
Lo que a la gente le molestó no fue la inexactitud. De las 190 respuestas que recibieron un "no me gusta", la razón más frecuente fue que la respuesta se quedaba corta, con el 65.8 por ciento13. Sobre la confianza, el cuestionario muestra un matiz: el 77.4 por ciento dijo que el chatbot ahorraría tiempo buscando respuestas, el 75.1 por ciento calificó la calidad como al menos aceptable, el 67.3 por ciento encontró las respuestas confiables y el 56.0 por ciento prefirió usarlo antes que una búsqueda general en internet como Google14. La confianza, entonces, va por debajo del gusto.
Aquí están los límites, y son serios. No hubo monitoreo de seguridad clínica en tiempo real ni una ruta para escalar un caso urgente15. El estudio no midió resultados en los pacientes, y los propios autores piden más estudios para evaluar la efectividad educativa y validar de forma independiente la calidad y la seguridad de las respuestas16. Las cifras positivas provienen de quienes decidieron calificar y de quienes decidieron contestar el cuestionario, no de todos los usuarios.
Otros trabajos que leímos, y que conocemos solo por su resumen porque el texto completo está detrás de una suscripción, permiten comparar dos maneras de usar la misma tecnología. En un ensayo con 1,297 participantes en Estados Unidos, Canadá y el Reino Unido, una conversación de tres minutos con un chatbot sobre la vacuna del papiloma humano aumentó la intención de vacunar al momento, pero a los 45 días ninguno de los dos estilos de chatbot mantenía ese efecto, mientras que los materiales públicos impresos sí conservaban uno modesto, y ninguna intervención aumentó la vacunación reportada17181920. En Hong Kong, un chatbot que aplicaba técnicas de entrevista motivacional dentro de centros de salud de primer nivel, probado en 627 adultos con hipertensión o diabetes o en riesgo de ellas, mostró mejoras en actividad física y alimentación a las 12 semanas, sostenidas a los nueve meses y sin eventos adversos graves reportados212223. La diferencia entre ambos casos está en el diseño y en lo que se mide: intención declarada en uno, conducta y presión arterial en el otro2425.
Hay una tercera comparación útil. En un estudio con 576 participantes sin conocimientos médicos, versiones simplificadas por ChatGPT de informes de resonancia, tomografía, cirugía, patología y alta hospitalaria se leyeron mejor y se entendieron más: en resonancia, por ejemplo, la comprensión pasó del 53,7 al 80.3 por ciento, sin pérdida de información médica observada262728. El problema que atiende es el mismo que el de los chatbots alemanes: los documentos médicos se escriben para profesionales y eso limita la comprensión del paciente29.
Así lo leemos nosotros. Cuando la consulta con el especialista tarda semanas, la gente no deja de tener preguntas: las guarda y las resuelve como puede. Por eso esperamos que quienes más aprovechen una herramienta así sean justamente los que viven lejos de un reumatólogo o esperan mucho entre citas, y que lo que más le pregunten sean las cosas prácticas que no alcanzan a caber en veinte minutos: dosis, efectos, qué esperar del tratamiento. Nos equivocaríamos si los usuarios con acceso frecuente al especialista resultaran ser los que más la usan y mejor la califican, o si casi todas las preguntas fueran definiciones generales de la enfermedad y no asuntos del día a día. Si usted vive con una enfermedad reumática y tarda en ver a su especialista, lo que puede hacer es anotar sus dudas entre una cita y otra, llevar esa lista a la consulta y usar una herramienta así solo para preguntar mejor, nunca para cambiar medicamentos por cuenta propia.
También esperamos que la voz amable y la seguridad con que responde un programa de estos inclinen a la gente a tomarlo como autoridad sin preguntar de dónde sale lo que dice. La confianza se apoya más en el tono que en la fuente. Lo veríamos al revés si los usuarios citaran las fuentes que el sistema muestra, compararan las respuestas con su médico o rechazaran una respuesta por no traer respaldo, en lugar de calificarla por lo larga o lo seca. Antes de creer una respuesta, pregunte de qué guía o documento sale y contrástela con su equipo de salud. Una respuesta segura no es prueba de que sea correcta.
Y hay un detalle que vale para cualquier familia. Quien solo recibe explicaciones hechas, sin espacio para preguntar, retiene poco. Quien formula sus propias preguntas aprende a pensar por sí mismo. Esperamos que quienes más sacaron provecho llegaran con una duda concreta y siguieran conversando varios turnos, y que una respuesta larga y sin diálogo se sintiera como un folleto más, aunque fuera correcta. Nos equivocaríamos si el número de turnos no cambiara nada en la valoración. Lo que puede hacer con esto es usar estas herramientas para afinar sus propias preguntas antes de una cita, y pedir a su equipo de salud que le explique lo que no entendió, en vez de quedarse solo con la indicación.
Para el lector de esta parte del mundo, la distancia es grande y hay que decirla. Los diez chatbots se construyeron sobre guías clínicas alemanas, en alemán, y responden sobre las diez enfermedades reumáticas que esas guías cubren; nada de esto se probó en México, en Colombia ni en una clínica de habla hispana. Lo que puede pedir donde vive es distinto: que su sistema de salud publique materiales de educación para pacientes en lenguaje llano, que su especialista le explique el plan por escrito y que cualquier herramienta digital que le ofrezcan le diga con claridad de qué documento sale cada respuesta. Eso sí se puede exigir en cualquier consultorio.
Volvamos a la sala de espera. Usted tiene el teléfono en la mano, una duda concreta y una cita en tres semanas. Un programa que responde con base en una guía puede ayudarle a llegar a esa cita con mejores preguntas. Lo que este estudio muestra es que a la mayoría de quienes calificaron les gustó la respuesta, y que la seguridad y la exactitud las midió una máquina, con un acuerdo flojo frente a los médicos solo al medir si la respuesta se apoyaba en la guía. La próxima vez que una herramienta así le conteste, pregúntele de qué guía sale, y lleve esa respuesta a su reumatólogo antes de tomar cualquier decisión.
De dónde sale cada dato de contexto, y cuánto leímos de cada documento
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Rheumatic diseases are chronic conditions that require lifelong management. Patients must continuously navigate questions related to symptoms, treatment options, side effects, lifestyle adaptations, and disease monitoring."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Yet many of these questions remain insufficiently addressed in routine care, which is often constrained by limited consultation time"
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Importantly, unanswered questions are not restricted to newly diagnosed patients. Over the course of years and decades, new questions arise as life circumstances change, from pregnancy planning to infection risk."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Ten disease-specific chatbots based on German guidelines were co-developed and deployed through 13 rheumatology centres and six patient organisations."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Between September 2025 and January 2026, 6291 questions were recorded."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Of 2671 responses rated by users, 2481 (92.9%) received a positive rating."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Among 602 questionnaire respondents, 84.6% reported that the chatbot was easy to use, 84.1% that answers were easy to understand, and 80.2% that it was a useful addition to patient education."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Retrieval-augmented generation (RAG) can improve reliability by anchoring chatbot responses in predefined source documents rather than relying solely on the model’s internal knowledge"
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The system prompt instructed the model to answer patient questions about rheumatic diseases in plain language, avoid complex medical terminology and unexplained abbreviations, and adhere strictly to the guideline-based knowledge base."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "In the LLM-based evaluation, 95.3% of answers were rated as completely safe and 79.1% as completely correct."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Guideline adherence was assessed separately, with 45.0% rated as fully adherent; agreement with physician assessment was weak."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Chatbot answer correctness was sometimes affected by strict adherence to the uploaded guideline documents, as these did not always reflect the most recent drug approvals."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The most common reason for a dislike was “insufficient” 65.8% (125/190)"
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Furthermore, 77.4% indicated that the chatbot would save time when searching for answers, 75.1% rated the answer quality as at least acceptable, 67.3% found the answers trustworthy, and 56.0% preferred using the chatbot over general internet searches (e.g. Google)."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "No real-time clinical safety monitoring or escalation pathway was implemented."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Further studies are needed to evaluate educational effectiveness and independently validate response quality and clinical safety."
- Sehgal NKR, Rai S, Tonneau M, Agarwal AK, Cappella J, Kornides ML, et al. (2026). Large Language Model Chatbot Conversations vs Public Health Materials and Parental HPV Vaccination Intentions. JAMA Network Open. 10.1001/jamanetworkopen.2026.16822 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Compared with no intervention, public health materials increased immediate vaccination intent (Cohen d = 0.53; 95% CI, 0.36-0.70), as did the default chatbot (d = 0.48; 95% CI, 0.30-0.65) and conversational chatbot (d = 0.33; 95% CI, 0.17-0.49)."
- Sehgal NKR, Rai S, Tonneau M, Agarwal AK, Cappella J, Kornides ML, et al. (2026). Large Language Model Chatbot Conversations vs Public Health Materials and Parental HPV Vaccination Intentions. JAMA Network Open. 10.1001/jamanetworkopen.2026.16822 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "At 45 days, neither chatbot increased intent relative to controls, whereas public health materials maintained modest effects."
- Sehgal NKR, Rai S, Tonneau M, Agarwal AK, Cappella J, Kornides ML, et al. (2026). Large Language Model Chatbot Conversations vs Public Health Materials and Parental HPV Vaccination Intentions. JAMA Network Open. 10.1001/jamanetworkopen.2026.16822 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "No intervention increased self-reported vaccination uptake."
- Sehgal NKR, Rai S, Tonneau M, Agarwal AK, Cappella J, Kornides ML, et al. (2026). Large Language Model Chatbot Conversations vs Public Health Materials and Parental HPV Vaccination Intentions. JAMA Network Open. 10.1001/jamanetworkopen.2026.16822 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "A total of 1297 participants (mean [SD] age, 42.84 [6.93] years; 935 [72.1%] female) were randomized."
- Wong CHY, Wong MNK, Hou WK, Ting FST, Ng YM, Che Hin Chan C. (2026). Motivational interviewing chatbot improves lifestyle in primary healthcare settings in a pragmatic randomised controlled trial. npj Digital Medicine. 10.1038/s41746-026-02728-w — solo el resumen - no se pudo obtener el texto completo — el pasaje: "This pragmatic, open-label, multicentre randomised controlled trial across three Hong Kong PHC facilities randomised 627 adults aged 45-75 with or at risk of hypertension or diabetes."
- Wong CHY, Wong MNK, Hou WK, Ting FST, Ng YM, Che Hin Chan C. (2026). Motivational interviewing chatbot improves lifestyle in primary healthcare settings in a pragmatic randomised controlled trial. npj Digital Medicine. 10.1038/s41746-026-02728-w — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Modified intention-to-treat analysis (n = 460) showed significant improvements in the intervention group compared to usual care in physical activity (576 MET-min/week), fruit/vegetable intake (0.27 portions/day), and committed action (0.95 a.u.) at 12 weeks."
- Wong CHY, Wong MNK, Hou WK, Ting FST, Ng YM, Che Hin Chan C. (2026). Motivational interviewing chatbot improves lifestyle in primary healthcare settings in a pragmatic randomised controlled trial. npj Digital Medicine. 10.1038/s41746-026-02728-w — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Effects were sustained at 9-month follow-up. No serious adverse events were reported."
- Sehgal NKR, Rai S, Tonneau M, Agarwal AK, Cappella J, Kornides ML, et al. (2026). Large Language Model Chatbot Conversations vs Public Health Materials and Parental HPV Vaccination Intentions. JAMA Network Open. 10.1001/jamanetworkopen.2026.16822 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "The findings of this randomized clinical trial suggest that well-designed public health messaging may match or exceed the impact of short chatbot conversations for HPV vaccine promotion."
- Wong CHY, Wong MNK, Hou WK, Ting FST, Ng YM, Che Hin Chan C. (2026). Motivational interviewing chatbot improves lifestyle in primary healthcare settings in a pragmatic randomised controlled trial. npj Digital Medicine. 10.1038/s41746-026-02728-w — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Exploratory subgroup analyses showed strong effects among participants with low baseline physical activity (537 MET-min/week) and systolic blood pressure reductions among hypertensive participants (-5.03 mmHg)."
- Stephan D, Schumacher S, Al-Nawas B, Kämmerer PW, Thiem DGE. (2026). Improving patient communication and understanding of medical reports through text simplification with ChatGPT: a comparative study. Oral and Maxillofacial Surgery. 10.1007/s10006-026-01643-4 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "A total of 576 participants without medical knowledge evaluated either a simplified or original report version and completed a standardized questionnaire assessing clarity, structure, and applicability, followed by comprehension testing."
- Stephan D, Schumacher S, Al-Nawas B, Kämmerer PW, Thiem DGE. (2026). Improving patient communication and understanding of medical reports through text simplification with ChatGPT: a comparative study. Oral and Maxillofacial Surgery. 10.1007/s10006-026-01643-4 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Across all document types, patients rated simplified texts significantly higher in clarity, structure, and usefulness (p < 0.001), with significantly improved comprehension accuracy (e.g., MRI 80.3% vs. 53.7%; p < 0.001)."
- Stephan D, Schumacher S, Al-Nawas B, Kämmerer PW, Thiem DGE. (2026). Improving patient communication and understanding of medical reports through text simplification with ChatGPT: a comparative study. Oral and Maxillofacial Surgery. 10.1007/s10006-026-01643-4 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "No loss of medical information was observed."
- Stephan D, Schumacher S, Al-Nawas B, Kämmerer PW, Thiem DGE. (2026). Improving patient communication and understanding of medical reports through text simplification with ChatGPT: a comparative study. Oral and Maxillofacial Surgery. 10.1007/s10006-026-01643-4 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Medical documentation is essential for clinical communication but is often intended for professional audiences, limiting patient understanding. This linguistic complexity can reduce health literacy and hinder shared decision-making."
Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6
Quién pagó: El estudio contó con una subvención de la Deutsche Rheumastiftung otorgada a J.K., y dos autores recibieron apoyo parcial del proyecto Horizon Europe SPIDERR (Acuerdo de subvención n.º 101080711); el artículo no indica si los financiadores tuvieron algún papel, y uno de los autores trabaja para el Zentrum für Telemedizin Bad Kissingen, cuya plataforma Buddy alojó los chatbots.
No tome esto como consejo médico profesional.
other · Journal of medical systems · the paper, 5 Oct 2026 · free
Ten Rheumatology Chatbots Got a Thumbs-Up on 92.9% of the Answers Users Rated
Most answers were rated helpful, but fewer than half of all answers were rated at all, and the chatbots knew only German guidelines.
Short version · the longer version follows, about 6 min
Ask the weeklyAI chat
Ask me about AI in your life: I point you to the stories weeklyAI has published on it, this week and before.
Conversations are saved for as long as weeklyAI exists, to improve the publication. Answers come in the language you write in.
- The study at a glance
- Who
- people with rheumatic diseases using ten German guideline-based chatbots
- How many
- 6291 questions recorded
- Where
- Germany
- When
- September 2025 to January 2026
- Kind of study
- analysis of what people did
- Who did it
- rheumatology centres and patient organisations in Germany
- The limit that matters
- Only 42.5 percent of answers were rated, and the questionnaire was optional.
These are shares of the 602 people who chose to fill in an optional questionnaire, so they speak for those who answered, not for all users.
The next time a health chatbot offers you an answer, look for one that shows where its information comes from, beside each answer, and says plainly when it does not know.
Take 2,671 answers and line them up. On 2,481 of them, someone pressed the thumbs-up button. That is 92.9 percent, about nine in ten.
Those answers came from ten chatbots built for people with rheumatic diseases, the long-lasting conditions that affect joints and other parts of the body. Each chatbot was tied to one German medical guideline, the official document that tells doctors how to treat that disease. The chatbots were available from September 2025 to January 2026. Thirteen rheumatology centres and six patient organisations in Germany helped develop and spread them. Doctors, patients, patient groups and technical experts worked together to develop them, using a commercial platform.
People typed 6,291 questions in 1,534 conversations. Half of the questions were about their own disease, such as how to prepare for a doctor's visit. More than a third were about medicines and monitoring, like when a drug starts to work.
The complaints were few. Of 190 thumbs-down, 125 said the answer was not detailed enough. In a questionnaire, 602 users answered: 84.6 percent agreed the chatbot was easy to use, 84.1 percent that the answers were easy to understand, and 80.2 percent that it was a useful addition to patient education.
Consider who is speaking. Only 42.5 percent of answers were rated at all, and the questionnaire was optional. These are the voices of people who chose to respond, and they cannot tell you how everyone felt.
The knowledge base was one guideline per disease, so its edges showed. The chatbot said it could not answer 4.2 percent of questions, 263 of 6,291. A computer program rated 45.0 percent of all answers as fully supported by the guidelines. The researchers warn that this rating is uncertain because it agreed poorly with a doctor's review.
A computer program judged every answer for safety and correctness. Those ratings said 95.3 percent of answers were completely safe and 79.1 percent completely correct. No live safety monitoring was running. A computer program flagged 11 answers as possibly unsafe. When two rheumatologists reviewed them, both rated one of the 11 as potentially unsafe.
The next time a health chatbot offers you an answer, look for one that shows where its information comes from, beside each answer, and says plainly when it does not know.
What this means for you
For now, treat a chatbot's thumbs-up as a sign that someone found an answer useful, not as proof that its advice was safe. If a health chatbot offers you an answer, look for one that shows its sources beside each answer and says plainly when it does not know, and ask your own rheumatologist about anything that would change your medicine.
Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6
Who paid: The study was supported by a grant from the Deutsche Rheumastiftung awarded to J.K., and two authors were partly supported by the Horizon Europe SPIDERR project (Grant Agreement No. 101080711); the article does not state whether the funders had any role, and one author is employed by the Zentrum für Telemedizin Bad Kissingen, whose Buddy platform hosted the chatbots.
Do not take this as professional medical advice.
The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 1104 words · about 6 minRead it →Close
A Chatbot Built on Guidelines Was Liked. Safe Is a Bigger Claim.
Germany's rheumatology network tested ten chatbots on patients' own questions. Most answers drew praise; the safety check was done by a machine.
Consider what a clinic visit covers. A person with rheumatoid arthritis or lupus waits weeks for an appointment, then has minutes to ask about a drug's side effects, a flare, a pregnancy, a strange new symptom. The questions keep coming after the door closes. Rheumatic diseases are chronic conditions that require lifelong management, and patients must continuously navigate questions about symptoms, treatment options, side effects, lifestyle changes and disease monitoring1. The article's authors note that many of these questions remain insufficiently addressed in routine care, which is often constrained by limited consultation time2. They add that the problem is not limited to people newly diagnosed: over years and decades, new questions arise as life circumstances change, from pregnancy planning to infection risk3.
Ten chatbots were built on German clinical guidelines and deployed through 13 rheumatology centres and six patient organisations4. The technology is worth understanding, because it is not a general chatbot with a medical tone. Retrieval-augmented generation, as the authors describe it, anchors a chatbot's responses in predefined source documents rather than relying solely on the model's internal knowledge5. In this case the source was the latest valid German guideline for each disease. When a patient typed a question, the system searched the guideline, pulled up to eight relevant passages, and instructed the model to answer in plain language, avoid unexplained medical jargon, and stick strictly to the guideline-based knowledge base6. The model itself was OpenAI's GPT-4o, reached through a Microsoft Azure endpoint in Germany; it was not fine-tuned, and patient inputs were not used to train it.
Between September 2025 and January 2026, 6,291 questions were recorded7. Of the 2,671 responses that users rated, 2,481, or 92.9 percent, received a positive rating8. Among 602 questionnaire respondents, 84.6 percent reported the chatbot was easy to use, 84.1 percent that answers were easy to understand, and 80.2 percent that it was a useful addition to patient education9. Users also said it would save them time searching (77.4 percent), rated answer quality at least acceptable (75.1 percent), found the answers trustworthy (67.3 percent), and preferred it over general internet searches (56.0 percent)10. When people disliked an answer, the most common reason was that it was insufficiently detailed: 65.8 percent of dislikes, 125 of 19011.
Now the part that needs care. The safety and correctness figures in this study came from an automated judge, a large language model rating the chatbot's answers, not from rheumatologists reviewing every response. In that evaluation, 95.3 percent of answers were rated as completely safe and 79.1 percent as completely correct12. Guideline adherence was assessed separately, with 45.0 percent rated as fully adherent, and agreement with physician assessment was weak13. That last clause matters: the machine's judgment of whether an answer followed the guideline did not reliably match what doctors thought. Treat those automated estimates as a screening tool, not a certificate.
The design also carried a known tension. Chatbot answer correctness was sometimes affected by strict adherence to the uploaded guideline documents, as these did not always reflect the most recent drug approvals14. One answer told a patient that no JAK inhibitors were specifically approved for ankylosing spondylitis — true of the guideline the system had, not of the current regulatory picture. And no real-time clinical safety monitoring or escalation pathway was implemented15. The system could decline to answer, which it did in 4.2 percent of interactions.
Here is how we read it. A tool that answers from an official document borrows the document's authority, and the person on the other end cannot see which passages were retrieved, which were ranked higher, or what the guideline simply left out. They can judge whether the answer sounds right; they cannot judge the selection behind it. Our expectation, not the study's finding, is that people will trust such a tool roughly in proportion to how official it looks, and that the answers most likely to mislead will be the confident ones about topics the source handles thinly or out of date.
You can test whether we are wrong. Watch whether your own confidence rises and falls with the source the tool names and the date it was last updated, rather than with how polished the answer sounds. If a clinic near you offers something like this, the useful questions are plain: Which document does it answer from? When was that document last revised? What does it do when it does not know?
A second pattern is worth naming. When a system presents itself as neutral and automatic, that presentation does much of the credibility work. The German team was unusually candid about this: they described their own tool's limits in the paper, including the missing safety net. Most products do not. A chatbot that declines to answer can look cautious when it is simply out of material.
What this study did not measure is the thing a patient would most want to know: whether using the chatbot improved anyone's knowledge, confidence, adherence to treatment or use of health services. The authors say further studies are needed to evaluate educational effectiveness and independently validate response quality and clinical safety16. The positive ratings also come from a self-selected group — only 42.5 percent of responses were rated at all, and the questionnaire was optional.
We also read, in another line of work, that a short conversation with a chatbot can shift health intentions in the moment but not necessarily later. In an online trial among parents in the US, Canada and the UK, a three-minute chat with GPT-4o prompted to encourage HPV vaccination raised vaccination intent right away; by 45 days, neither chatbot version had increased intent relative to controls, while ordinary public health materials held a modest effect, and no intervention increased actual vaccination. We could read only the summary of that study; the full paper is behind a subscription. It is a caution about enthusiasm, not a verdict on this one.
So what does the German work make possible? A person with a chronic illness, holding a phone at nine at night, can get a plain-language answer drawn from a guideline their own rheumatologist would recognize — and can bring the answer, and the question behind it, to the next appointment. That is a supplement to care, not a replacement for it. The tool that answered you was built by people who chose what it would read. Ask them what they chose, and when they last changed it.
Where each piece of context comes from, and how much of it we read
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Rheumatic diseases are chronic conditions that require lifelong management. Patients must continuously navigate questions related to symptoms, treatment options, side effects, lifestyle adaptations, and disease monitoring."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Yet many of these questions remain insufficiently addressed in routine care, which is often constrained by limited consultation time"
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Importantly, unanswered questions are not restricted to newly diagnosed patients. Over the course of years and decades, new questions arise as life circumstances change, from pregnancy planning to infection risk."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Ten disease-specific chatbots based on German guidelines were co-developed and deployed through 13 rheumatology centres and six patient organisations."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Retrieval-augmented generation (RAG) can improve reliability by anchoring chatbot responses in predefined source documents rather than relying solely on the model’s internal knowledge"
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "The system prompt instructed the model to answer patient questions about rheumatic diseases in plain language, avoid complex medical terminology and unexplained abbreviations, and adhere strictly to the guideline-based knowledge base."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Between September 2025 and January 2026, 6291 questions were recorded."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Of 2671 responses rated by users, 2481 (92.9%) received a positive rating."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Among 602 questionnaire respondents, 84.6% reported that the chatbot was easy to use, 84.1% that answers were easy to understand, and 80.2% that it was a useful addition to patient education."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Furthermore, 77.4% indicated that the chatbot would save time when searching for answers, 75.1% rated the answer quality as at least acceptable, 67.3% found the answers trustworthy, and 56.0% preferred using the chatbot over general internet searches (e.g. Google)."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "The most common reason for a dislike was “insufficient” 65.8% (125/190)"
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "In the LLM-based evaluation, 95.3% of answers were rated as completely safe and 79.1% as completely correct."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Guideline adherence was assessed separately, with 45.0% rated as fully adherent; agreement with physician assessment was weak."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Chatbot answer correctness was sometimes affected by strict adherence to the uploaded guideline documents, as these did not always reflect the most recent drug approvals."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "No real-time clinical safety monitoring or escalation pathway was implemented."
- Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6 - the article this story is about — the whole article — the passage: "Further studies are needed to evaluate educational effectiveness and independently validate response quality and clinical safety."
Wilhelmi, T., Bartsch, V., Platt, M. et al. (2026). Development and Nationwide Multicentre Evaluation of Guideline-Grounded Large Language Model Chatbots to Support Patient Self-Management and Education in Rheumatology. Journal of Medical Systems. https://doi.org/10.1007/s10916-026-02470-6
Who paid: The study was supported by a grant from the Deutsche Rheumastiftung awarded to J.K., and two authors were partly supported by the Horizon Europe SPIDERR project (Grant Agreement No. 101080711); the article does not state whether the funders had any role, and one author is employed by the Zentrum für Telemedizin Bad Kissingen, whose Buddy platform hosted the chatbots.
Do not take this as professional medical advice.

