"It Is Not Always 100% Correct. I Find It Very Useful and Amazing."

survey · PLoS ONE · la publicación, 23 sep 2026 · gratis

"No siempre es 100% correcto, pero me resulta muy útil y asombroso"

Un estudio con 1,300 cuidadores en Estados Unidos encontró que una cuarta parte de los que usan chatbots los usan para cuidar a alguien. Es una sola foto en el tiempo y no puede decir si eso ayuda o perjudica.

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: personas que cuidan a un familiar o amigo
Cuántos: 1,300
Dónde: Estados Unidos
Cuándo: entre agosto y octubre de 2025
Tipo de estudio: survey
Quién lo hizo: Yale, Universidad del Sur de California y Michigan
El límite que importa: Es una sola foto en el tiempo: no puede probar que una cosa cause la otra.

Para qué usan el chatbot quienes cuidan

Para pedir información sobre el cuidado58%

Para apoyo emocional relacionado con el cuidado34%

Para otros usos37%

Porcentajes de las 121 personas que dijeron usar el chatbot para cuidar; cada persona pudo marcar más de una opción, por eso suman más de 100.

Es una fotografía de un momento, no una película: el estudio no siguió a nadie en el tiempo ni comparó a nadie con nadie, así que no puede decir si apoyarse en un chatbot ayuda o perjudica.
Lectura de weeklyAI

Ver · 0:49

Los investigadores de Yale, la Universidad del Sur de California y Michigan preguntaron a 1,300 personas que cuidan a un familiar o amigo en Estados Unidos. Las encuestas se hicieron entre agosto y octubre de 2025, dentro de un panel nacional llamado Understanding America Study.

De esas 1,300 personas, 487 habían usado alguna vez un chatbot como ChatGPT, Gemini o Copilot. De ellos, 121, es decir una de cada cuatro, lo habían usado para su tarea de cuidar.

La mayoría fue a buscar información: el 58%. Otros buscaron apoyo emocional, el 34%, y un 37% dijo que lo usaba para otras cosas. Una persona lo resumió así: "Al darme cuenta de que no siempre es 100% correcto, me resulta muy útil, muy útil y asombroso".

Quienes más recurrían al chatbot tenían mayor apertura a lo nuevo, menor conocimiento de internet y menor puntaje en una prueba de razonamiento. También consultaban más quienes ayudaban con más tareas y quienes se sentían más inseguros en su papel de cuidadores.

Eso no significa que el chatbot cause nada. Es un estudio de un solo momento, una foto, y no puede probar que una cosa lleve a la otra. Los propios autores lo advierten: puede ser que quienes manejan peor otras fuentes de información recurran a la máquina porque es fácil, o puede ser que confíen en ella sin notar sus errores.

Hay otra razón para tomar el 25% con pinzas. Cada persona decidió por sí misma qué contaba como uso para cuidar. Al leer sus respuestas, los investigadores vieron que varias hablaban de trabajo o de preguntas generales, no de cuidar a alguien.

Todo esto pasó en Estados Unidos. El estudio no dice nada sobre cómo se usan estas herramientas en América Latina ni en Canadá. Tampoco midió si los cuidadores están mejor o peor por usarlas.

El estudio encontró que quienes se sentían más inseguros en su papel de cuidadores usaban más el chatbot. Los autores también advierten que quienes manejan peor otras fuentes de información pueden confiar en la máquina sin notar sus errores.

Qué significa para usted

Ese 25% es una foto de un solo momento en Estados Unidos, así que no concluya que el chatbot ayude ni que perjudique a quien cuida. Fíjese más bien en lo que usted le pide: si es información para cuidar a alguien, conviene confirmarla con un profesional antes de actuar.

Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753

Quién pagó: El estudio fue financiado por el National Institute on Aging (NIH/NIA, subvención R01AG083097, UAS-CLEAR); el artículo no indica qué papel tuvo el financiador, y no se menciona ningún proveedor de equipos o software.

No tome esto como consejo médico profesional.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1506 palabras · unos 8 minLeerla →Cerrar

Quienes cuidan y dudan son los que más le preguntan a un chatbot

Un sondeo nacional en Estados Unidos encontró que una de cada cuatro personas que usan chatbots de inteligencia artificial también los usan para cuidar a alguien, sobre todo para pedir información.

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

Una encuesta hecha entre agosto y octubre de 2025 a 1,300 personas de Estados Unidos que cuidan a un familiar o amigo encontró que 487 de ellas usaban aplicaciones de inteligencia artificial como ChatGPT, Gemini, Copilot o Grok. De esas 487, una de cada cuatro —121 personas— dijo que las usaba también para su tarea de cuidado¹². Es una fotografía de un momento, no una película: el estudio no siguió a nadie en el tiempo ni comparó a nadie con nadie, así que no puede decir si apoyarse en un chatbot ayuda o perjudica¹.

¿Para qué las usan? Más de la mitad, el 58%, para pedir información sobre el cuidado; un 34% para apoyo emocional relacionado con el cuidado; y un 37% marcó "otros" usos³. En las respuestas abiertas, que fueron 128, el tema más común fue otra vez la búsqueda de información: 72 personas describieron eso, frente a 9 que describieron apoyo emocional. Un 38% dijo que el chatbot les resultaba útil, con frases como "rápido y fácil"⁴.

El estudio también buscó qué tipo de persona tiende a usar estas herramientas para cuidar. Encontró que quienes tienen mayor apertura a lo nuevo —una de las cinco grandes rasgos de personalidad— mostraban una probabilidad algo mayor de hacerlo⁵. Y encontró lo contrario de lo que uno esperaría: quienes tenían mayor capacidad cognitiva y mayor alfabetización digital eran menos propensos a apoyarse en un chatbot para cuidar⁶. Quienes se sentían más inseguros sobre cómo cuidar a la otra persona eran más propensos a usarlo⁷. Nada de esto es causa y efecto: son asociaciones dentro de un solo corte en el tiempo¹.

Vale la pena detenerse en qué es esta tecnología, porque no es la de antes. El artículo describe un estudio en el que se explica que un chatbot de inteligencia artificial es un programa que conversa de manera parecida a una persona. Los primeros eran de reglas fijas: respondían con guiones ya escritos según lo que uno escribía. Los actuales generan la respuesta en el momento, de forma dinámica, y por eso pueden dar una experiencia personalizada y ajustada al contexto⁸. El artículo describe un estudio en el que se señala su atractivo para quien cuida: están disponibles a toda hora, no hay que pedir turno, uno puede preguntar sin vergüenza y casi siempre son gratis o muy baratos⁹.

El atractivo no nace de la nada. La propia investigación recuerda que hoy hay unos 63 millones de adultos en Estados Unidos cuidando a alguien con un problema serio de salud, frente a 43,5 millones en 2015, y que la cifra subirá mucho en las próximas tres décadas por el envejecimiento de la población¹⁰. El artículo describe un estudio en el que se señala que, al mismo tiempo, la distancia física, los horarios impredecibles, los costos, el estigma social y el simple desconocimiento de las opciones disponibles hacen que muchos cuidadores no lleguen a los apoyos que existen¹¹. Un chatbot no arregla eso, pero entra por donde esas barreras dejan un hueco.

No es un fenómeno de nicho. El artículo describe un estudio en el que, entre julio de 2023 y comienzos de 2025, el porcentaje de adultos estadounidenses que usaban ChatGPT pasó del 18% al 34%¹². Es decir, en menos de dos años se duplicó la porción de una población entera que ya conversa con una máquina. Y el uso que este estudio observa en el cuidado es, en su mayoría, el más básico: pedir datos concretos.

Así lo leemos nosotros. Cuando un chatbot responde algo sobre salud o sobre cuidados, anote la pregunta exacta que hizo y contrástela después con una persona o una fuente que pueda verificar. Si la respuesta no cambia al reformularla, trátela como un dato a confirmar, no como un consejo. Y ojo con una trampa silenciosa: si la respuesta es siempre segura, siempre amable, siempre de acuerdo con usted, eso no es señal de que sea correcta. Es la forma que tiene un programa de no incomodarlo. Usted puede notar que nos equivocamos si los cuidadores que más dudan resultan ser, con el tiempo, los que menos consultan al chatbot, o si lo consultan sobre todo para sentirse acompañados y no para pedir información.

El estudio encontró que quienes tenían mayor capacidad cognitiva y mayor alfabetización digital eran menos propensos a apoyarse en un chatbot para cuidar, y que quienes se sentían más inseguros sobre cómo cuidar a la otra persona eran más propensos a usarlo⁶⁷. Eso admite dos lecturas opuestas, y el estudio no puede separarlas: puede ser que el chatbot sea un canal accesible para quien no sabe navegar la literatura médica, los profesionales o los programas comunitarios¹³; o puede ser que quien tiene menos capacidad de detectar los límites de una respuesta sea justamente quien más se apoye en ella sin buscar confirmación, mientras que los más alfabetizados digitalmente eviten el chatbot a propósito para decisiones de cuidado¹⁴. Las dos caben en los datos. Ninguna está probada. Lo que sí sabemos es que estas características —capacidad cognitiva y alfabetización digital— cambian muy poco con el tiempo, así que más encuestas difícilesmente aclaren el mecanismo¹⁴.

Nos detenemos también en cómo se midió la palabra "cuidado". Los investigadores decidieron a propósito no imponer una definición estrecha y dejar que cada persona juzgara qué cuenta como uso para cuidar¹⁵. Eso hace que el 25% sea más flojo de lo que suena: el examen de las respuestas abiertas indica que un subconjunto de quienes reportaron uso para cuidar describió usos solo indirectamente relacionados o no relacionados, como tareas de trabajo, fines creativos o búsqueda general de información¹⁵. Si usted ve una cifra sobre cuánta gente usa estas herramientas para cuidar, pregunte cómo se definió esa palabra en la encuesta. Y cuando hable con alguien que cuida, déjelo contar con sus propias palabras en vez de encasillarlo en opciones.

Lo que todavía no existe es lo que más falta hace. El artículo describe un estudio en el que la investigación actual sobre cuidadores y chatbots se concentra en diseñar y evaluar chatbots hechos específicamente para cuidar, en medir qué tan buenas son las respuestas de los chatbots generales a preguntas de cuidado, y en estudios cualitativos pequeños¹⁶. Ningún estudio previo había mirado los rasgos de personalidad ligados al uso de estas herramientas para cuidar¹⁷. El artículo describe un estudio en el que las advertencias son serias: información inexacta por "alucinaciones" del sistema, consejos médicos o emocionales potencialmente peligrosos, empatía insuficiente, mal manejo de situaciones de crisis, y problemas de privacidad y seguridad¹⁸. Quien pide consejo sobre curas o tratamientos para la persona que cuida puede quedar expuesto a información falsa, sobre todo si no tiene cómo verificarla.

Otros trabajos que leímos, y de los que solo pudimos leer el resumen, apuntan en la misma dirección. Una revisión de ensayos sobre chatbots comerciales de salud mental encontró mejoras en depresión, resultados inciertos en ansiedad y una reducción parcial de la soledad, pero advierte que la evidencia sobre soledad es limitada y frágil y que, en ansiedad, no puede descartarse un daño neto. Esa revisión también señala que la mayoría de los ensayos no monitoreaba efectos adversos de forma activa, y que la retención en el mundo real probablemente sea menor que la de los estudios. Otra revisión, cuyo resumen pudimos leer, encontró que la mayoría de las intervenciones con chatbots de salud mental incorporaba al menos un mecanismo técnico de seguridad, pero que la supervisión humana era limitada y que los protocolos de derivación en crisis solían estar poco desarrollados; entre las fallas documentadas figuran casos en que no se detectó ideación suicida y casos en que se dio información clínica inexacta. Esas revisiones no evaluaron a cuidadores ni a los chatbots de uso general, así que no hablan de este estudio: hablan del terreno que lo rodea.

Lo que esto deja claro es que la pregunta útil no es si el chatbot es bueno o malo, sino para qué exactamente lo está usando usted y quién puede confirmarlo. La investigación propone que el próximo paso sea ayudar a los cuidadores a distinguir qué información es confiable y qué requiere consultar a un tercero o a un profesional, como un médico o una enfermera¹⁹. También propone estudiar si este uso se relaciona con mejor bienestar con el tiempo, y para quién resulta más o menos útil²⁰. Nada de eso está resuelto. Mientras tanto, usted puede hacer algo concreto: lleve un registro breve de las veces que el chatbot le sirvió y de las veces se repitió o se contradijo. Ese registro le sirve para decidir cuándo apoyarse en él y cuándo llevar la pregunta a una persona.

¿Cuál fue la última pregunta que le hizo a un chatbot y todavía no le ha contado a nadie que puede verificarla?

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

Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The sample consisted of 1,300 caregivers, with data collected between August and October 2025. We employed a cross-sectional, mixed-methods design, combining quantitative analysis of structured survey items with thematic coding of open-ended responses."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Of the 487 respondents who reported being AI chatbot users, 25% used the technology for caregiving. Those who were more likely to use AI for caregiving were higher in openness, lower in internet literacy, lower in cognitive ability, helped with more caregiving tasks, and were more uncertain in their caregiving role."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Over half (58%) used AI chatbots for information about caregiving, 34% for emotional support related to caregiving, and 37% reported other uses."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Forty-eight (38%) responses referred to how an AI chatbot is useful for caregiving purposes. For example, people described it as “quick and easy” or stated, “Realizing that it is not always 100% correct, I find it very useful and helpful and amazing”"
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Consistent with the patterns observed for general chatbot use (see S1Table ), a one-standard deviation increase in openness was associated with a 4–5 percentage point increase in the likelihood of using AI chatbots for caregiving support."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "In contrast, caregivers with higher cognitive ability and greater internet literacy were significantly less likely to rely on AI chatbots for caregiving assistance."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Notably, greater self-reported uncertainty about how to care for the care recipient was associated with a higher likelihood of using AI chatbots for caregiving support."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "AI chatbots are computer programs that engage in human-like conversations with users [ 8 ]. Previously, chatbots were rule-based programs, producing various scripted replies after processing user input. However, recent advancements in AI have resulted in the development of AI-powered chatbots that dynamically and interactively generate responses to user inputs, providing a personalized and context-dependent experience [ 9 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The primary advantages of using AI chatbots rather than or in addition to traditional resources are their constant availability, scalability, anonymity, and personalization capabilities. Additionally, most AI chatbots are inexpensive or entirely free to use [ 12 – 14 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "An estimated 63 million adults in the United States served as caregivers for people living with serious health conditions, up from 43.5 million in 2015 [ 1 ]. These numbers are expected to climb dramatically over the next 30 years due to population aging, as more caregivers need support."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Physical distance, unpredictable and busy schedules, costs, social stigma, and a lack of awareness regarding available options all contribute to caregivers not accessing informational and emotional support [ 5 – 7 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "From July 2023 to early 2025, the percentage of American adults using ChatGPT increased from 18% to 34%, demonstrating the rapid rise of AI chatbots and, therefore, the importance of understanding their possible societal impacts, from work to social relationships [ 10 , 11 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "One plausible mechanism, consistent with open-ended responses describing chatbots as “quick and easy” to use, is that AI chatbots provide an accessible information channel for caregivers who may find it harder to navigate other sources (e.g., medical literature, healthcare professionals, or community-based programs)."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "First, caregivers with lower cognitive ability or internet literacy may be less able to identify the limitations of chatbot output, leaving them more reliant on chatbots without seeking corroboration; conversely, more digitally literate caregivers may be more aware of these risks and deliberately avoid chatbots for caregiving-related decisions."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The question eliciting this key measure intentionally relied on respondents’ own judgment of what counts as caregiving-related use, rather than imposing a narrow functional criterion."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Current research on caregivers’ use of AI chatbots is concentrated primarily on the design and assessment of caregiving-specific chatbots, the evaluation of responses given by various chatbots to caregiving-related questions, and small-scale qualitative studies of caregiver perspectives and feedback on chatbot usage [ 12 , 20 , 21 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "No prior study has examined personality traits related to AI use for caregiving purposes. Hence, we investigate heterogeneity in the use of AI chatbots for caregiving purposes not only by demographic and socio-economic characteristics, but also by personality traits."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "These include providing inaccurate information due to AI hallucinations, potentially dangerous medical and emotional advice, inadequate empathy, mismanagement of crisis scenarios, and privacy and security issues [ 15 – 19 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Future research could address this by helping caregivers understand which information is accurate and which requires follow-up by a third party or a professional, such as a medical doctor or nurse."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Another future direction of this research is to examine whether the use of chatbots for caregiving is associated with improvements in well-being over time as well as to identify individual and contextual factors that predict for whom they are more or less useful in improving well-being."

No tome esto como consejo médico profesional.

Los hallazgos de otros estudios que aquí se mencionan los conocemos por este documento, que fue el que leímos; no abrimos cada uno de esos estudios.

survey · PLoS ONE · the paper, 23 Sep 2026 · free

"It Is Not Always 100% Correct. I Find It Very Useful and Amazing."

A snapshot of 1,300 US caregivers shows who leans on a chatbot, and it cannot show whether the leaning helps.

Short version · the longer version follows, about 5 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: family caregivers in the United States
How many: 1,300 caregivers
Where: United States
When: August to October 2025
Kind of study: survey
Who did it: Yale, University of Southern California and University of Michigan
The limit that matters: One snapshot in time, so it cannot show cause or effect

What the 121 caregivers used chatbots for

Information about caregiving58%

Emotional support about caregiving34%

Other uses37%

Shares of the 121 caregivers who used a chatbot for caregiving; people could pick more than one reason, so the shares add up to more than 100.

When a role feels uncertain, that is when the machine gets opened, and that is exactly when a wrong answer is hardest to catch.
weeklyAI's reading

Watch · 0:50

One caregiver wrote down what many people seem to feel. "Realizing that it is not always 100% correct, I find it very useful and helpful and amazing."

Researchers at Yale, the University of Southern California and the University of Michigan asked 1,300 family caregivers across the United States about AI chatbots. The caregivers answered between August and October 2025.

Of the 1,300, 487 had used a chatbot. One in four of those, 121 people, had used it for caregiving.

Information was the main draw. Of those 121, 58 percent asked about caregiving, 34 percent went for emotional support, and 37 percent for other reasons. In longer written answers, the pattern held. One caregiver said the chatbot helped with "basic questions," then added a line worth keeping: major medical issues, that person would not let it advise.

The people most likely to lean on a chatbot shared a few traits. They scored higher on openness. They helped with more caregiving tasks. They felt more uncertain about their role. And those with lower scores on internet skills and on a thinking test were more likely to use chatbots for care.

That last part cuts against common sense. One likely reason is that a chatbot is quick and easy when other sources, like medical offices or community programs, are hard to reach. The researchers also say people with lower digital skills may be less able to spot when a chatbot is wrong. The study cannot tell which explanation is right.

It is one snapshot, taken once, so it cannot show that chatbots help caregivers or harm them. The survey also asked caregivers how often they felt short of time, stressed, strained or unsure about their role.

Caregiving use was also whatever each person called it. Some written answers were really about work tasks or general questions, so the one-in-four figure is looser than it sounds.

The findings come from US caregivers only. Nothing here says how people in Latin America or Canada use these tools.

The study found that caregivers who felt more uncertain about their role were more likely to use a chatbot for care.

What this means for you

Among 1,300 U.S. caregivers surveyed in 2025, 487 had used an AI chatbot. Of those, about one in four said they used it for caregiving, mostly to ask for information. That does not mean it helps or harms, since this study only looked once and cannot say which. Watch for wrong answers, and ask a professional about medical decisions.

Who paid: The study was funded by the National Institute on Aging (NIH/NIA, grant R01AG083097, UAS-CLEAR); the article does not state any role played by the funder, and no equipment or software lenders are mentioned.

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 · 1071 words · about 5 minRead it →Close

Caregivers Are Asking Chatbots for Help. The Least Sure of Themselves Ask Most.

A survey of 1,300 U.S. caregivers finds a quarter of chatbot users bring them into caregiving — mostly for information, and most often when they feel unsure of the role.

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

Picture a caregiver with a hundred open questions and nobody free to answer them. In a survey of 1,300 caregivers across the United States, 487 said they used AI chatbots at all. Of those, 25 percent — 121 people — said they used them for caregiving, according to the study by Joan K. Monin and colleagues at Yale, the University of Southern California and the University of Michigan, funded by the National Institute on Aging¹². The data were collected between August and October 2025.

What did they ask the machine? More than half — 58 percent — used it for information about caregiving. About a third, 34 percent, used it for emotional support, and 37 percent reported other uses³. Asked to describe their experience in their own words, 128 people wrote something. The most common theme was seeking informational support, found in 72 of the 128 responses. Another 48 responses described the chatbot as useful, with phrases like "quick and easy"⁴.

Who leaned on it most? Caregivers who scored higher on openness were somewhat more likely to use a chatbot for caregiving. Caregivers with higher cognitive ability and greater internet literacy were significantly less likely to do so. And those who said they felt uncertain about how to care for their person were more likely to use one²⁵⁶⁷.

Three limits matter here. This was one snapshot in time, so it can show associations, not causes — nothing here says a chatbot helped or harmed anyone. "Caregiving use" was whatever each person called caregiving; the authors say they deliberately left that to the respondent's own judgment rather than a narrow definition, and some open-ended answers described work tasks or general questions instead⁸. And the sample is U.S. caregivers only.

The technology is easy to describe and hard to police. The article describes a study in which older chatbots followed scripts: you typed, they matched a canned reply, while today's versions generate answers on the spot, shaped by what you just said, which is what makes them feel personal⁹. That fluency is also the problem. The article describes a study in which reported concerns include inaccurate information from AI "hallucinations," potentially dangerous medical and emotional advice, thin empathy, mishandled crises, and privacy¹⁰. The article describes a study in which the pull is obvious: a chatbot is always available, scales to any number of users, lets you ask anonymously, and is usually free¹¹.

The backdrop is need. The article describes a study in which, in the United States, family members do this work in enormous numbers — an estimated 63 million adults, up from 43.5 million in 2015, and the figure is expected to climb over the next 30 years as the population ages¹². The article describes a study in which what they need is personalized information, training, help balancing care with the rest of life, and support managing emotion, and what they hit instead are distance, unpredictable schedules, cost, stigma and not knowing what exists¹³. General chatbots have risen fast in that gap: The article describes a study in which, among American adults, ChatGPT use went from 18 percent in July 2023 to 34 percent by early 2025¹⁴.

The article describes a study in which research so far has mostly looked elsewhere: at purpose-built caregiving chatbots, at how various chatbots answer caregiving questions, and at small qualitative studies of caregiver feedback¹⁵. The authors say no prior study has examined personality traits alongside AI use for caregiving¹⁶.

Here is how we read it. People defer to whatever sounds confident, and a machine that answers smoothly can take that seat even when nobody vouches for it — especially for someone who is unsure and has no easy way to check. So we would expect the caregivers who feel least certain of their role, and least practiced at navigating other sources, to be the ones who bring the bot into care decisions — and the least positioned to notice when it is wrong. If instead the most uncertain caregivers avoided chatbots for care, or the least digitally skilled turned out to be the most skeptical of the answers, our expectation would fail. What you can do with it: notice when you open a chatbot because you feel unsure rather than because it is the best source. Before acting on anything about a person's health or care, ask who else could confirm it.

We would add a second reading. A program that mirrors your words and asks for more feels like being listened to, and being listened to feels like respect; the study's own numbers suggest that is a small slice next to plain information-seeking. If caregivers who used bots for emotional support described concrete help with their feelings rather than the comfort of being answered, that reading would fail. When a bot's warmth feels like understanding, ask yourself whether you got information or just the feeling of being attended to — and whether a person could give you the second.

And a third, about tools. A tool you master extends what you can do; a tool that masters you quietly sets what you expect of yourself. We would expect caregivers who hand over their care questions to gradually stop building their own judgment, especially when the bot is quick and easy and no one teaches them its limits. If heavy users grew more confident and skilled in their own decisions over time, or kept checking other sources just as much, that would fail. Keep one caregiving question a week that you work out yourself or with a person, so the bot stays something you pick up.

What happens next is not mysterious. The authors call for research that helps caregivers tell which chatbot information is accurate and which needs follow-up from a third party or a professional, such as a doctor or nurse¹⁷. They also want studies that follow people over time to see whether chatbot use for caregiving is tied to well-being, and for whom it helps more or less¹⁸.

So the useful question is not whether chatbots belong in caregiving. It is which questions you hand over, and which you keep.

Which caregiving question will you work out yourself this week?

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

Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "The sample consisted of 1,300 caregivers, with data collected between August and October 2025. We employed a cross-sectional, mixed-methods design, combining quantitative analysis of structured survey items with thematic coding of open-ended responses."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "Of the 487 respondents who reported being AI chatbot users, 25% used the technology for caregiving. Those who were more likely to use AI for caregiving were higher in openness, lower in internet literacy, lower in cognitive ability, helped with more caregiving tasks, and were more uncertain in their caregiving role."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "Over half (58%) used AI chatbots for information about caregiving, 34% for emotional support related to caregiving, and 37% reported other uses."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "Forty-eight (38%) responses referred to how an AI chatbot is useful for caregiving purposes. For example, people described it as “quick and easy” or stated, “Realizing that it is not always 100% correct, I find it very useful and helpful and amazing”"
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "Consistent with the patterns observed for general chatbot use (see S1Table ), a one-standard deviation increase in openness was associated with a 4–5 percentage point increase in the likelihood of using AI chatbots for caregiving support."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "In contrast, caregivers with higher cognitive ability and greater internet literacy were significantly less likely to rely on AI chatbots for caregiving assistance."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "Notably, greater self-reported uncertainty about how to care for the care recipient was associated with a higher likelihood of using AI chatbots for caregiving support."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "The question eliciting this key measure intentionally relied on respondents’ own judgment of what counts as caregiving-related use, rather than imposing a narrow functional criterion."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "AI chatbots are computer programs that engage in human-like conversations with users [ 8 ]. Previously, chatbots were rule-based programs, producing various scripted replies after processing user input. However, recent advancements in AI have resulted in the development of AI-powered chatbots that dynamically and interactively generate responses to user inputs, providing a personalized and context-dependent experience [ 9 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "These include providing inaccurate information due to AI hallucinations, potentially dangerous medical and emotional advice, inadequate empathy, mismanagement of crisis scenarios, and privacy and security issues [ 15 – 19 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "The primary advantages of using AI chatbots rather than or in addition to traditional resources are their constant availability, scalability, anonymity, and personalization capabilities. Additionally, most AI chatbots are inexpensive or entirely free to use [ 12 – 14 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "An estimated 63 million adults in the United States served as caregivers for people living with serious health conditions, up from 43.5 million in 2015 [ 1 ]. These numbers are expected to climb dramatically over the next 30 years due to population aging, as more caregivers need support."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "Physical distance, unpredictable and busy schedules, costs, social stigma, and a lack of awareness regarding available options all contribute to caregivers not accessing informational and emotional support [ 5 – 7 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "From July 2023 to early 2025, the percentage of American adults using ChatGPT increased from 18% to 34%, demonstrating the rapid rise of AI chatbots and, therefore, the importance of understanding their possible societal impacts, from work to social relationships [ 10 , 11 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "Current research on caregivers’ use of AI chatbots is concentrated primarily on the design and assessment of caregiving-specific chatbots, the evaluation of responses given by various chatbots to caregiving-related questions, and small-scale qualitative studies of caregiver perspectives and feedback on chatbot usage [ 12 , 20 , 21 ]."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "No prior study has examined personality traits related to AI use for caregiving purposes. Hence, we investigate heterogeneity in the use of AI chatbots for caregiving purposes not only by demographic and socio-economic characteristics, but also by personality traits."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "Future research could address this by helping caregivers understand which information is accurate and which requires follow-up by a third party or a professional, such as a medical doctor or nurse."
Monin, J. K., Angrisani, M., Dunbar, T. et al. (2026). Prevalence and predictors of artificial intelligence chatbots use for caregiving: Evidence from the Understanding America Study. PLOS One. https://doi.org/10.1371/journal.pone.0357753 - the article this story is about — the whole article — the passage: "Another future direction of this research is to examine whether the use of chatbots for caregiving is associated with improvements in well-being over time as well as to identify individual and contextual factors that predict for whom they are more or less useful in improving well-being."

Do not take this as professional medical advice.

The findings of other studies mentioned here are known to us through this document, which is the one we read; we did not open each of those studies.