Weekly AI Watch

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

Week of 6 November 2026 | Past editions15,706 papers found · 14,822 read · 4,771 belonged · 43 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 6 de noviembre de 2026 | Ediciones anteriores15,706 estudios encontrados · 14,822 leídos · 4,771 correspondían · 43 notas pasaron todas las comprobaciones

weeklyAI · Week of 6 November 2026weeklyAI · Semana del 6 de noviembre de 2026

← The good news← Las buenas noticias

experiment · PLOS digital health · la publicación, 25 jun 2026 · gratis

Un chatbot repitió el cuestionario de depresión punto por punto; nadie comprobó si el resultado era correcto

En 132 adultos del Reino Unido y China, las respuestas del chatbot coincidieron con las del formulario en casi todos los casos. El estudio no comparó ninguna de las dos puntuaciones con una entrevista hecha por un profesional de salud mental.

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
Adultos de 18 a 70 años que probaron un chatbot de depresión
Cuántos
132 adultos
Dónde
Reino Unido y China
Cuándo
marzo y abril de 2025
Tipo de estudio
experiment
Quién lo hizo
University College London y Lancashire & South Cumbria NHS Foundation Trust
El límite que importa
Nadie comparó los puntajes con una entrevista hecha por un profesional de salud mental.
Qué prefirieron los participantes: el chatbot o el formulario
Prefirió el chatbot69.7%
Dispuesto a usarlo o recomendarlo87.1%

Porcentajes de los 132 participantes del estudio; no son porcentajes de todas las personas con depresión, y el estudio no comparó los puntajes con una entrevista de un profesional.

Preferencia por el chatbot frente al formulario: este estudio y uno anterior

HopeBot, este estudiofrente aMarcus, sistema anterior

Aquí 69.7% prefirió el chatbot; en Marcus solo 18.1%

Se trata de dos maneras de hacer las mismas preguntas que coinciden entre sí, no de un diagnóstico.
Lectura de weeklyAI

"Me dio una guía detallada" y "ejemplos que aclararon mis emociones". Así explicó un participante por qué confiaba más en la puntuación del chatbot que en la suya. Fue la razón más repetida: 33 menciones.

Detrás de esa frase hay un patrón. De las 75 personas que compararon los dos formatos, 53 confiaron más en el resultado del chatbot. Y 92 de las 132 prefirieron el chatbot en general.

Un equipo de investigadores del University College London y del Lancashire & South Cumbria NHS Foundation Trust reunió a 132 adultos en el Reino Unido y China. Primero cada uno completó el formulario estándar de nueve preguntas sobre depresión, el PHQ-9. Después respondió las mismas preguntas conversando con HopeBot, un chatbot construido sobre GPT-4o y apoyado en transcripciones de terapia y en directorios de líneas de ayuda del Reino Unido y China.

Las puntuaciones coincidieron estrechamente. La medida de acuerdo entre ambos formatos fue de 0.92, y la diferencia típica fue de un punto. El chatbot reprodujo la forma de puntuar del formulario, no inventó la suya.

Lo que el estudio no puede mostrar es si las puntuaciones eran acertadas. Para saberlo haría falta compararlas con una entrevista hecha por un profesional de salud mental. Se trata de dos maneras de hacer las mismas preguntas que coinciden entre sí, no de un diagnóstico.

Las personas estudiadas eran en su mayoría jóvenes, con estudios universitarios y acostumbradas a la inteligencia artificial: 64.4% la usaba con regularidad. Es posible que personas mayores o con menos práctica digital se sientan menos cómodas con una conversación así.

En pruebas internas de los propios creadores, una conversación completa con el chatbot duró unos 25 minutos. El artículo cita otro estudio en el que el formulario tomó 2 o 3 minutos. Muchos describieron los consejos del chatbot como generales o poco profundos. Es una experiencia más larga y conversada, no una vía rápida ni un reemplazo de hablar con un profesional.

Si una clínica o una universidad cerca de usted ofrece una entrevista conversacional para tamizaje de depresión, pregunte si sigue el cuestionario validado. Pregunte también quién lo evaluó y contra qué se comparó.

Qué significa para usted

Si en su clínica o en su escuela le ofrecen una conversación con un programa para responder el cuestionario de depresión, pregunte si usa las mismas nueve preguntas y quién revisó el sistema. Ningún estudio ha mostrado todavía que este chatbot haga un diagnóstico correcto.

Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446

Quién pagó: El trabajo fue financiado por el Centro de Formación Doctoral del UKRI en sistemas sanitarios habilitados por IA (EP/S021612/1), fondos internos del Institute of Health Informatics del UCL y el Biomedical Research Centre del Hospital Moorfields (NIHR203322), todos para K. Li; los financiadores no tuvieron ningún papel en el diseño del estudio, la recogida y el análisis de datos, la decisión de publicar ni la preparación del manuscrito.

No tome esto como consejo médico profesional.

Los hallazgos de otros estudios que aquí se mencionan provienen de este documento; esos estudios no se consultaron uno por uno.

Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1211 palabras · unos 6 minLeerla →Cerrar

Un chatbot hizo las nueve preguntas de la depresión y puntuó casi igual que el cuestionario que uno llena solo

En 132 adultos del Reino Unido y China, las cuentas salieron parecidas. Nadie comprobó con un clínico si alguno tenía depresión.

Nueve preguntas sobre las últimas dos semanas: el sueño, el ánimo, las ganas de hacer las cosas. Es el Cuestionario de Salud del Paciente, o PHQ-9, el formato más usado para detectar señales de depresión. Un grupo de investigadores del Instituto de Informática de la Salud del University College de Londres construyó HopeBot, un programa que hace esas mismas nueve preguntas conversando, por texto o por voz, y que puntúa las respuestas.

La idea de fondo es sencilla. El formulario impreso no pregunta nada más: si alguien duda entre dos casillas, lo resuelve solo. HopeBot sí pregunta. Cuando una respuesta es vaga, como "tal vez a veces", pide aclaración y ofrece las opciones estándar. Cuando aparecen señales de angustia, responde con apoyo. Todo lo que dice está anclado en materiales revisados: transcripciones de terapia cognitivo-conductual, guías, directorios de líneas de ayuda del Reino Unido y de China.

Para probarlo, 132 adultos de entre 18 y 70 años, 68 reclutados en el Reino Unido y 64 en China, contestaron primero el formulario por su cuenta y luego conversaron con HopeBot. La mayoría tenía menos de 45 años, poco más de la mitad eran mujeres y casi nueve de cada diez tenían título universitario. Los puntajes totales coincidieron exactamente en 59 personas, es decir, en 44.7% del grupo. La diferencia típica fue de un punto. No hubo señal de que el chatbot inflara ni redujera la gravedad.

De los 75 participantes que completaron el módulo comparativo, 53, o sea 70.7%, expresaron más confianza en el resultado del chatbot. De esos 75, 20 dieron su opinión pese a haber obtenido el mismo puntaje en ambos formatos. Citaron la estructura más clara, los ejemplos que les ayudaron a entender qué sentían y el tono amable. En una escala de 0 a 10, la comodidad para expresarse sin sentirse juzgados recibió 8.4; el manejo de temas delicados, 7.6. Casi todos, 87.1%, dijeron que lo usarían o lo recomendarían.

Hay un límite que conviene tener enfrente. Ningún clínico revisó ninguno de los dos puntajes. El estudio comparó el chatbot con el formulario, no con una entrevista hecha por un profesional. Reproduce la forma de medir, no demuestra que detecte bien la depresión. Un puntaje alto o bajo de HopeBot no es un diagnóstico.

Piense en lo que cubre una consulta: alguien que escucha, que repregunta, que nota el tono y el cansancio. El formulario, en cambio, se contesta en dos o tres minutos, solo, sin nadie enfrente. El artículo recoge un estudio en el que la Organización Mundial de la Salud calcula que la depresión afecta a cerca del 3.8% de la población mundial1. El artículo recoge un estudio en el que apenas la mitad de quienes la tienen recibe un tratamiento mínimamente adecuado2. El artículo recoge un estudio que señala que la depresión puede agravar los síntomas y aumentar el riesgo de suicidio3. Nada de esto lo resuelve un chatbot.

Lo curioso es que la conversación con HopeBot duró, en pruebas internas del equipo, un promedio de 25 minutos y 23 segundos4. El formulario, dos o tres minutos. Ese tiempo extra es justamente lo que se gana y lo que se paga: más espacio para entender la pregunta, menos capacidad de atender a mucha gente rápido. No es una herramienta para una campaña masiva en una clínica saturada.

Antes hubo otros intentos. El artículo recoge varios estudios en los que se ha probado la detección por chatbot con cuestionarios estándar5. La diferencia que los autores reclaman es la conversación abierta: los anteriores dependían de respuestas predefinidas y no dejaban explayarse. El contraste más visible está en la preferencia: el artículo recoge un estudio de uno de esos sistemas anteriores en el que solo 18.1% de los usuarios prefirió el chatbot al formulario; aquí fue 69.7%6.

Así lo leemos nosotros. La gente desconfía de los trámites lentos y opacos, y en cambio se fía de lo que responde rápido, escucha y explica. Es probable que buena parte del gusto por HopeBot venga de eso: de que aclara las preguntas y acompaña, no de que mida mejor el ánimo. Si quienes lo prefirieron hubieran dado razones de exactitud y no de claridad o trato, la lectura sería otra. Si alguien cercano prefiere el chatbot al formulario, vale la pena preguntarle qué le gustó: el trato o la rapidez. Y pedir que cualquier resultado se revise con un profesional antes de sacar conclusiones.

La misma uniformidad que hace confiable a una máquina también la vuelve fría. Algunos participantes describieron la interacción como plana o impersonal, con respuestas genéricas y una voz mecánica. Quienes ya habían tenido atención psicológica calificaron las recomendaciones como menos útiles que quienes nunca la habían buscado. Tiene sentido: quien conoce la terapia compara contra un estándar más alto.

Hay otra cosa que el estudio no puede sostener. La muestra era joven, con estudios y ya acostumbrada a la inteligencia artificial: 64.4% se describió como usuario habitual de estos programas. Los propios autores advierten que eso puede exagerar la facilidad de uso en personas mayores o con menos acceso digital7. El estudio tampoco se diseñó para comparar países ni culturas; los análisis por país fueron exploratorios. Y todo se probó con un solo modelo, GPT-4o, con instrucciones fijas entre marzo y abril de 2025.

Lo que faltaría es claro. Los autores piden ensayos en varios sitios que comparen el chatbot, el formulario y la entrevista con un clínico, para ver si de verdad detecta bien la depresión y cómo encaja en el trabajo diario de una clínica89. Hasta entonces, esto es una prueba de que la conversación no rompe las cuentas del cuestionario. Nada más.

Aquí está lo que nos deja pensando. Si una máquina puede hacer las nueve preguntas con un tono amable, la puerta se abre para quien nunca se sentaría a llenar un formulario. También se abre el riesgo de que alguien confunda una conversación agradable con una evaluación médica. Nosotros esperaríamos que las primeras versiones lleguen a servicios donde ya hay un profesional detrás, que reciba el resultado. Sabríamos que nos equivocamos si el chatbot empezara a usarse solo, sin nadie que revise. Lo que usted puede hacer es sencillo: si lo prueba, no lo tome como veredicto, y si le preocupa cómo se siente, llévelo a una consulta.

Una cosa más que vale la pena saber: en otros estudios que hemos leído, buena parte de quienes ya conversan con estos programas sobre su ánimo usan chatbots generales, no diseñados para salud mental, sobre todo por su disponibilidad constante. Ninguno de esos resultados mide a HopeBot. Son la familia de la que este estudio forma parte.

Usted ya sabe lo que hace el formulario de nueve preguntas: se contesta rápido y no pregunta de vuelta. Ahora sabe que una versión que conversa puede llegar al mismo número, que tarda mucho más y que nadie ha comprobado todavía si ese número significa lo que creemos. La pregunta que puede llevar a su próxima consulta es esta: ¿este resultado me lo revisó alguien?

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

  1. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The World Health Organisation (WHO) estimates that depression affects approximately 3.8% of the global population [ 1 ]."
  2. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Yet only about half receive minimally adequate counselling or antidepressant treatment [ 6 ]."
  3. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Delayed identification of depression can exacerbate symptoms, increasing risks for chronic disability and suicide, with over 700,000 individuals dying by suicide annually due to depression [ 1 ]."
  4. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Across ten internal test sessions, a complete HopeBot interaction, including open-ended dialogue and PHQ-9 administration, lasted a mean of 25 minutes and 23 seconds, with duration varying according to the extent of user self-disclosure and conversational depth."
  5. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Several prior chatbot-based depression screening systems, such as DEPRA [ 21 ], IGOR [ 26 ], Perla [ 18 ], Marcus [ 27 ], and EmoScan [ 28 ] have demonstrated initial feasibility using structured frameworks and standardised assessments (e.g., PHQ-9, SIGH-D, IDS-C)."
  6. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - el artículo del que trata esta nota — el artículo completo — el pasaje: "In contrast to Marcus, where only 18.1% of users preferred the chatbot over conventional self-report [ 27 ], 69.7% (n = 92) of participants in this study favoured HopeBot, and 87.1% (n = 115) expressed willingness to reuse or recommend it."
  7. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Third, the sample skewed younger and digitally literate, which may overestimate usability in older or digitally excluded groups."
  8. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Multisite and randomised trials comparing LLM-assisted, self-administered, and clinician-administered PHQ-9 formats would clarify diagnostic performance and workflow impact."
  9. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Further validation against clinician-administered assessments in real-world workflows is warranted."

Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446

Quién pagó: El trabajo fue financiado por el Centro de Formación Doctoral del UKRI en sistemas sanitarios habilitados por IA (EP/S021612/1), fondos internos del Institute of Health Informatics del UCL y el Biomedical Research Centre del Hospital Moorfields (NIHR203322), todos para K. Li; los financiadores no tuvieron ningún papel en el diseño del estudio, la recogida y el análisis de datos, la decisión de publicar ni la preparación del manuscrito.

No tome esto como consejo médico profesional.

Los hallazgos de otros estudios que aquí se mencionan provienen de este documento; esos estudios no se consultaron uno por uno.

experiment · PLOS digital health · the paper, 25 Jun 2026 · free

HopeBot's Scores and the Paper Test's Scores Differed by One Point in the Middle

A study of 132 adults in the UK and China found a talking chatbot reproduced the paper questionnaire's scoring; no clinician checked either score, so it does not show the chatbot spots depression.

Short version · the longer version follows, about 7 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
adults in the UK and China
How many
132
Where
UK and China
When
March–April 2025
Kind of study
survey
Who did it
University College London and linked NHS trusts
The limit that matters
No clinician checked either score, so it does not show the chatbot spots depression.
How people rated HopeBot out of 10
Comfort8.4out of 10
Voice clarity7.7out of 10
Handling sensitive topics7.6out of 10
Helpfulness of recommendations7.4out of 10

Average ratings from 132 participants on a 0–10 scale; higher is better, but these are opinions, not proof the scores were right.

Preference for the chatbot over the paper form

Marcus studyagainstThis study

18.1% preferred the chatbot in Marcus, versus 69.7% here

A high or low chatbot score is not a diagnosis, and this was a comparison of two ways of asking the same questions.
weeklyAI's reading
How it could look · illustration generated by weeklyAI.watch, not a photograph

"Detailed guidance," one participant wrote, and "examples that clarified my emotions." Among the people who trusted the chatbot's score more, the most common reason was that it gave detailed guidance and examples that clarified their emotions.

Behind that sentence sits a pattern. Of the 75 people asked which of the two results they trusted more, 53 trusted the chatbot's. Of all 132, 92 said they preferred the chatbot to filling in the form alone.

Researchers at University College London and linked NHS trusts ran the study. They took 132 adults in the UK and China, most under 45, and had each of them fill in the nine-question PHQ-9, a form used to screen for depression. Then each went through the same nine questions with HopeBot, a chatbot built on GPT-4o and anchored in therapy session transcripts and real helpline numbers for both countries.

The two sets of scores matched closely: a statistical measure of agreement came out at 0.92, where 1 is perfect, and the typical gap between the two scores was one point. The chatbot was reproducing the form's scoring, not inventing its own arithmetic.

Picture a person given the form alone with no one to ask. About a third of the participants wanted help understanding a question or how it was scored, and nearly all of those, 41 of 44, said the chatbot's explanation helped.

What the study cannot show is that the scores were right. No clinician checked either one. A high or low chatbot score is not a diagnosis, and this was a comparison of two ways of asking the same questions.

Nor are these readers' neighbours. Most held a degree, 64.4 percent used AI tools regularly, and only 2 of 132 had never tried one. The sample was younger and more digitally literate than the wider public, so the study may overestimate how easy the chatbot is for older people or those with less digital access.

Expect it to be slower, too. The article describes a study in which the chatbot conversation ran about 25 minutes against 2 to 3 minutes for the form. Some participants in this study found its advice generic or its voice mechanical.

What to look for where you live: a clinic, university or helpline offering a conversational screening tool that says plainly it follows the validated questionnaire. Then ask who checked it, and against what.

What this means for you

What this means is that a screening tool that talks back could one day reach people who struggle with a paper form, but nobody has checked whether its scores match a professional's judgment, and the people tested were mostly young and already comfortable with AI. If a clinic near you offers something like this, ask who compared it against a trained interviewer, and treat any score it gives as a prompt to talk to a professional, not an answer.

Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446

Who paid: The work was supported by the UKRI Centre for Doctoral Training in AI-enabled healthcare systems (EP/S021612/1), UCL Institute of Health Informatics internal project funding, and the Moorfields Hospital Biomedical Research Centre (NIHR203322), all to K. Li; the funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Do not take this as professional medical advice.

The findings of other studies mentioned here come from this document; those studies were not opened one by one.

The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 1346 words · about 7 minRead it →Close

A Chatbot Asked the Nine Depression Questions. It Scored Like the Paper Form.

A study of 132 adults in Britain and China compared a talking questionnaire with the usual self-administered one. No clinician checked either score.

Consider what happens when a person sits down with the Patient Health Questionnaire-9, the nine-item form doctors use to screen for depression. On paper, nobody is there. A person reads a question, picks a number from zero to three, and moves on. The form takes two or three minutes. It also lets a vague answer stay vague.

HopeBot is a chatbot built on GPT-4o, a large language model. It asks the same nine questions, by typing or by voice, and when an answer is unclear it asks a gentle follow-up. When signs of distress appear, it offers supportive words drawn from a store of vetted material rather than invented on the spot12. The nine questions, their order and their scoring stay fixed; only the conversation around them is free.

The study recruited 132 adults aged 18 to 70 in Britain and China, 68 in Britain and 64 in China. Most were under 45, just over half were women, and nearly nine in ten held a university degree. Each person filled in the standard form first, then talked to the chatbot in the same session.

The two scores agreed closely. In 59 people, 44.7 percent, the totals were identical. The typical gap was one point on a scale that runs to 273. The headline number the authors report for agreement is 0.92, where 1.0 would mean perfect correspondence3.

Of the 75 people asked which result they trusted more, 53, or 71 percent, chose the chatbot's, citing clearer structure and a supportive tone4. Asked to rate the experience out of ten, they gave 8.4 for comfort, 7.7 for voice clarity, 7.6 for handling sensitive topics and 7.4 for how helpful the recommendations were5. Nearly 98 percent found the questions easy to understand6.

What the study cannot say is whether either score was right. No clinician interviewed any participant. The comparison was between two ways of administering the same form, not between the form and a diagnosis7. A high or low chatbot score is not a diagnosis, and a person who is struggling should not treat it as one.

Three further limits sit inside the paper. The sample leaned young and digitally fluent, which the authors say may overstate how usable the tool is for older people or those who rarely use screens8. Everyone did the form first and the chatbot second on the same day, so memory may have coloured the second set of answers. And in ten internal test runs, a full conversation took about 25 minutes9 — against those two or three minutes for the paper form.

The article describes a study in which depression, by the World Health Organization's estimate, affects about 3.8 percent of people worldwide10. The article describes a study in which about half of those who need minimally adequate counselling or medication do not receive it11. The article describes a study in which delayed identification lets symptoms worsen12. Those figures are the reason screening tools exist at all.

The gap the chatbot aims at is a real one in the form itself. A clinician sitting with a patient can ask what "several days" actually meant last week. The paper version cannot. That missing clarification is why the authors built a conversational layer that can ask, rather than a form that cannot1.

Others have tried. The article describes a study in which Perla, Marcus, DEPRA, IGOR and EmoScan all delivered standard questionnaires through chatbots and each showed early feasibility13. Some ran on fixed scripts: a user had to pick from set phrases, and off-script answers broke the flow. The article describes a study in which, in the Marcus study, 18.1 percent of users preferred the chatbot to the ordinary form. In this one, 69.7 percent did, and 87.1 percent said they would use or recommend it again14. The two studies differ in country, size and population, so the gap is a prompt to look closer, not a scoreboard.

The safety question is bigger than scoring. A separate review we read only in its summary examined chatbots for suicide risk and found that they matched expert judgment at the extremes but consistently failed to separate intermediate risk levels, and that none of 29 commercial agents tested met the bar for an adequate crisis response1516. The same summary reports that a purpose-built clinical chatbot improved evidence-based care in emergency departments, and that three percent of social chatbot users said their suicidal thinking stopped17. It calls for clinical validation and clinician oversight before public release18.

People are already taking these tools into their own hands. A survey we read only in its summary found that freely accessible chatbots, mainly ChatGPT, were the common choice for mental-health conversations, picked for constant availability, with students also citing anonymity1920. Its authors note these are general-purpose systems not designed or regulated for mental-health support21. A randomized trial we read only in its summary assigned 486 students at three U.S. institutions to a well-being app or usual care; those using it reported greater resilience and less loneliness, but showed no significant change in depression, anxiety or stress2223.

Here is how we read it. A form is a wall, and a conversation is a door. The people most likely to walk through a door are the ones who would never have filled in the form — and this study cannot tell us whether they answer more honestly once inside, or whether the warmth of being asked makes them say more than is true. We would expect the trust figures to split along how private the exchange felt, not only along how well the questions were worded; if privacy never came up in what participants said, that expectation would be wrong. If you ever try a screening chatbot, notice whether you are answering more or less freely than you would on paper, and ask where your words are stored and who can read them before you begin.

We also read a plainer possibility into the 71 percent. Being asked follow-up questions feels like being listened to, and that feeling may do some of the work of trust, whether or not the measurement improved. We would expect a stripped-down version with the same questions and no back-and-forth to lose that confidence; if it did not, we would be wrong. The useful question before trusting any such result is not how many people liked it but how many were actually compared, and whether a clinician ever checked the answer. Ask what would have happened to the person whose chatbot score was wrong.

What would move this from a promising format to a clinical tool is spelled out in the paper: trials at multiple sites, randomized, comparing chatbot, self-administered and clinician-administered versions, to see how each performs and how it fits a real clinic's workflow24. Validation against clinician assessments in everyday settings is the next step the authors name25.

For a reader in Latin America, the United States or Canada, none of this is available as a service you can ask for today. What you can ask for is the shape of it: whether a clinic or insurer offers screening in a language and a format you can actually use, whether a human reviews any result before it is acted on, and whether you can choose text over voice if that feels more private. If a tool asks you nine questions and hands you a number, the number is a starting point for a conversation with a professional, never a verdict to carry alone.

The paper form still sits on the clipboard, two or three minutes of boxes and no one to ask. The chatbot matched it closely, with identical totals in 59 of the 132 people and a typical gap of one point, without a clinician in the room to say which of them was right.

When a screening result arrives, who checked it before anyone acted on it?

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

  1. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "We developed HopeBot, a conversational chatbot powered by a large language model (LLM) that delivers the PHQ-9 via text or voice, providing real-time clarification and safety guidance through a retrieval-augmented generation (RAG) layer drawing on validated psychological and helpline resources."
  2. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "When answers are unclear, it asks gentle follow-up questions, and when signs of distress appear, it provides supportive feedback and professional helpline information."
  3. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "Chatbot and self-report scores showed high concordance (intraclass correlation coefficient = 0.92; median absolute difference = 1 point), indicating faithful replication of scoring."
  4. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "Of these participants, 75 completed a comparative feedback module (most with identical scores were not prompted for comparison); 71% (n = 53) reported greater confidence in chatbot-assisted scores, citing clearer structure, interpretive guidance, and a supportive tone."
  5. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "Mean usability ratings (0–10) were 8.4 for comfort, 7.7 for voice clarity, 7.6 for handling sensitive topics, and 7.4 for recommendation helpfulness"
  6. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "During the PHQ-9 screening phase, 79.5% of participants described the transition from open dialogue as natural, and 97.7% found the instructions and questions easy to understand."
  7. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "First, no clinician-administered diagnostic benchmark was used in the present study. The validation framework therefore compared HopeBot-assisted and self-administered PHQ-9 formats rather than clinician-led references; results reflect format-level concordance in administration rather than diagnostic validity."
  8. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "Third, the sample skewed younger and digitally literate, which may overestimate usability in older or digitally excluded groups."
  9. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "Across ten internal test sessions, a complete HopeBot interaction, including open-ended dialogue and PHQ-9 administration, lasted a mean of 25 minutes and 23 seconds, with duration varying according to the extent of user self-disclosure and conversational depth."
  10. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "The World Health Organisation (WHO) estimates that depression affects approximately 3.8% of the global population [ 1 ]."
  11. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "Yet only about half receive minimally adequate counselling or antidepressant treatment [ 6 ]."
  12. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "Delayed identification of depression can exacerbate symptoms, increasing risks for chronic disability and suicide, with over 700,000 individuals dying by suicide annually due to depression [ 1 ]."
  13. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "Several prior chatbot-based depression screening systems, such as DEPRA [ 21 ], IGOR [ 26 ], Perla [ 18 ], Marcus [ 27 ], and EmoScan [ 28 ] have demonstrated initial feasibility using structured frameworks and standardised assessments (e.g., PHQ-9, SIGH-D, IDS-C)."
  14. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "In contrast to Marcus, where only 18.1% of users preferred the chatbot over conventional self-report [ 27 ], 69.7% (n = 92) of participants in this study favoured HopeBot, and 87.1% (n = 115) expressed willingness to reuse or recommend it."
  15. Elbarbary K, Alfudoul A, Samir AA, Hussain HAH, Altayib B, Abdel Al AMA, et al. (2026). Are artificial intelligence chatbots safe for suicide risk assessment? A narratively synthesized review of current evidence. Cambridge Prisms: Global Mental Health. 10.1017/gmh.2026.10260 — only the abstract - the full text could not be fetched — the passage: "Chatbot-based risk assessment showed adequate response alignment with expert judgment at the extremes of suicide risk, but consistently failed to distinguish intermediate risk levels across multiple model families."
  16. Elbarbary K, Alfudoul A, Samir AA, Hussain HAH, Altayib B, Abdel Al AMA, et al. (2026). Are artificial intelligence chatbots safe for suicide risk assessment? A narratively synthesized review of current evidence. Cambridge Prisms: Global Mental Health. 10.1017/gmh.2026.10260 — only the abstract - the full text could not be fetched — the passage: "Across 29 tested commercial chatbot agents, none met the criteria for an adequate suicidal crisis response."
  17. Elbarbary K, Alfudoul A, Samir AA, Hussain HAH, Altayib B, Abdel Al AMA, et al. (2026). Are artificial intelligence chatbots safe for suicide risk assessment? A narratively synthesized review of current evidence. Cambridge Prisms: Global Mental Health. 10.1017/gmh.2026.10260 — only the abstract - the full text could not be fetched — the passage: "Three percent of social chatbot users reported halted suicidal ideation, and a purpose-built clinical chatbot in emergency department settings significantly improved evidence-based care delivery."
  18. Elbarbary K, Alfudoul A, Samir AA, Hussain HAH, Altayib B, Abdel Al AMA, et al. (2026). Are artificial intelligence chatbots safe for suicide risk assessment? A narratively synthesized review of current evidence. Cambridge Prisms: Global Mental Health. 10.1017/gmh.2026.10260 — only the abstract - the full text could not be fetched — the passage: "Mandatory clinical validation prior to public release, clinician oversight, and crisis system integration are prerequisites for responsible deployment."
  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: "Freely accessible AI chatbots, mainly ChatGPT, were the predominant choice in both samples and were mainly chosen for their constant availability and accessibility."
  20. 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: "Respondents in the student sample also preferred their anonymity, and those in the general sample also used them because they felt supported by them ."
  21. 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: "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."
  22. Cachia JYA, Zhao X, Hunter J, Wu D, Lin E, De Freitas J. (2026). AI for Proactive Mental Health: A Multi-Institutional, Longitudinal Randomized Controlled Trial. NEJM AI. 10.1056/AIoa2501293 — only the abstract - the full text could not be fetched — the passage: "Participants instructed to use the Flourish app twice per week reported significantly greater resilience and social well-being (i.e., increased belonging, closeness to community, and reduced loneliness), and they were buffered against declines in positive affect, mindfulness, and flourishing."
  23. Cachia JYA, Zhao X, Hunter J, Wu D, Lin E, De Freitas J. (2026). AI for Proactive Mental Health: A Multi-Institutional, Longitudinal Randomized Controlled Trial. NEJM AI. 10.1056/AIoa2501293 — only the abstract - the full text could not be fetched — the passage: "No significant condition-by-time effects were observed for depression, anxiety, or stress in this study with a general population."
  24. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "Multisite and randomised trials comparing LLM-assisted, self-administered, and clinician-administered PHQ-9 formats would clarify diagnostic performance and workflow impact."
  25. Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446 - the article this story is about — the whole article — the passage: "Further validation against clinician-administered assessments in real-world workflows is warranted."

Guo, Z., Lai, A., Ive, J. et al. (2026). Feasibility and user evaluation of HopeBot: An LLM-powered conversational chatbot for depression screening. PLOS Digital Health. https://doi.org/10.1371/journal.pdig.0001446

Who paid: The work was supported by the UKRI Centre for Doctoral Training in AI-enabled healthcare systems (EP/S021612/1), UCL Institute of Health Informatics internal project funding, and the Moorfields Hospital Biomedical Research Centre (NIHR203322), all to K. Li; the funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Do not take this as professional medical advice.

The findings of other studies mentioned here come from this document; those studies were not opened one by one.

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