weeklyAI · Week of 9 October 2026weeklyAI · Semana del 9 de octubre de 2026

← Your parents← Sus padres

experiment · Antimicrobial Stewardship & Healthcare Epidemiology · la publicación, 1 ago 2026 · gratis

En 22 de 24 respuestas los robots dijeron que sí era una infección de orina

Un estudio pequeño escribió ocho preguntas inventadas de familiares y las puso en tres chatbots. En 22 de 24 respuestas el robot estuvo de acuerdo con que era una infección urinaria.

Versión breve · la versión detallada sigue, unos 8 min

Pregunte al chat de weeklyAI

Pregúnteme por este estudio: a quiénes se estudió, qué encontró y qué no dice.

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
Preguntas simuladas de familiares sobre un análisis de orina positivo sin síntomas claros
Cuántos
24 respuestas
Dónde
No lo dice el artículo
Cuándo
15 de diciembre de 2025
Tipo de estudio
analysis of what people did
Quién lo hizo
University of Utah, University of Alabama at Birmingham
El límite que importa
Las preguntas eran inventadas y no hubo personas reales seguidas después
Lo que dijeron los tres robots en sus 24 respuestas
Dijeron que parecía una infección urinaria91.7%
Mencionaron explicaciones alternativas70.9%
Nombraron la bacteriuria asintomática66.7%
Mencionaron antibióticos33.3%
Recomendaron buscar atención médica inmediata79.2%
Incluyeron citas o referencias29.2%

Porcentajes de las 24 respuestas de tres robots a ocho preguntas inventadas sobre un análisis de orina positivo sin síntomas claros; las preguntas no las hizo ningún familiar real y no se siguió a ninguna persona después.

Cómo cambiaron las respuestas según el robot, el sexo y la edad

Geminifrente aChatGPT y Copilot

Gemini nombró la bacteriuria asintomática en 0 de 4 respuestas; ChatGPT en 2 de 4 y Copilot en 2 de 4

Padrefrente aMadre

Sugirieron antibióticos en 1 de 6 respuestas cuando era el padre y en 3 de 6 cuando era la madre

68 añosfrente a84 años

Plantearon diagnósticos alternativos en 4 de 6 respuestas a los 68 años y en 1 de 6 a los 84

Cuando consulte a un robot por un resultado de orina, fíjese en dos cosas antes de que aparezca la palabra antibiótico: si nombra la bacteriuria asintomática y si pregunta por los síntomas.
Lectura de weeklyAI
Ver · 0:52

Investigadores escribieron ocho preguntas inventadas de familiares sobre un resultado de orina positivo, sin los síntomas claros de una infección urinaria. Las metieron en tres robots conversacionales públicos: ChatGPT v5, Gemini v3 y Copilot v5. Eso lo hicieron el 15 de diciembre de 2025, en sesiones privadas y sin memoria.

Las preguntas cambiaban el sexo del padre o de la madre y la edad, 68 u 84 años. Una de las dos historias añadía fiebre y orina con mal olor.

De las 24 respuestas, 22 dijeron que sí, que parecía una infección urinaria. Son casi todas.

Sólo 16 de las 24, dos de cada tres, nombraron la bacteriuria asintomática. Es el nombre de tener bacterias en la orina sin síntomas, y es justamente el caso de las preguntas.

Los robots también cambiaron según la historia y según a quién se refería. Cuando la historia no traía fiebre ni mal olor, algunos mencionaron los daños de los antibióticos; cuando sí los traía, no. Ante una madre sugirieron antibióticos 3 de 6 veces; ante un padre, 1 de 6. Y ante una persona de 68 años propusieron otras explicaciones 4 de 6 veces; ante una de 84, 1 de 6.

Estas preguntas las escribieron los investigadores, no las hizo ningún familiar real. Nadie siguió después a una persona para ver si tomó antibióticos o se enfermó. Tampoco puede decirse que estos robots causen un mal; sólo se anotó lo que respondieron.

Los tres robots se probaron un solo día. Otra aplicación, u otra versión más nueva, puede contestar distinto. Y como las preguntas eran inventadas, nada de esto dice qué responderá su robot ante el caso concreto de su familia.

Los tres robots no respondieron igual entre sí. Por ejemplo, Gemini nombró la bacteriuria asintomática en 0 de 4 respuestas, ChatGPT en 2 de 4 y Copilot en 2 de 4. Se contó lo que dijeron, nada más.

Cuando consulte a un robot por un resultado de orina, fíjese en dos cosas antes de que aparezca la palabra antibiótico: si nombra la bacteriuria asintomática y si pregunta por los síntomas.

Qué significa para usted

Cuando usted o su familia consulten a un robot por un resultado de orina, conviene fijarse en si nombra la bacteriuria asintomática y si pregunta por los síntomas, como muestra el estudio. Si el robot habla de antibióticos sin mencionarlos, esa respuesta no basta para decidir nada. La persona indicada para valorar ese resultado es un profesional de la salud.

Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512

Quién pagó: El artículo no dice quién financió el estudio, si los financiadores tuvieron alguna influencia, ni quién prestó equipos o programas.

No tome esto como consejo médico profesional.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1537 palabras · unos 8 minLeerla →Cerrar

Los chatbots de IA ante un análisis de orina sin síntomas: qué respondieron y qué no podemos concluir

Un estudio puso ocho preguntas simuladas de cuidadores a tres chatbots. La mayoría respondió que era una infección urinaria.

Así podría verse · ilustración generada por weeklyAI.watch, no es una fotografía

Ocho preguntas escritas como las haría un hijo o una hija preocupados por un padre o una madre se introdujeron en tres chatbots de uso público —ChatGPT v5, Gemini v3 y Copilot v5— el 15 de diciembre de 2025, en sesiones privadas y sin memoria activada1. Las preguntas describían a un padre o una madre con un análisis de orina positivo pero sin los síntomas típicos de una infección urinaria, y preguntaban directamente si esa persona tenía una infección2. Los investigadores crearon dos escenarios: uno con solo el análisis positivo, y otro que añadía fiebre y orina con mal olor3. Para cada escenario variaron el sexo del padre o la madre (madre o padre) y la edad (68 o 84 años)4.

De las 24 respuestas recogidas, 22 (91.7%) respaldaron la idea de que se trataba de una infección urinaria5. Un 70.9% mencionó explicaciones alternativas para los síntomas y algo más de la mitad —16 de 24, un 66.7%— mencionó la bacteriuria asintomática6. Los antibióticos aparecieron en 8 respuestas (33.3%), y en 6 de esas se señalaron posibles daños7. La mayoría, 19 de 24 (79.2%), recomendó buscar atención médica inmediata8. El tono varió: 9 respuestas (37.5%) subrayaron la urgencia, 5 (20.8%) fueron tranquilizadoras y 10 (41.7%) neutras9. Solo 7 respuestas (29.2%) incluyeron citas o referencias10.

Las respuestas no fueron iguales entre escenarios. Por ejemplo, los posibles daños de los antibióticos se mencionaron en el escenario A pero no en el B11. Dentro de cada escenario también hubo diferencias: al mencionar la bacteriuria asintomática, Gemini lo hizo en 0 de 4 respuestas, ChatGPT en 2 de 4 y Copilot en 2 de 4; los antibióticos se sugirieron en 1 de 6 respuestas cuando el paciente era el padre y en 3 de 6 cuando era la madre; y los diagnósticos alternativos se plantearon en 4 de 6 respuestas cuando la persona tenía 68 años y en 1 de 6 cuando tenía 8412.

Conviene detenerse en lo que este estudio no hizo. Las preguntas eran simuladas, no de cuidadores reales, y no hubo personas participantes: nadie siguió a una familia para ver si tomó antibióticos ni si se enfermó. Se contó lo que dijeron los chatbots, no lo que ocurre después. Se probaron tres aplicaciones en un solo día, así que otra aplicación o una versión posterior puede responder distinto. Fueron 24 respuestas repartidas en 8 preguntas y 3 plataformas, sin pruebas estadísticas de las diferencias por plataforma, sexo o edad. Las respuestas las codificaron dos miembros del equipo, con dos revisores adicionales resolviendo desacuerdos.

Para entender por qué esto importa, hay que empezar por lo que es un chatbot: un programa al que usted le escribe una pregunta en lenguaje corriente y que le devuelve una respuesta conversacional, como si charlara con alguien13. Cada vez más pacientes y cuidadores los usan para eso: respuestas rápidas a dudas médicas13. Y ahí está el nudo del asunto: la preocupación es que esas respuestas pueden influir en lo que la persona cree sobre su enfermedad, en lo que espera del tratamiento y en la decisión de buscar atención14.

El caso de la orina es un buen ejemplo de por qué esa influencia pesa. La bacteriuria asintomática —bacterias en la orina sin síntomas claros— se confunde con frecuencia con una infección urinaria, y esa confusión lleva a usar antibióticos innecesarios15. Es decir: la pregunta que la familia escribe al chatbot suele nacer ya torcida, y la respuesta puede terminar de torcerla.

Así podría verse · ilustración generada por weeklyAI.watch, no es una fotografía

Este estudio no salió de la nada. Sus autores se propusieron examinar cómo responden los chatbots a preguntas simuladas de cuidadores sobre bacteriuria asintomática e infecciones urinarias16. Y otros equipos han hecho preguntas parecidas en otros terrenos. Según el resumen de un estudio sobre decisiones de fin de vida en adultos mayores —pudimos leer solo el resumen, el texto completo está detrás de una suscripción—, los grandes modelos de lenguaje se usan mucho para buscar información de salud y facilitar la comunicación con el médico, pero su valor y sus riesgos en ese terreno aún no se han evaluado de forma sistemática17.

Ese mismo trabajo, siempre según su resumen, puso a prueba cinco sistemas de IA de uso público con preguntas estandarizadas para familiares que deben decidir por un adulto mayor18. Las preguntas se armaron a partir de guías clínicas, documentos de consenso, literatura científica y estudios con cuidadores reales19. Las tasas de respuestas consideradas seguras fueron del 76.7% al 86.7%, sin que se hallara una diferencia general entre sistemas20. La exactitud sí varió entre ellos, aunque el efecto fue pequeño21. Las diferencias en empatía, según la evaluación de expertos, fueron más marcadas22. También hubo diferencias en varias medidas de calidad y transparencia de la información23. Cada pregunta se envió una sola vez, en una sesión nueva, a cinco sistemas, en un solo día de junio de 202624. Dos médicos evaluaron las respuestas sin saber cuál sistema las había producido25.

Otro trabajo, también leído solo en su resumen, envió cada pregunta una vez a cinco chatbots en sus interfaces públicas entre el 1 y el 10 de mayo de 2026, y reunió 210 respuestas26. Con una instrucción que pedía expresamente seguridad, 199 de esas 210 respuestas (94.8%) cumplieron el criterio de seguridad fijado de antemano; las 11 que no lo cumplieron se concentraron en tres preguntas27. La exactitud fue alta en todos los modelos, sin diferencias detectables28. La empatía sí varió, con Gemini en la puntuación mediana más alta29. Y la legibilidad dejó mucho que desear: solo 2 de 210 respuestas (1.0%) alcanzaron el nivel de lectura que se suele recomendar30. Los propios autores advierten que estos resultados se obtuvieron con una instrucción concreta y no demuestran que los modelos sean seguros por sí solos31.

Un tercer estudio, del que también leímos solo el resumen, evaluó cinco modelos de lenguaje con 20 preguntas frecuentes de pacientes con cataratas, repartidas en cinco temas32. Su objetivo era medir qué tan legibles, de qué calidad y qué tan adecuadas para educar eran las respuestas sobre cataratas33. Hallaron diferencias entre modelos en legibilidad, calidad de la información y adecuación educativa34. El modelo con mejores puntuaciones de calidad también mostró índices de dificultad de lectura más altos35. Los temas de manejo posoperatorio y de prevención tendieron a ser más adecuados, mientras que los de diagnóstico, tratamiento y manejo preoperatorio resultaron más difíciles de leer36. La relación entre calidad y legibilidad fue en general débil37.

Así podría verse · ilustración generada por weeklyAI.watch, no es una fotografía

Lea esos tres trabajos como lo que son: estudios que también evaluaron chatbots con preguntas escritas de antemano, no con pacientes reales. Todos comparten el mismo diseño de fondo que el que nos ocupa. Y todos apuntan a lo mismo: la calidad de estas respuestas cambia según el modelo y según el tema38.

Lo que los autores del estudio sobre la orina proponen como siguiente paso es concreto: vigilar y mejorar lo que producen los chatbots, y dar a los pacientes orientación sobre cómo usar e interpretar esas respuestas, para que la atención se mantenga alineada con las guías clínicas39. Otros investigadores, en los resúmenes que leímos, sugieren que estos sistemas se usen solo como apoyo educativo supervisado, con sus respuestas planteadas como complemento —nunca como sustituto— de la atención profesional para cada procedimiento40.

Así lo leemos nosotros. Cuando una voz segura afirma algo, sobre todo si uno está solo frente a ella y no tiene a mano otra versión de los hechos, lo más fácil es aceptarlo. Por eso esperamos que muchas familias, al leer que el chatbot dice "es una infección urinaria", se orienten hacia los antibióticos aunque el análisis no muestre síntomas claros. Sabríamos que nos equivocamos si viéramos que la familia, ante esa respuesta, la compara con lo que dice su médico, pregunta por la diferencia entre tener bacterias en la orina y tener una infección, o simplemente no actúa hasta tener más información. Lo que usted puede hacer con esto es sencillo: cuando un chatbot le diga que es una infección urinaria, anote esa respuesta y llévela a su médico o al médico de su familiar para contrastarla, en lugar de tratarla como una receta. Y pregúntese en voz alta: ¿esto lo dice un aparato o una persona que conoce a mi familia?

Hay algo más que nos parece importante. Estas herramientas no lo escuchan a usted ni conocen el caso concreto: responden con seguridad, pero no preguntan por el contexto completo ni reconocen sus límites. Si un chatbot le afirma un diagnóstico sin haberle preguntado nada sobre la persona real, trátelo como una voz que no conoce a su familiar y busque a alguien que sí lo conozca. Fíjese también en si le ofrece citas o referencias: en este estudio, muy pocas respuestas lo hicieron. Y en la consulta, pregunte sin miedo: ¿este análisis solo, sin síntomas, necesita antibiótico? Pida que le expliquen la diferencia entre tener bacterias en la orina y tener una infección que requiere tratamiento. Esa pregunta, hecha a una persona que sí puede examinar a su familiar, vale más que cualquier respuesta rápida en una pantalla.

De dónde sale cada dato de contexto, y cuánto leímos de cada documento

  1. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The eight questions were entered into three publicly available AI chatbots (ChatGPT v5, Gemini v3, and Copilot v5) on December 15, 2025, using private sessions with no memory enabled."
  2. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "We created two clinical scenarios of hypothetical medical advice questions involving a positive urinalysis without specific UTI symptoms (i.e., ASB) that asked explicitly whether the parent had a UTI (Table 1)."
  3. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Scenario A described only a positive test, whereas scenario B included a fever and malodorous urine."
  4. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "For each scenario, we varied the parents’ gender (mother vs. father) and age (68 vs. 84 years)."
  5. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Most responses endorsed the suggestion of UTI (22/24, 91.7%; Table 2)."
  6. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Responses frequently acknowledged alternative explanations for symptoms (17/24, 70.9%) and over half mentioned ASB (16/24, 66.7%)."
  7. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Antibiotics were mentioned in 8 responses (33.3%), with 6 of those noting potential harms."
  8. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Most responses recommended seeking immediate care (19/24, 79.2%)."
  9. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Tone varied: 9 (37.5%) stressed urgency (e.g., “this is not something to wait on—he needs urgent medical evaluation.”), 5 (20.8%) were reassuring (e.g., “I hear your concern”), and 10 (41.7%) neutral."
  10. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Only 7 responses (29.2%) provided citations."
  11. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Responses varied between scenarios. For instance, antibiotic harms were given for scenario A but not scenario B."
  12. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Within scenarios (Table 3), responses varied by chatbot (e.g., mentioning ASB [0/4; Gemini] vs. [2/4; ChatGPT] vs. [2/4; Copilot]), parent’s gender (e.g., antibiotics suggested [1/6; father] vs. [3/6; mother]); and age (e.g., alternative diagnoses suggested [4/6; 68 years] vs. [1/6; 84 years])."
  13. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Artificial intelligence (AI) chatbots are increasingly used by patients and caregivers seeking quick, conversational answers to medical questions."
  14. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "This trend raises concerns about the clinical accuracy and appropriateness of responses, which can influence illness beliefs, treatment expectations, and care-seeking."
  15. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "These concerns are relevant for conditions like asymptomatic bacteriuria (ASB), which is often mistaken for urinary tract infection (UTI), leading to unnecessary antibiotic use."
  16. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship & Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "We examined how AI chatbots respond to simulated caregiver questions about ASB/UTIs."
  17. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Large language models (LLMs) are being widely used for accessing health information and facilitating doctor-patient communication, but their value and risks in end-of-life decision-making support for older adults have yet to be systematically evaluated."
  18. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "This study aimed to benchmark the safety, accuracy, observable empathetic communication, information quality, transparency indicators, and readability of responses generated by five publicly accessible consumer-facing generative AI systems to standardized end-of-life questions relevant to surrogate decision-makers for older adults."
  19. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "The question set was developed using an evidence-informed approach incorporating clinical guidelines, consensus statements, relevant literature, published qualitative studies involving surrogate decision-makers and family caregivers, and public search terminology."
  20. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Safe-response rates ranged from 76.7% to 86.7% across systems, with no evidence of an overall between-system difference in safety (Cochran’s Q = 1,368, df = 4, P = 0.850; effect size = 0.011, 95% CI 0.005–0.117)."
  21. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Accuracy differed significantly across systems, although the overall effect was small ( P = 0.003; Kendall’s W = 0.132, 95% CI 0.036–0.332)."
  22. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Differences in expert-rated textual empathy were more pronounced ( P < 0.001; Kendall’s W = 0.373, 95% CI 0.194–0.598)."
  23. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Significant between-system differences were also observed for DISCERN ( P < 0.001; W = 0.332), EQIP ( P = 0.016; W = 0.102), JAMA transparency benchmarks ( P < 0.001; W = 0.751), and GQS ( P = 0.001; W = 0.148)."
  24. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Each question was submitted once, in a new session, to five consumer-facing generative AI systems under their default web-interface conditions on June 14, 2026, yielding 150 paired system–question outputs."
  25. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Two blinded clinical experts independently evaluated the outputs using predefined criteria for safety, accuracy, textual empathy, DISCERN, EQIP, JAMA transparency benchmarks, GQS, and six readability indices."
  26. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Between May 1 and May 10, 2026, each question was submitted once to five generative artificial intelligence-driven chatbots through their official public interfaces, yielding 210 responses."
  27. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Under the standardized safety-oriented prompt, 199 of 210 responses (94.8%) met the study's prespecified binary safety criterion; the 11 responses that did not meet the criterion were concentrated in three questions and were summarized descriptively."
  28. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Accuracy ratings were high across models, with no statistically detectable difference ( P = 0.572)."
  29. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Empathy differed across models ( P = 0.003; Kendall's W = 0.096), with Gemini showing the highest median score."
  30. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Readability was suboptimal: only 2 of 210 responses (1.0%) met the commonly recommended sixth-grade threshold by FKGL."
  31. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "These prompt-conditioned findings do not establish unguided LLM responses as generally safe."
  32. Hou Y, Liu M, Wang Z, Jiang X. (2026). Performance evaluation of mainstream large language models in cataract science popularization Q&A: a comparative study of readability, quality, and educational suitability. Frontiers in Ophthalmology. https://doi.org/10.3389/fopht.2026.1943376 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Five mainstream LLMs - Doubao, DeepSeek, Wenxin Yiyan, Tongyi Qianwen, and GPT-5 - were evaluated based on their responses to 20 frequently asked questions (FAQs) for cataract patient, which covered five subject categories."
  33. Hou Y, Liu M, Wang Z, Jiang X. (2026). Performance evaluation of mainstream large language models in cataract science popularization Q&A: a comparative study of readability, quality, and educational suitability. Frontiers in Ophthalmology. https://doi.org/10.3389/fopht.2026.1943376 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "This study systematically evaluated the readability, quality, and educational suitability of mainstream LLMs when answering questions related to cataract."
  34. Hou Y, Liu M, Wang Z, Jiang X. (2026). Performance evaluation of mainstream large language models in cataract science popularization Q&A: a comparative study of readability, quality, and educational suitability. Frontiers in Ophthalmology. https://doi.org/10.3389/fopht.2026.1943376 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "There were significant differences among LLMs in terms of readability, information quality, and educational suitability (all p < 0.05)."
  35. Hou Y, Liu M, Wang Z, Jiang X. (2026). Performance evaluation of mainstream large language models in cataract science popularization Q&A: a comparative study of readability, quality, and educational suitability. Frontiers in Ophthalmology. https://doi.org/10.3389/fopht.2026.1943376 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "GPT-5 had the highest c-PEMAT-P and GQS scores, however, several readability difficulty indices of GPT-5 and Tongyi Qianwen were also higher."
  36. Hou Y, Liu M, Wang Z, Jiang X. (2026). Performance evaluation of mainstream large language models in cataract science popularization Q&A: a comparative study of readability, quality, and educational suitability. Frontiers in Ophthalmology. https://doi.org/10.3389/fopht.2026.1943376 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "The postoperative management and risk/prevention topics tended to have better educational suitability, whereas surgical diagnosis, treatment, and preoperative management were more difficult to read."
  37. Hou Y, Liu M, Wang Z, Jiang X. (2026). Performance evaluation of mainstream large language models in cataract science popularization Q&A: a comparative study of readability, quality, and educational suitability. Frontiers in Ophthalmology. https://doi.org/10.3389/fopht.2026.1943376 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Correlation analysis demonstrated that the correlation between the quality indicators and the readability-related indicators is generally weak."
  38. Hou Y, Liu M, Wang Z, Jiang X. (2026). Performance evaluation of mainstream large language models in cataract science popularization Q&A: a comparative study of readability, quality, and educational suitability. Frontiers in Ophthalmology. https://doi.org/10.3389/fopht.2026.1943376 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "In terms of generating educational materials for cataract patient, LLMs have potential, but the output quality varies depending on the model and the topic."
  39. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Monitoring and improvement of chatbot outputs, alongside patient-facing guidance on how to use and interpret responses, are needed to support guideline-aligned care."
  40. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "LLMs may serve only as supervised adjunctive educational tools, with outputs framed as complementary to procedure-specific professional care."

Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512

Quién pagó: El artículo no dice quién financió el estudio, si los financiadores tuvieron alguna influencia, ni quién prestó equipos o programas.

No tome esto como consejo médico profesional.

experiment · Antimicrobial Stewardship & Healthcare Epidemiology · the paper, 1 Aug 2026 · free

A Chatbot Says It Is a UTI. It Often Is Not.

In a small test of 24 made-up questions about a positive urine test, three chatbots went along with the UTI idea 22 times. The questions were not from real families, and the researchers did not follow anyone.

Short version · the longer version follows, about 7 min

Ask the weeklyAI chat

Ask me about this study: who was studied, what it found, and what it does not say.

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
made-up questions about an older parent, typed as a caregiver would
How many
24 answers
Where
not stated in the passages
When
December 15, 2025
Kind of study
analysis of what people did
Who did it
University of Utah and University of Alabama at Birmingham
The limit that matters
Questions were made up, not from real families; no one was followed.
What the 24 chatbot answers said about a positive urine test
went along with the idea it was a urinary tract infection91.7%
named asymptomatic bacteriuria66.7%
told the family to seek care right away79.2%
gave any citations29.2%

Shares of the 24 answers the three chatbots gave to eight made-up questions about a positive urine test with no clear urinary symptoms; the questions were not from real families.

The same kind of question, answered differently

a 68-year-old parentagainstan 84-year-old parent

other possible causes came up in 4 of 6 answers for the 68-year-old and 1 of 6 for the 84-year-old

a fatheragainsta mother

antibiotics were suggested in 1 of 6 answers for a father and 3 of 6 for a mother

GeminiagainstChatGPT and Copilot

asymptomatic bacteriuria was named in 0 of 4 answers by Gemini, 2 of 4 by ChatGPT and 2 of 4 by Copilot

When a chatbot answers a urine-test worry, look for whether it names asymptomatic bacteriuria and asks about symptoms before it mentions antibiotics.
weeklyAI's reading
Watch · 0:52

The idea that a positive urine test means an infection is wrong for many older adults, and chatbots repeat it.

Researchers typed eight made-up questions into three chatbots: ChatGPT v5, Gemini v3 and Copilot v5. All eight described a parent with a positive urine test and no clear urinary symptoms. Some of the questions said the parent was a mother, some a father, and some said the parent was 68 years old, some 84. One version added a fever and smelly urine.

Of the 24 answers, 22 went along with the idea that it was a UTI. That is 91.7 percent.

Only 16 answers, two out of three, used the name for the real condition: asymptomatic bacteriuria. That means bacteria in the urine without symptoms. It is common in older adults, and it does not need antibiotics.

Antibiotics came up in 8 answers. Six of those mentioned that they can cause harm. Most answers, 19 of 24, told the family to seek care right away. The article says most answers told the family to seek care right away. It does not say whether that advice was right or wrong.

The answers also disagreed with each other. The same question got different replies from different apps, and replies changed with the parent's age or gender. A mention of other possible causes appeared in 4 of 6 answers about a 68-year-old, but only 1 of 6 about an 84-year-old. The study does not say why the answers changed with age.

The questions were made up, not taken from real caregivers, so no one can say how these apps would answer your own family's situation. Only three apps were tested, on one day, December 15, 2025, so a different app or a later version may answer differently. The study counted what the chatbots said. It did not follow anyone to see whether they took antibiotics or got sick.

When a chatbot answers a urine-test worry, look for whether it names asymptomatic bacteriuria and asks about symptoms before it mentions antibiotics.

What this means for you

When a chatbot answers a urine-test worry about your parent, notice whether it names asymptomatic bacteriuria and asks about symptoms before it brings up antibiotics, because the study found this did not always happen. The questions were made up, so no one can say how these apps would answer your own family's case, and a doctor is the one to ask about any test result.

Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512

Who paid: The article does not say who funded the study, whether funders had any say, or who lent equipment or software.

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 · 1415 words · about 7 minRead it →Close

A Urine Test, a Chatbot, and a UTI That May Not Be There

A small study typed eight worried questions about older parents into three AI chatbots. Almost every answer said the same thing: it's a urinary tract infection. The test alone doesn't say that.

How it could look · illustration generated by weeklyAI.watch, not a photograph

The study set out to see how artificial intelligence chatbots answer the kind of medical question a caregiver might type late at night, worried about an older parent and a urine test that came back positive1. The researchers wrote two pretend scenarios, each one a question about an older parent that asked, in effect, "does my mother have a urinary tract infection?" In both scenarios the parent had a positive urine test but no specific urinary symptoms — the situation doctors call asymptomatic bacteriuria, or ASB2. One scenario mentioned only the positive test; the other added a fever and strong-smelling urine3. For each scenario they changed the parent's gender (mother or father) and age (68 or 84)4. That made eight questions.

Those eight questions went into three widely used chatbots — ChatGPT v5, Gemini v3, and Copilot v5 — on December 15, 2025, in private sessions with no memory switched on5. Two team members coded the answers separately, and two more reviewers settled disagreements. There were 24 answers in all.

Most of them — 22 of 24, or just over nine in ten — endorsed the suggestion that the parent had a urinary tract infection6. About seven in ten also mentioned other possible explanations for the symptoms, and about two-thirds brought up asymptomatic bacteriuria by name7. Antibiotics came up in 8 answers, and in 6 of those the chatbot noted possible harms8. Most answers, 19 of 24, told the caregiver to seek care right away9. The tone split three ways: 9 stressed urgency, 5 were reassuring, 10 were neutral10. Only 7 answers, under three in ten, gave any citations11.

The answers also disagreed with each other. Harms from antibiotics were mentioned in the first scenario but not the second12. Within the same scenario, answers changed with the chatbot, the parent's gender, and the parent's age: one chatbot mentioned ASB in none of four answers, another in two of four, a third in two of four; antibiotics were suggested in one of six answers for a father and three of six for a mother; other possible diagnoses appeared in four of six answers for a 68-year-old and one of six for an 84-year-old13.

What this study did not do matters as much as what it did. The questions were written by the researchers as pretend scenarios, not asked by real caregivers. Only three chatbots were tested, on a single day, so a different app or a later version may answer differently. The study counted what the chatbots said; it did not follow anyone to see whether they took antibiotics or got sick. The study reported that answers varied by platform, gender and age.

How it could look · illustration generated by weeklyAI.watch, not a photograph

The reason researchers are looking at this at all is that people are already asking chatbots their medical questions. The study's authors note that AI chatbots are increasingly used by patients and caregivers looking for quick, conversational answers to medical questions14. They add that this trend raises concerns about how clinically accurate and appropriate those answers are, because they can shape what people believe about an illness, what treatment they expect, and whether they go for care15. Asymptomatic bacteriuria is a good test case: it is often mistaken for a urinary tract infection, and that mistake leads to unnecessary antibiotic use16.

The mechanism here is simple enough to picture. You type a question the way you would ask a relative: my mother's urine test was positive, does she have a UTI? The chatbot produces a fluent, conversational reply, and in this study only 7 of 24 answers gave any citations. In this study, that reply usually took the shape of the question — a UTI — even when the facts described were the textbook picture of ASB, a positive test without the symptoms that define an infection2. Some answers did mention ASB7; most did not. The inconsistency across platforms, gender and age13 means the same family could get different answers from different apps about the same test result.

This is one study in a growing family of studies checking whether these tools give sound health advice. We could read only the summaries of the others. One benchmarked five publicly accessible AI systems against standardized end-of-life questions for people making decisions on behalf of older adults; safe-response rates ranged from 76.7% to 86.7% across systems, with no evidence of an overall difference between them in safety17. In that study, accuracy did differ between systems, though the overall effect was small18, and differences in expert-rated empathy were more pronounced19. Another study submitted questions to five chatbots through their public interfaces and found that under a standardized safety-oriented prompt, 199 of 210 responses met the study's safety criterion20, while accuracy ratings were high with no statistically detectable difference between models21; the authors caution that these prompt-conditioned findings do not establish unguided chatbot responses as generally safe22. A third evaluated five mainstream language models on 20 common questions from cataract patients and found significant differences between them in readability, information quality and educational suitability23.

Put the studies side by side and a pattern emerges that no single one of them proves. In each case the tool being tested was a general-purpose chatbot answering a health question in ordinary conversation, and in each case the researchers had to build a scoring system — safety, accuracy, readability, whether the answer named the right condition. The ASB study used eight questions, one day, three apps. What it shares with the others is the finding that quality varies — by system, by question, by wording.

How it could look · illustration generated by weeklyAI.watch, not a photograph

Here is how we read it. The pattern in this study is not that chatbots are wrong. It is that they answer the question you asked, and the question you asked already contained the answer. The caregiver's question was whether the parent had a UTI2, and the reply mostly agreed6. We would expect that in homes like yours, the same thing happens: a worried son or daughter types a urine result into a phone, receives a confident paragraph that says infection, and feels the matter is settled — even though a positive urine test without symptoms is exactly the situation where the label does not fit16. What you can do with this: when a chatbot tells you a urine result means infection, notice whether it gave you the symptoms that would make that true — burning, urgency, frequency, fever with no other cause — and whether it mentioned asymptomatic bacteriuria at all. If it did neither, you are holding a confident answer to a different question than the one your family is facing.

We also read other work on how these tools handle health questions more broadly. In one study of end-of-life questions for people deciding on behalf of older adults, accuracy differed between systems, though the overall effect was small, and differences in expert-rated empathy were more pronounced, while there was no evidence of an overall difference in safety. In another, the findings under a standardized safety-oriented prompt do not establish unguided chatbot responses as generally safe. In a third, on cataract questions, the models differed significantly in readability, information quality and educational suitability. None of this tells us what happened in this study.

What would have to happen next, according to the study's authors, is monitoring and improvement of chatbot outputs, alongside guidance for patients on how to use and interpret the responses, so that care follows guidelines24. One of the other teams we read said that these tools may serve only as supervised educational adjuncts, with their output framed as complementary to procedure-specific professional care25. Another found that the models have potential for producing educational material but that output quality varies by model and topic26.

So here is the thing to carry with you. The next time a chatbot, or any confident voice, tells you a urine test means infection, you can ask a question that costs nothing and changes the conversation: which symptoms does this person actually have? A positive test on its own is not that answer, and no app can examine your parent. When the answer comes, bring it to a clinician who can — and notice, while you wait, whether what convinced you was the medical claim or the warm way it was delivered.

Where each piece of context comes from, and how much of it we read

  1. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "We examined how AI chatbots respond to simulated caregiver questions about ASB/UTIs."
  2. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "We created two clinical scenarios of hypothetical medical advice questions involving a positive urinalysis without specific UTI symptoms (i.e., ASB) that asked explicitly whether the parent had a UTI (Table 1)."
  3. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Scenario A described only a positive test, whereas scenario B included a fever and malodorous urine."
  4. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "For each scenario, we varied the parents’ gender (mother vs. father) and age (68 vs. 84 years)."
  5. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "The eight questions were entered into three publicly available AI chatbots (ChatGPT v5, Gemini v3, and Copilot v5) on December 15, 2025, using private sessions with no memory enabled."
  6. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Most responses endorsed the suggestion of UTI (22/24, 91.7%; Table 2)."
  7. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Responses frequently acknowledged alternative explanations for symptoms (17/24, 70.9%) and over half mentioned ASB (16/24, 66.7%)."
  8. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Antibiotics were mentioned in 8 responses (33.3%), with 6 of those noting potential harms."
  9. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Most responses recommended seeking immediate care (19/24, 79.2%)."
  10. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Tone varied: 9 (37.5%) stressed urgency (e.g., “this is not something to wait on—he needs urgent medical evaluation.”), 5 (20.8%) were reassuring (e.g., “I hear your concern”), and 10 (41.7%) neutral."
  11. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Only 7 responses (29.2%) provided citations."
  12. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Responses varied between scenarios. For instance, antibiotic harms were given for scenario A but not scenario B."
  13. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Within scenarios (Table 3), responses varied by chatbot (e.g., mentioning ASB [0/4; Gemini] vs. [2/4; ChatGPT] vs. [2/4; Copilot]), parent’s gender (e.g., antibiotics suggested [1/6; father] vs. [3/6; mother]); and age (e.g., alternative diagnoses suggested [4/6; 68 years] vs. [1/6; 84 years])."
  14. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Artificial intelligence (AI) chatbots are increasingly used by patients and caregivers seeking quick, conversational answers to medical questions."
  15. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "This trend raises concerns about the clinical accuracy and appropriateness of responses, which can influence illness beliefs, treatment expectations, and care-seeking."
  16. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "These concerns are relevant for conditions like asymptomatic bacteriuria (ASB), which is often mistaken for urinary tract infection (UTI), leading to unnecessary antibiotic use."
  17. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — only the abstract - the full text could not be fetched — the passage: "Safe-response rates ranged from 76.7% to 86.7% across systems, with no evidence of an overall between-system difference in safety (Cochran’s Q = 1.368, df = 4, P = 0.850; effect size = 0.011, 95% CI 0.005–0.117)."
  18. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — only the abstract - the full text could not be fetched — the passage: "Accuracy differed significantly across systems, although the overall effect was small ( P = 0.003; Kendall’s W = 0.132, 95% CI 0.036–0.332)."
  19. Liu Z, Zhou X, Zhu Z, Ling J, Cao F. (2026). Quality, safety, and readability of consumer-facing generative AI responses to end-of-life questions relevant to surrogate decision-makers for older adults: a benchmark evaluation. BMC Palliative Care. https://doi.org/10.1186/s12904-026-02360-1 — only the abstract - the full text could not be fetched — the passage: "Differences in expert-rated textual empathy were more pronounced ( P < 0.001; Kendall’s W = 0.373, 95% CI 0.194–0.598)."
  20. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — only the abstract - the full text could not be fetched — the passage: "Under the standardized safety-oriented prompt, 199 of 210 responses (94.8%) met the study's prespecified binary safety criterion; the 11 responses that did not meet the criterion were concentrated in three questions and were summarized descriptively."
  21. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — only the abstract - the full text could not be fetched — the passage: "Accuracy ratings were high across models, with no statistically detectable difference ( P = 0.572)."
  22. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — only the abstract - the full text could not be fetched — the passage: "These prompt-conditioned findings do not establish unguided LLM responses as generally safe."
  23. Hou Y, Liu M, Wang Z, Jiang X. (2026). Performance evaluation of mainstream large language models in cataract science popularization Q&A: a comparative study of readability, quality, and educational suitability. Frontiers in Ophthalmology. https://doi.org/10.3389/fopht.2026.1943376 — only the abstract - the full text could not be fetched — the passage: "There were significant differences among LLMs in terms of readability, information quality, and educational suitability (all p < 0.05)."
  24. Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512 - the article this story is about — the whole article — the passage: "Monitoring and improvement of chatbot outputs, alongside patient-facing guidance on how to use and interpret responses, are needed to support guideline-aligned care."
  25. Hu G, Lai Q, Liu S, Xie B, Wu Z, Zhuang Q, et al. (2026). Performance of large language models in postoperative hip fracture rehabilitation counseling for older adults: a comparative evaluation of safety, accuracy, reliability, readability, and empathy. Frontiers in Public Health. https://doi.org/10.3389/fpubh.2026.1961597 — only the abstract - the full text could not be fetched — the passage: "LLMs may serve only as supervised adjunctive educational tools, with outputs framed as complementary to procedure-specific professional care."
  26. Hou Y, Liu M, Wang Z, Jiang X. (2026). Performance evaluation of mainstream large language models in cataract science popularization Q&A: a comparative study of readability, quality, and educational suitability. Frontiers in Ophthalmology. https://doi.org/10.3389/fopht.2026.1943376 — only the abstract - the full text could not be fetched — the passage: "In terms of generating educational materials for cataract patient, LLMs have potential, but the output quality varies depending on the model and the topic."

Luhan, E., Lee, R., Fagerlin, A. et al. (2026). Evaluation of AI Chatbot Responses to Simulated Caregiver Questions About Asymptomatic Bacteriuria. Antimicrobial Stewardship &amp; Healthcare Epidemiology. https://doi.org/10.1017/ash.2026.10512

Who paid: The article does not say who funded the study, whether funders had any say, or who lent equipment or software.

Do not take this as professional medical advice.