User Experience Design MSc student Yu-Ru Lin developed and tested two AI chatbots as part of her postgraduate research project, investigating whether psychological support delivered through conversational AI could help address one of the main barriers to breast cancer screening: fear.
The study involved 24 women, each of whom used both chatbots before taking part in a follow-up interview.
One chatbot, named Ember, was designed using principles from Acceptance and Commitment Therapy (ACT), a psychological approach that encourages people to acknowledge difficult thoughts and feelings while taking actions aligned with their values. The second chatbot provided factual information about breast screening but did not respond to users' emotions.
Both chatbots were built using the same interface and large language model, Anthropic's Claude, and differed only in the instructions guiding their responses.
Ember was designed around four ACT-informed processes: acknowledging feelings, helping users view worrying thoughts as thoughts rather than facts, connecting screening to what matters to them personally, and encouraging one small practical next step. The comparison chatbot focused solely on providing factual answers.
The research found that fear levels fell significantly after conversations with both chatbots, but the reduction was greater following interactions with Ember. Mean fear scores decreased from 21.0 to 14.4 after using the ACT-informed chatbot, compared with 17.0 for the facts-only version.
Participants' willingness to attend screening increased significantly only after using Ember. Women who reported higher levels of fear at the start of the study appeared to benefit most from the ACT-informed approach.
Researchers also found that participants were substantially more engaged with Ember. On average, women typed almost twice as many words when interacting with the chatbot, and 23 out of 24 participants wrote more to Ember than to the facts-only version.
The order in which participants used the chatbots also appeared to matter. Fear levels increased again for five of the 12 women who used the factual chatbot after Ember. No such increase was observed among participants who used Ember second.
Participants reported that having their fears acknowledged and normalised made them feel more comfortable. Some said that factual information alone introduced new concerns, such as potential pain during screening, without helping them manage those worries.
Psychological support approaches such as ACT have previously been studied primarily among people who have already received a diagnosis, while AI-based mental health chatbots have tended to focus on cognitive behavioural therapy. According to the project report, the use of ACT-informed conversational AI to support women who are avoiding breast screening had not previously been explored.
Research from the project is currently being prepared for journal publication.
Speaking about the project, Yu-Ru said:
Many women delay or avoid breast cancer screening, and psychological barriers are often among the main reasons. This matters because screening helps find cancer early, when it is easier to treat.
In my project, I designed Ember, an AI chatbot based on Acceptance and Commitment Therapy, which helps people make room for their worries instead of pushing them away. When I tested it with 24 participants, they felt less worried and more willing to go for screening after talking with Ember than after a chatbot that only gave information.
I hope this shows that AI health tools can do more than share facts and can support people emotionally to take the next step for their health.
As a UX designer, it was really rewarding to see design and AI work together to support people’s health decisions, and I’m excited to share this work more widely.
Explore Yu-Ru’s research and visuals featured alongside other innovative postgraduate work at the SDCA Postgraduate Showcase.