Weekly AI Watch

Every week, what changed and what it means for you. No hurry, no alarm.

Week of 13 November 2026 | Past editions15,811 papers found · 14,944 read · 4,785 belonged · 46 stories passed every check

Observatorio de la
Inteligencia Artificial

Cada semana, qué cambió y qué significa para usted. Sin prisa y sin alarma.

Semana del 13 de noviembre de 2026 | Ediciones anteriores15,811 estudios encontrados · 14,944 leídos · 4,785 correspondían · 46 notas pasaron todas las comprobaciones

weeklyAI · Week of 13 November 2026weeklyAI · Semana del 13 de noviembre de 2026

← Your mind← Su mente

experiment · Frontiers in psychology · la publicación, 1 sep 2026 · gratis

Un aviso claro y una advertencia clara cambiaron cómo 768 estudiantes juzgaron a un chatbot de salud mental

Leían una situación inventada y dijeron qué harían; el estudio no siguió a nadie en una conversación real.

Versión breve · la versión detallada sigue, unos 6 min

Pregunte al chat de weeklyAI

Pregúnteme sobre la IA en su vida: le señalo las notas que weeklyAI ha publicado sobre el tema, esta semana y antes.

Las conversaciones se guardan mientras exista weeklyAI, para mejorar la publicación. Se responde en el idioma en que usted escribe.

El estudio, de un vistazo
Quiénes
Estudiantes universitarios
Cuántos
768
Dónde
China
Cuándo
No lo dice el pasaje
Tipo de estudio
experiment
Quién lo hizo
Universidades de China
El límite que importa
Leyeron una situación inventada, no conversaron con un chatbot real; midieron intenciones.

Lo que cambió cada mensaje por separado

Con aviso de privacidadfrente aSin aviso de privacidad

Los estudiantes se sintieron más protegidos al compartir información

Con advertencia de límitesfrente aSin advertencia de límites

Los estudiantes entendieron mejor hasta dónde llega el chatbot

Los dos mensajes juntosfrente aUno solo o ninguno

La confianza que reconoce los límites fue más alta

La próxima vez que abra un chatbot para hablar de cómo se siente, fíjese en dos cosas antes de escribir: si dice con claridad qué hace con sus datos y dónde termina su ayuda.
Lectura de weeklyAI

Abre un chatbot para hablar de estrés, de sueño o de la presión de los estudios, y contesta con soltura, como si conociera a la persona.

Investigadores de universidades chinas querían saber si dos mensajes breves en la pantalla cambian la manera en que los estudiantes juzgan a un chatbot de salud mental. Reunieron a 768 estudiantes universitarios. Cada uno leyó una sola situación inventada, entre cuatro posibles.

Los dos mensajes eran cortos. Uno decía que los datos de salud mental se recogen lo mínimo y se protegen. El otro advertía que el chatbot no sustituye un diagnóstico, una psicoterapia, una intervención en crisis ni un tratamiento médico.

Los estudiantes que vieron la advertencia entendieron mejor los límites de la herramienta. Esa comprensión se relacionó con un menor riesgo de depender del chatbot y con una intención más firme de buscar ayuda profesional. Quienes vieron el aviso de privacidad se sintieron más protegidos al compartir información.

La confianza mejor calibrada, la que reconoce lo que la herramienta puede y no puede hacer, fue más alta cuando aparecieron los dos mensajes juntos. El promedio fue de 3.98 en una escala de 1 a 5, frente a 3.34 sin ningún mensaje.

Los participantes leyeron un texto, no conversaron con un chatbot de verdad. Lo que reportaron fueron sus intenciones, y las intenciones no siempre se convierten en actos. El estudio tampoco midió si alguien pidió ayuda real ni si dejó de apoyarse en la máquina en la vida diaria.

Los 768 estudiantes eran de un solo país, China, y todos estaban inscritos en la universidad. No se sabe si los adultos de otras edades y de otros lugares reaccionarían igual. El equipo probó dos mensajes; nada dice sobre otros avisos que pudieran añadirse.

El estudio probó dos mensajes en una situación inventada, así que no dice qué pasa con el uso diario de estos programas.

La próxima vez que abra un chatbot para hablar de cómo se siente, fíjese en dos cosas antes de escribir: si dice con claridad qué hace con sus datos y dónde termina su ayuda.

Qué significa para usted

Cuando abra un chatbot para hablar de lo que siente, vale la pena mirar si el programa explica qué hace con sus datos y si aclara que no reemplaza a un profesional. En el estudio, esos dos mensajes juntos bastaron para que los estudiantes confiaran con más cuidado. Nadie conversó de verdad con un chatbot, así que conviene esperar antes de sacar conclusiones sobre el uso diario.

Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264

Quién pagó: Los autores declararon que no recibieron apoyo financiero para este trabajo ni para su publicación, y que no tenían relaciones comerciales o financieras que pudieran constituir un conflicto de interés.

No tome esto como consejo médico profesional.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1224 palabras · unos 6 minLeerla →Cerrar

Un aviso de privacidad y un aviso de límites: lo que 768 estudiantes opinaron de un chatbot de salud mental

El estudio no observó conversaciones reales. Midió cómo juzgaron una situación imaginaria.

Imagine que su universidad pone a su disposición un chatbot que responde preguntas sobre estrés, sueño, presión de estudios y a dónde acudir si algo le preocupa. Usted no lo usa: lee una descripción. En la mitad de los casos, esa descripción incluye una frase que explica cómo se cuidan sus datos; en la otra mitad, no. En la mitad, además, incluye una advertencia de que el chatbot no sustituye a un profesional; en la otra mitad, no. Un equipo de investigadores chinos repartió al azar a 768 estudiantes universitarios entre esas cuatro versiones, 192 en cada una, y les preguntó qué opinaban1.

Los mensajes cambiaron lo que los estudiantes percibían. El aviso de privacidad subió la sensación de que sus datos estarían protegidos; la advertencia de límites subió la claridad sobre hasta dónde llega el chatbot2. La confianza mejor medida, es decir, la que reconoce lo que la herramienta hace bien y lo que no puede hacer, fue más alta cuando estaban los dos mensajes juntos que cuando había uno solo o ninguno3. El estudio aparece en Frontiers in Psychology; lo firmaron Zeyu Zhang, Xiaomei Lu, Yinlan Zhang, Hao Zhang y Min Zhang, y los autores declararon no haber recibido financiamiento ni tener conflictos comerciales.

Hay que decir con claridad qué se midió. Los participantes leyeron un escenario inventado y contestaron un cuestionario sobre intenciones y percepciones; no conversaron con un sistema real, y el estudio no probó si alguien buscaría ayuda de verdad ni si dejaría de apoyarse de más en el chatbot4. Tampoco probó nada clínico: no hubo diagnóstico, terapia ni atención de crisis. Y la muestra fueron estudiantes universitarios de un solo país, así que el tamaño de estos efectos en adultos de otras edades y otros lugares no se conoce5.

Piense en lo que pasa cuando alguien escribe a las dos de la mañana que no puede dormir de preocupación. Un amigo puede estar dormido; un profesional puede estar lejos, costar dinero o dar pena. Los propios autores señalan que muchos estudiantes dudan en contar lo que les pasa por estigma, incertidumbre, falta de tiempo o falta de servicios, y que un chatbot se siente disponible, sin juicio y fácil de abordar6. Esa comodidad tiene un filo: los modelos de lenguaje responden con soltura sobre casi cualquier tema, y esa soltura puede tapar la incertidumbre, los datos inventados y la ausencia de responsabilidad profesional7.

El aviso de privacidad, tal como lo define el estudio, no es una promesa vaga: es una frase clara sobre cómo se maneja, se protege y se limita el uso de la información de salud mental8. La advertencia de límites es igual de concreta: dice que el chatbot no sustituye la atención calificada, el diagnóstico, la psicoterapia, la intervención en crisis ni los servicios de emergencia9. Son dos mensajes distintos porque atienden dos riesgos distintos. Uno responde si es razonable contar algo íntimo; el otro aclara hasta dónde llega la ayuda que se puede esperar10.

¿Y qué tan común es ya pedirle apoyo emocional a una máquina? Según el resumen de una encuesta hecha en Líbano, lo único disponible de ese trabajo, el 62.8 por ciento de los participantes había usado chatbots de inteligencia artificial para apoyo de salud mental11; quienes los usaban eran con más frecuencia mujeres, gente joven y personas que reportaban problemas de salud mental12, y los consultaban por ansiedad, ánimo bajo, agotamiento, dolor emocional, problemas de sueño y síntomas de pánico13. Según el resumen de un estudio con dos muestras belgas, lo único disponible de él, los chatbots gratuitos de uso general, sobre todo ChatGPT, eran la opción predominante, elegidos por su disponibilidad constante14, y muchos de esos usuarios ya tenían o habían tenido apoyo profesional15.

Hubo otro intento de medir esto con más realismo. Según el resumen de un experimento con 285 adultos estadounidenses, lo único disponible de ese trabajo, las personas conversaron unos diez minutos con ChatGPT o navegaron en sitios de salud mental que no eran conversacionales16. El estrés bajó en ambos grupos y no hubo diferencia clara entre ellos17. Los autores de ese trabajo advierten que sus hallazgos no hablan de resultados clínicos, de beneficios sostenidos ni del reemplazo de la atención profesional18.

Aquí está la lectura de weeklyAI. La escena de este estudio no es la de alguien hablando con un chatbot: es la de alguien leyendo la letra chica antes de decidir si se acerca. Y lo que se ve es que dos frases colocadas donde se ven, antes de que uno cuente nada, mueven la manera en que la gente juzga la herramienta. Eso tiene una consecuencia práctica: quien diseña estos sistemas puede poner esos avisos a la vista, sin esconderlos en los términos del servicio, y el usuario puede exigirlos ahí.

Concede el estudio algo a favor de los chatbots: mucha gente los usa porque no tiene otra cosa a mano, y eso no es ingenuidad sino falta de opciones. Lo que el estudio no muestra es si el aviso basta cuando la alternativa real sigue estando lejos o es cara. Esta lectura se equivocaría si resultara que quienes tienen menos acceso a servicios son justamente los que más buscan ayuda después de leer el aviso. Con lo que hay, la expectativa razonable es la contraria: el aviso orienta, pero no abre consultorios.

Otros trabajos matizan el asunto. Los propios autores señalan que las herramientas de inteligencia artificial podrían reducir las barreras para buscar ayuda o, al contrario, aumentar la evitación si alguien trata el chatbot como un reemplazo de los servicios profesionales. Son señales de que un aviso amable no siempre gana la partida.

También conviene mirar la palabra de la empresa. Un aviso que suena a fórmula legal, sin decir qué se guarda y quién lo ve, puede no generar confianza real aunque el chatbot hable con calidez. Antes de contarle algo íntimo, fíjese si lo que promete sobre sus datos aparece explicado en algún lado concreto; si solo hay frases bonitas, trátelo como una promesa sin respaldo. Y si usted es de los que saltan esos mensajes, pruebe leer una vez qué se guarda de sus conversaciones y decida con eso, no con la costumbre.

Los autores proponen algo verificable. Dicen que, en lugar de asegurar de forma amplia, las interfaces deberían indicar qué información se recoge, qué no se recoge, si se guarda la conversación, quién puede verla, cuánto tiempo se conserva y cómo se puede detener o borrar una sesión19. Y que un chatbot responsable debería decir desde el inicio que no diagnostica trastornos, no da psicoterapia, no reemplaza a consejeros ni médicos y no atiende emergencias20. Esas son las preguntas que usted puede llevar a cualquier herramienta que le ofrezca su escuela, su trabajo o su servicio de salud.

Vuelva a la escena del principio: usted frente a la pantalla, antes de escribir la primera palabra. Lo que este estudio sugiere es que esas dos frases iniciales no son adorno. La próxima vez que abra un chatbot para hablar de lo que le pesa, busque primero qué dice de sus datos y qué dice de sus límites. Si no lo encuentra, ya tiene su respuesta.

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

  1. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "We conducted a 2 × 2 randomized vignette experiment with 768 college students. Participants were assigned to one of four chatbot scenarios that either included or omitted privacy assurance and professional-boundary warning."
  2. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Privacy assurance increased perceived privacy protection, F (1, 764) = 159.30, p < 0.001, ηp 2 = 0.172, d = 0.91. Professional-boundary warning increased boundary awareness, F (1, 764) = 176.40, p < 0.001, ηp 2 = 0.188, d = 0.96."
  3. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Calibrated trust was highest when both messages were present."
  4. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Because this study used hypothetical vignette scenarios rather than actual chatbot interactions, the findings should be interpreted as evidence about students' safety-related evaluations and intentions, not as evidence of real-world chatbot use or clinical help-seeking behavior."
  5. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The sample consisted of college students in one national context. The findings should be interpreted within the cultural and institutional context in which the study was conducted."
  6. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Students may hesitate to disclose distress to peers, teachers, or professionals because of stigma, uncertainty, time constraints, or lack of access to services. A chatbot can appear available, non-judgmental, and easy to approach."
  7. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Unlike earlier rule-based or scripted conversational agents, large language models can generate fluent and context-sensitive responses across a very wide range of topics. Fluency may increase perceived competence, but it may also obscure uncertainty, hallucinated information, or the lack of professional accountability."
  8. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Privacy assurance refers to a clear statement that informs users how mental health-related information is handled, protected, and limited in use."
  9. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Professional-boundary warning refers to a clear statement that the chatbot is not a substitute for qualified mental health care, clinical diagnosis, psychotherapy, crisis intervention, or emergency services."
  10. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A privacy assurance alone may make the tool easier to approach, but it does not by itself clarify when human care is needed. A boundary warning alone may limit inappropriate reliance, but it does not fully address disclosure concerns."
  11. Boutros T, Fadel DA, Chalhoub G, Ghadban C, Fadel J, Akl C, et al. (2026). Exploring the use of artificial intelligence chatbots for mental health support in Lebanon: a cross-sectional study. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08846-9 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "62.8% of participants reported using artificial intelligence chatbots for mental health support."
  12. Boutros T, Fadel DA, Chalhoub G, Ghadban C, Fadel J, Akl C, et al. (2026). Exploring the use of artificial intelligence chatbots for mental health support in Lebanon: a cross-sectional study. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08846-9 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Users were more frequently females, of younger age, and more likely to report mental health problems."
  13. Boutros T, Fadel DA, Chalhoub G, Ghadban C, Fadel J, Akl C, et al. (2026). Exploring the use of artificial intelligence chatbots for mental health support in Lebanon: a cross-sectional study. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08846-9 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Users commonly consulted chatbots for anxiety, low mood, burnout, emotional pain, sleep problems, and panic symptoms."
  14. Bernaerts S, Buelens F, De Witte NAJ, Balaskas A, Colman T, Van Daele T. (2026). Exploring Real-World Use of AI Chatbots for Mental Health Support: Cross-Sectional Survey Study. JMIR Mental Health. 10.2196/104316 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Freely accessible AI chatbots, mainly ChatGPT, were the predominant choice in both samples and were mainly chosen for their constant availability and accessibility."
  15. Bernaerts S, Buelens F, De Witte NAJ, Balaskas A, Colman T, Van Daele T. (2026). Exploring Real-World Use of AI Chatbots for Mental Health Support: Cross-Sectional Survey Study. JMIR Mental Health. 10.2196/104316 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "To conclude, this study confirms that general-purpose AI chatbots that are not designed or regulated for mental health support, specifically ChatGPT, are being used to obtain social and emotional support, often by individuals already familiar with professional mental health support."
  16. Kim H, Han JY, Zhang W, Yoon HJ, Seo JK. (2026). LLM-Based AI Chatbot Use and Short-Term Issue-Specific Stress: The Role of Perceived Message Contingency. Healthcare. https://doi.org/10.3390/healthcare14223878 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "In a randomized online experiment, 285 U.S. adults identified a personally relevant stressor and conversed with ChatGPT (GPT-3.5) or browsed non-conversational mental health websites for about 10 min."
  17. Kim H, Han JY, Zhang W, Yoon HJ, Seo JK. (2026). LLM-Based AI Chatbot Use and Short-Term Issue-Specific Stress: The Role of Perceived Message Contingency. Healthcare. https://doi.org/10.3390/healthcare14223878 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Stress decreased in both conditions. Contrary to our hypothesis, post-session stress adjusted for pre-session stress did not differ between conditions (adjusted difference = 0.05, 95% CI [−0.16, 0.26]); an exploratory equivalence test supported equivalence within ±0.2 SD."
  18. Kim H, Han JY, Zhang W, Yoon HJ, Seo JK. (2026). LLM-Based AI Chatbot Use and Short-Term Issue-Specific Stress: The Role of Perceived Message Contingency. Healthcare. https://doi.org/10.3390/healthcare14223878 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "The findings do not speak to clinical outcomes, sustained benefits, or the replacement of professional care."
  19. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Rather than using broad assurances, interfaces should state what information is collected, what is not collected, whether conversation content is stored, who can access it, how long it is retained, and how users can stop or delete a session where applicable."
  20. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A responsible chatbot should clearly state at onboarding that it cannot diagnose mental disorders, provide psychotherapy, replace counselors or physicians, or handle emergencies."

Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264

Quién pagó: Los autores declararon que no recibieron apoyo financiero para este trabajo ni para su publicación, y que no tenían relaciones comerciales o financieras que pudieran constituir un conflicto de interés.

No tome esto como consejo médico profesional.

experiment · Frontiers in psychology · the paper, 1 Sep 2026 · free

A Chatbot Can Sound Like It Knows You. Two Plain Messages Changed How Students Judged It.

In a made-up chatbot scenario, 768 college students read one of four versions. A privacy note and a "not a therapist" warning shifted what they said they intended, not what they did.

Short version · the longer version follows, about 5 min

Ask the weeklyAI chat

Ask me about AI in your life: I point you to the stories weeklyAI has published on it, this week and before.

Conversations are saved for as long as weeklyAI exists, to improve the publication. Answers come in the language you write in.

The study at a glance
Who
college students
How many
768
Where
China
When
March 2026 to April 2026
Kind of study
experiment
Who did it
Judicial Police College, Xinjiang University of Political Science and Law, Tumxuk, China
The limit that matters
Students read a made-up scenario, not a real chatbot, and reported intentions.

What each message did on its own

privacy note aloneagainstboundary warning alone

The privacy note made the tool easier to approach but did not by itself clarify when human care is needed; the warning limited inappropriate reliance but did not fully address disclosure concerns.

both messages togetheragainsteither message alone

A balanced view of the bot as useful but limited was highest when both messages were present.

A plain "not a therapist" line may work less by lowering trust than by giving permission to doubt the machine.
weeklyAI's reading

You open a chatbot to talk about stress, or sleep, or the pressure of exams. It answers fluently, in full sentences, and it never seems tired or bored.

Researchers wanted to know whether two short messages at the start of such a chatbot would change how students judged it. They recruited 768 college students and randomly assigned each to read one of four made-up chatbot scenarios. Each person read a single scenario and then answered questions. The students did not talk to a real chatbot, and nobody watched what they did afterward.

The scenarios described a chatbot a university had introduced. It could answer general questions about stress, sleep and study pressure, offer coping suggestions, and point people toward help. In one version, a privacy assurance said the chatbot would collect as little personal information as possible, protect mental health data, and explain how information would be handled. In another, a professional-boundary warning said the chatbot was not a substitute for diagnosis, psychotherapy, crisis intervention or medical treatment. The other versions left these messages out.

Students who saw the privacy note felt their information was better protected. Students who saw the boundary warning better understood the chatbot's limits. That understanding went with a lower stated risk of leaning on the chatbot too much, and with a stronger intention to seek professional help when needed. Calibrated trust, a balanced view of the bot as useful but limited, was highest when both messages appeared together.

The study cannot show that these messages change what people do in a live chat. The students read a made-up scenario and reported what they thought they would do. The sample was college students in one country, so how strong these effects would be for adults elsewhere and of other ages is unknown. The study tested only two messages, so other design features may also shape how students use such a chatbot.

When you next open a mental health chatbot, look for whether it says plainly what it does with your data and where its help stops.

What this means for you

For now, treat those two messages as a small test you can run on yourself, watching whether a chatbot's plain words about your data and its limits change how much you lean on it. What the students said they would do is not what anyone did.

Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264

Who paid: The authors declared that no financial support was received for this work or its publication, and they declared no commercial or financial relationships that could be a conflict of interest.

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 · 989 words · about 5 minRead it →Close

A Privacy Note and a "Not a Therapist" Line Changed How 768 Students Judged a Chatbot

In a made-up chatbot scenario, two plain interface messages shifted students' trust and their stated intentions. Nobody was observed in a real conversation.

A university announces a chatbot for stress, sleep and study pressure. Before a student types anything, two short lines appear. One says what happens to the information. The other says the bot is not a therapist, not a diagnosis, not emergency care. That is the intervention tested in a new study: a written scenario describing the chatbot, with a privacy statement added in some versions and a boundary warning added in others.

The researchers, led by Zeyu Zhang of the Judicial Police College at Xinjiang University of Political Science and Law in Tumxuk, China, ran a randomized experiment with 768 college students. Each was assigned to one of four versions of the same written scenario: no messages, privacy note only, boundary warning only, or both1. The study declares no financial support and no commercial conflicts.

What the students saw was a description, not a working product. Generative AI chatbots are not scripted reply trees; large language models produce fluent, context-sensitive text across a very wide range of topics, and that fluency can raise perceived competence while hiding uncertainty, invented information, or the absence of professional accountability2. A privacy assurance, in the study's terms, is a clear statement of how mental health-related information is handled, protected, and limited in use3. A professional-boundary warning is a clear statement that the chatbot is not a substitute for qualified mental health care, clinical diagnosis, psychotherapy, crisis intervention, or emergency services4. Picture the two lines on a phone screen before the first message: one answers "is it safe to type this," the other answers "how far can this go."

The appeal of such tools is easy to picture. Students may hesitate to disclose distress to peers, teachers, or professionals because of stigma, uncertainty, time constraints, or lack of access to services, and a chatbot can appear available, non-judgmental, and easy to approach5. Use is not hypothetical: in a survey of 349 people in two Belgian samples, general-purpose chatbots, mainly ChatGPT, were the predominant choice for mental health conversations, picked for their constant availability and accessibility6, and most respondents were women who had received or were receiving professional mental health support7. In a Lebanese survey, 62.8% of participants reported using AI chatbots for mental health support8, commonly for anxiety, low mood, burnout, emotional pain, sleep problems and panic symptoms9. Both of those are abstracts only; the full papers could not be fetched.

The students noticed the difference. Those who read the privacy line rated the chatbot's privacy protection higher; those who read the boundary line showed greater awareness of its limits10. Calibrated trust, a balanced view of the bot as useful but limited, was highest when both messages were present11. In the path model, boundary awareness went with lower overreliance risk and stronger intention to seek professional help12, and calibrated trust went with safer-use intention13. The authors are careful about the pairing: a privacy assurance alone may make the tool easier to approach without clarifying when human care is needed, and a boundary warning alone may limit inappropriate reliance without addressing disclosure concerns14.

Consider what this study did not do. Participants read a hypothetical scenario rather than interacting with a chatbot, so the findings speak to evaluations and intentions, not real-world chatbot use or clinical help-seeking behavior15. The outcomes were self-reported intentions and risk perceptions, and the sample was college students in one national context16. The study provided no clinical support and did not test crisis intervention. A separate randomized experiment with 285 U.S. adults, available only as an abstract, found stress fell after about ten minutes with ChatGPT or with ordinary mental health websites, with no clear difference between the two17, and its findings do not speak to clinical outcomes, sustained benefits, or the replacement of professional care18. A Belgian survey, abstract only, notes that knowledge of how, why, and which individuals use chatbots for mental health is limited19.

Here is weeklyAI's reading. A confident voice tends to collect trust, and people keep following confident advice unless something openly invites them to question it. A plain "not a therapist" line may work less by lowering trust than by giving permission to doubt the machine. That leads weeklyAI to expect the warning to matter most at the moment a chat turns serious, when the missing human care becomes visible. This reading would be wrong if students who saw the warning were just as ready to follow chatbot advice unchecked as those who did not. Those are about carrying over to other settings, not about this result.

What would have to happen next is concrete. The authors argue interfaces should state what information is collected, what is not, whether conversations are stored, who can access them, how long they are kept, and how a session can be stopped or deleted20. They argue a responsible chatbot should say at onboarding that it cannot diagnose mental disorders, provide psychotherapy, replace counselors or physicians, or handle emergencies21. A field test would mean watching what people actually do in a live chat, including the moment distress escalates.

The authors also note that AI can improve access, personalization, and communication efficiency while producing privacy concerns and overreliance when users treat automated systems as primary sources of support; that is the trade-off this study sits inside. For a reader in Latin America or North America, the useful move is to look at the first screen of any mental health chatbot and check whether both lines are there, in plain words, before typing. If the bot opens with confident expertise and no limits, that is the signal to slow down.

The next time a chatbot asks how you are sleeping, read the first screen before you answer. What does it say it cannot do?

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

  1. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "We conducted a 2 × 2 randomized vignette experiment with 768 college students. Participants were assigned to one of four chatbot scenarios that either included or omitted privacy assurance and professional-boundary warning."
  2. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "Unlike earlier rule-based or scripted conversational agents, large language models can generate fluent and context-sensitive responses across a very wide range of topics. Fluency may increase perceived competence, but it may also obscure uncertainty, hallucinated information, or the lack of professional accountability."
  3. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "Privacy assurance refers to a clear statement that informs users how mental health-related information is handled, protected, and limited in use."
  4. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "Professional-boundary warning refers to a clear statement that the chatbot is not a substitute for qualified mental health care, clinical diagnosis, psychotherapy, crisis intervention, or emergency services."
  5. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "Students may hesitate to disclose distress to peers, teachers, or professionals because of stigma, uncertainty, time constraints, or lack of access to services. A chatbot can appear available, non-judgmental, and easy to approach."
  6. Bernaerts S, Buelens F, De Witte NAJ, Balaskas A, Colman T, Van Daele T. (2026). Exploring Real-World Use of AI Chatbots for Mental Health Support: Cross-Sectional Survey Study. JMIR Mental Health. 10.2196/104316 — only the abstract - the full text could not be fetched — the passage: "Freely accessible AI chatbots, mainly ChatGPT, were the predominant choice in both samples and were mainly chosen for their constant availability and accessibility."
  7. Bernaerts S, Buelens F, De Witte NAJ, Balaskas A, Colman T, Van Daele T. (2026). Exploring Real-World Use of AI Chatbots for Mental Health Support: Cross-Sectional Survey Study. JMIR Mental Health. 10.2196/104316 — only the abstract - the full text could not be fetched — the passage: "Most respondents in the present samples were women and indicated having received or receiving professional mental health support."
  8. Boutros T, Fadel DA, Chalhoub G, Ghadban C, Fadel J, Akl C, et al. (2026). Exploring the use of artificial intelligence chatbots for mental health support in Lebanon: a cross-sectional study. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08846-9 — only the abstract - the full text could not be fetched — the passage: "62.8% of participants reported using artificial intelligence chatbots for mental health support."
  9. Boutros T, Fadel DA, Chalhoub G, Ghadban C, Fadel J, Akl C, et al. (2026). Exploring the use of artificial intelligence chatbots for mental health support in Lebanon: a cross-sectional study. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08846-9 — only the abstract - the full text could not be fetched — the passage: "Users commonly consulted chatbots for anxiety, low mood, burnout, emotional pain, sleep problems, and panic symptoms."
  10. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "Privacy assurance increased perceived privacy protection, F (1, 764) = 159.30, p < 0.001, ηp 2 = 0.172, d = 0.91. Professional-boundary warning increased boundary awareness, F (1, 764) = 176.40, p < 0.001, ηp 2 = 0.188, d = 0.96."
  11. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "Calibrated trust was highest when both messages were present."
  12. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "Boundary awareness was associated with calibrated trust, β = 0.25, p < 0.001, lower overreliance risk, β = −0.35, p < 0.001, and stronger professional help-seeking intention, β = 0.35, p < 0.001."
  13. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "Calibrated trust was associated with safe-use intention, β = 0.42, p < 0.001, lower overreliance risk, β = −0.14, p = 0.006, and higher professional help-seeking intention, β = 0.18, p < 0.001."
  14. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "A privacy assurance alone may make the tool easier to approach, but it does not by itself clarify when human care is needed. A boundary warning alone may limit inappropriate reliance, but it does not fully address disclosure concerns."
  15. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "Because this study used hypothetical vignette scenarios rather than actual chatbot interactions, the findings should be interpreted as evidence about students' safety-related evaluations and intentions, not as evidence of real-world chatbot use or clinical help-seeking behavior."
  16. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "The sample consisted of college students in one national context. The findings should be interpreted within the cultural and institutional context in which the study was conducted."
  17. Kim H, Han JY, Zhang W, Yoon HJ, Seo JK. (2026). LLM-Based AI Chatbot Use and Short-Term Issue-Specific Stress: The Role of Perceived Message Contingency. Healthcare. https://doi.org/10.3390/healthcare14223878 — only the abstract - the full text could not be fetched — the passage: "Stress decreased in both conditions. Contrary to our hypothesis, post-session stress adjusted for pre-session stress did not differ between conditions (adjusted difference = 0.05, 95% CI [−0.16, 0.26]); an exploratory equivalence test supported equivalence within ±0.2 SD."
  18. Kim H, Han JY, Zhang W, Yoon HJ, Seo JK. (2026). LLM-Based AI Chatbot Use and Short-Term Issue-Specific Stress: The Role of Perceived Message Contingency. Healthcare. https://doi.org/10.3390/healthcare14223878 — only the abstract - the full text could not be fetched — the passage: "The findings do not speak to clinical outcomes, sustained benefits, or the replacement of professional care."
  19. Bernaerts S, Buelens F, De Witte NAJ, Balaskas A, Colman T, Van Daele T. (2026). Exploring Real-World Use of AI Chatbots for Mental Health Support: Cross-Sectional Survey Study. JMIR Mental Health. 10.2196/104316 — only the abstract - the full text could not be fetched — the passage: "First studies indicate a growing prevalence of individuals talking to AI chatbots about mental health-related topics; yet, knowledge of how, why, and which individuals are using these AI chatbots for their mental health is limited."
  20. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "Rather than using broad assurances, interfaces should state what information is collected, what is not collected, whether conversation content is stored, who can access it, how long it is retained, and how users can stop or delete a session where applicable."
  21. Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264 - the article this story is about — the whole article — the passage: "A responsible chatbot should clearly state at onboarding that it cannot diagnose mental disorders, provide psychotherapy, replace counselors or physicians, or handle emergencies."

Zhang, Z., Lu, X., Zhang, Y. et al. (2026). Privacy assurances and professional-boundary warnings in generative AI mental health chatbots: a randomized vignette experiment on calibrated trust, overreliance risk, and professional help-seeking intentions. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2026.1934264

Who paid: The authors declared that no financial support was received for this work or its publication, and they declared no commercial or financial relationships that could be a conflict of interest.

Do not take this as professional medical advice.

The free weekly drip by email

One email a week: the free story of every section, with its picture, and the headlines of the rest.

No payment, nothing to cancel. Unsubscribe from any issue.

El goteo semanal gratis por correo

Un correo por semana: la nota gratuita de cada sección, con su imagen, y los titulares de las demás.

Sin pago y sin nada que cancelar. Puede darse de baja desde cualquier envío.

From the house

About

Every week the new research on AI is read here, and this page tells you what was found, how far it can be trusted, and what you can do with it. This page is free, and every line on it says what was found.

The studies are read by a machine built and ruled by a professor of communication at Sacramento State and Universidad Nacional Autónoma de México (UNAM). Who makes this →

Diego Bonilla, Ph.D.

Full Professor of Communication Studies, Sacramento State, California, United States
Graduate instructor, Universidad Nacional Autónoma de México (UNAM), Mexico

hypergraphia.com

I am a professor of communication at Sacramento State, where I created the course Information Management and Privacy and have taught it many times, and at UNAM, in Mexico City, I teach the graduate course Electronic and Generative Literature with AI. In both I teach AI. I have taught since 1989, at first mathematics and physics to children who found them hard, and since 1997 one subject that has changed its name many times: media literacy, computer literacy, digital literacy, information literacy, ICT literacy, and now AI literacy.

This page exists because most people cannot go back to a university, and these systems are changing their work, their children's schooling, their parents' care and their vote all the same. What a university teaches about them should reach the people who need it, every week, in plain language, without enrolling anywhere. That is what I do here, and it is what I do on Thursday mornings in a seminar for the Renaissance Society at Sacramento State, built on this publication, where anyone can send a question and hear it answered.

In 2003 my doctorate at Syracuse measured what a machine can infer about a person from the traces they leave, and what I found became the privacy course I still teach. In 2007 and 2009 I sat on the committee at ETS in Princeton that set the national standard for digital literacy, and from 2014 to 2016 I was the California State University's ICT Literacy Ambassador. Six names, one job: helping people not be at the mercy of the newest machine.

I have also worked on the other side of the counter. In 2000 I began making films that a machine assembles differently for every viewer; the first of them took first prize at the Moscow International Film Festival in 2004, two decades before the industry shipped a generative model. A generative video poem I made with a poet at UNAM is in the field's definitive anthology, and in 2025 my AI short film The Human Futures Market took Best AI Short Film at the Red Rocks AI Film Festival. My recent published research is on AI and elections, and on how these models represent the cultures they were not trained on. I do not read the papers myself, and this page will never tell you I do. I built the machine that reads them and wrote the rules it reads by, and I analyze what it gives me the way I have taught my students to analyze: asking how it knows, and who benefits from what it says.

The vendors began teaching AI literacy two years ago, because before that there was nothing to sell. I am not against them. I am simply not one of them, and no one who sells these systems pays for this page. Your subscription does, and that is what keeps it honest.

The record, dated
  • 1989 Teaching begins, Mexico City
  • 1995 Hypergraphia, an early Spanish-language web and hypermedia publisher
  • 1996 Mercadotecnia e imagen en Internet, with Jesús del Toro, a book on marketing on the Internet, among the first in Spanish, Editorial Iberoamérica
  • 1999 Autorretrato, first generative poem, published by UNAM's Instituto de Investigaciones Filológicas
  • 2003 Ph.D., Syracuse University, Doctoral Prize, for the dissertation The Medium Is the Measure of Itself
  • 2004 A Space of Time: First Prize, Internet/Multimedia, XXVI Moscow International Film Festival
  • 2007, 2009 ETS iSkills committee, Princeton
  • 2008 Making Sense of Tracking Data, a book on how the traces a person leaves online allow inferences about them, including how they think, VDM Verlag
  • 2011 Keynote on generative work, UNAM
  • 2014–2016 ICT Literacy Ambassador, California State University · California Open Educational Resources Council
  • 2019 Big Data, with Rodolfo Mata: Electronic Literature Collection, Volume 4
  • 2020 Tablada Hipertextual, with Rodolfo Mata, UNAM: the computer reads a poet's words and builds a different reading from them, the ground the language models now work on
  • 2021 Uku Pacha, a generative VR film: world premiere, Vancouver International Film Festival
  • 2023 Keynote on the representation of Latin American cultures in AI, Ibero-American Meeting of Audiovisual Narratives, Trujillo, Peru · Graduate teaching at UNAM begins
  • 2023–2026 Directed the faculty learning community Advancing AI Literacy, Sacramento State · Principles for Cognitive and Pedagogical Sovereignty
  • 2024 A weekly AI publication run on real pipelines, every week since
  • 2025 Best AI Short Film, Red Rocks AI Film Festival
  • 2026 Keynote, with Nancy Huante-Tzintzun, Latin American and Caribbean Digital Humanities symposium, UT San Antonio · AI Watch, a weekly seminar for the Renaissance Society at Sacramento State

The research record →

TertulIA: weeklyAI.watch/tertulia/en/Pictures are illustrations generated by weeklyAI.watch, never photographs