It is two in the morning. Someone has just received a confusing diagnosis, or is struggling with a symptom their doctor brushed aside, or simply feels scared and alone. They do not call a hospital. They open an online forum and type out their worries to strangers. And somewhere, perhaps on another continent, a stranger who went through something similar reads the post and tries to find the right words to reply.
That quiet, invisible act of kindness is what keeps online health communities alive. A study presented at the International Conference on Information Systems (ICIS) 2025 by Danielly de Paula, Nikita Shishelyakin, Vanessa Kipping, Yuexing Hao, Aparna Balagopalan, Marzyeh Ghassemi, and Ariel Stern asks whether artificial intelligence can help these everyday helpers do their job better, and under what conditions such teamwork is actually worth it.
The hidden fatigue of helping
Online health communities are digital spaces where patients and caregivers share experiences, advice, and comfort. Think of forums for people living with diabetes, new parents dealing with sleepless nights, or the women’s health communities on Reddit that the researchers draw on. These spaces fill gaps that formal healthcare often leaves open. Doctors have limited time, and many questions are not strictly medical. How do you cope with fatigue at work? What does it feel like after a particular procedure? Is it normal to feel this anxious?
But helping is hard work. Writing a thoughtful, accurate, compassionate reply takes time and emotional energy. Regular contributors can burn out. When they do, questions go unanswered and communities wither. The authors call this response fatigue, and it threatens the long-term survival of these valuable spaces.
This is where generative AI enters the picture. If an AI tool could help a volunteer polish, organise, or enrich their reply, perhaps more people would contribute, and contribute more often. The key word is help. The idea is not to replace the human with a chatbot, but to pair the human’s lived experience and empathy with the AI’s speed and ability to structure information.
Not all questions are the same
The authors make a clever observation. People in health forums typically ask for one of two things. Some want instructional support: practical information, such as how to manage a medication schedule or what questions to ask a specialist. Others want emotional support: reassurance, validation, and the feeling that someone understands.
It seems reasonable to expect AI to shine at the first kind and struggle with the second. AI is good at organising facts. But can it help someone sound genuinely caring? Or will an AI-polished message of sympathy feel hollow, like a greeting card written by a committee? The authors build this distinction into their research, predicting that the value of human–AI collaboration will depend on the situation.
Trust matters too. People who are sceptical of AI may judge a response more harshly if they suspect a machine had a hand in it, while AI enthusiasts may rate the same response more generously. In other words, our opinion of a message may depend as much on who we think wrote it as on what it actually says.
Two experiments in the making
The paper sets out two experiments. To prepare, the team built a dataset of real posts and replies from a public Reddit women’s health community. Using GPT-4o, they sorted the posts into emotional and instructional requests and then created an AI-enhanced version of each human reply, so every original response had an improved twin.
In the first study, about 300 participants will rate pairs of responses, original and enhanced, on several quality measures, without being told which is which. Participants will also say whether they suspect a response was written by AI, and report how much they trust AI in general. This will show whether AI-enhanced replies are actually seen as better, for which kind of question, and how much people’s prejudices about AI colour their judgement.
The second study goes a step further. The lessons from the first study will shape a prototype writing assistant. Participants will then use it to compose their own replies to support-seeking posts. The question here is not just whether the replies get better, but whether the people writing them feel more confident. Researchers call this knowledge self-efficacy: your belief that you have something valuable to offer. Earlier research suggests that people who feel this way share knowledge more often and stay engaged longer. If an AI assistant boosts that confidence, it could help keep communities alive, not just make individual posts shinier.
Why this is about more than technology
This research is still in progress, so the results are yet to come. But the questions it asks are important. Much of the public debate about AI in health focuses on diagnosis, or on whether chatbots can replace doctors. This study looks at something more modest and perhaps more realistic: AI as a helpful sidekick for ordinary people trying to support each other.
The approach is also refreshingly human-centred. Rather than asking whether AI can write better replies than people, it asks when a human and AI working together produce something more valuable than either alone, and what that partnership does to the human involved. Does it make them more confident, or more dependent? Does it make their kindness more effective, or does it sand away the rough edges that made it feel real?
There is a delicate balance here. Someone in distress does not want a perfectly formatted essay. They want to feel heard by another human being. If AI helps a nervous volunteer organise their thoughts, add a useful resource, or avoid an accidental unkind phrase, that is a win. If it turns every heartfelt reply into polished corporate prose, something precious is lost.
For platform designers, the lesson may be to offer AI help selectively, perhaps more actively for practical questions and more gently for emotional ones. For the rest of us, the study is a reminder that the best uses of AI may not be the flashy ones. Sometimes the most meaningful role for technology is simply helping one stranger find the right words for another at two in the morning.
Reference
de Paula, D., Shishelyakin, N., Kipping, V., Hao, Y., Balagopalan, A., Ghassemi, M., & Stern, A. (2025). Are we better together? Investigating the value of human–AI collaboration for online health communities. In ICIS 2025 Proceedings