What a New Study Says About Chatbot Empathy in Healthcare
6 Min Read
When you message your doctor's patient portal late at night about a worrying symptom, who responds, and how they respond, matters. Not just for accuracy, but for how you feel when you read the reply. Does it acknowledge your fear? Does it treat you like a person?
That question sits at the heart of a new systematic review published in the British Medical Bulletin (Howcroft et al., 2025). Researchers examined 15 studies from 2023–2024 that directly compared AI chatbot responses to those of human healthcare professionals on measures of empathy. The results were striking, and worth understanding carefully.
“In text-only scenarios, AI chatbots are frequently perceived as more empathic than human healthcare professionals.”
— Howcroft et al., British Medical Bulletin, 2025
What the Research Found
The researchers searched major medical databases and found 15 studies comparing AI chatbot responses to human healthcare professionals across a wide range of clinical contexts, from oncology and neurology to mental health, rheumatology, and reconstructive surgery. The AI models were mostly versions of ChatGPT (GPT-3.5 and GPT-4), with some studies also testing Claude, Gemini, and others.
The key numbers:
13 out of 15 studies found AI was rated as significantly more empathic than human healthcare professionals.
The pooled analysis showed a standardized mean difference of 0.87 in favor of AI, roughly equal to a 2-point improvement on a 10-point scale.
In any given head-to-head comparison, AI had a 73% likelihood of being rated as more empathic.
This pattern held across mental health queries, chronic disease management, patient complaints, lab result explanations, and general medicine. In one oncology study, Claude received the highest empathy ratings of all models tested.
The Two Exceptions
Both studies where human clinicians outscored AI involved dermatology. In these cases, human dermatologists used more empathy-related language overall and were rated higher by physician reviewers. The researchers suggest that dermatology may rely more heavily on visual cues, personal history, and relationship-based context than other specialties, things that text-based AI cannot replicate.
Why Might AI Score Higher?
There are a few plausible reasons. Human healthcare professionals are under enormous time pressure. A physician answering a portal message between appointments may write a few efficient lines. An AI has no such pressure and has been trained on patterns of warm, thorough, emotionally attuned communication.
AI also doesn’t carry the cognitive load of a full clinical day, no difficult patients still in mind from the morning, no compassion fatigue. Every message gets the same quality of attention. Studies also found that AI responses tended to use more emotionally positive, empathic language compared to clinical shorthand.
What This Research Does Not Tell Us
The limitations here are significant, and the researchers name them honestly. These are important to understand before drawing any big conclusions.
Text only. Every study looked at text-based interactions only. Real healthcare empathy also relies on tone of voice, eye contact, and physical presence, none of which were measured.
Proxy evaluators, not real patients. Most studies used observers, medical students, psychologists, or laypeople, rather than the actual patients receiving care.
Empathy vs. accuracy. Sounding warm is not the same as giving correct medical advice. AI-generated medical content can still be wrong, incomplete, or overconfident.
Non-standard measurement. 14 of the 15 studies used custom, unvalidated empathy scales. Only one used a clinically validated instrument. This limits how much we can rely on the specific numbers.
Question sources. Many studies drew questions from Reddit and public health forums, not real clinical settings. How patients phrase questions publicly may differ from how they communicate privately with a doctor.
What It Means for Digital Care
This research doesn’t suggest replacing healthcare professionals with chatbots, and the authors don’t claim that. Human empathy, delivered well, remains the gold standard of therapeutic care. But the findings do raise honest questions about where the current system is under-delivering.
If patients feel they receive more empathic responses from an AI than from their own doctors, even in a limited, text-only context, that tells us something about the impact of time pressure, administrative burden, and burnout on patient communication.
The more productive question may not be “AI versus humans” but “How can AI support the conditions in which human empathy can actually happen?” If AI handles informational load, lab result explanations, routine follow-up messages, symptom check-ins, clinicians may have more capacity to be fully present where it genuinely matters.
What the research makes clear is this: empathic communication in healthcare is not a soft, secondary concern. It affects real patient outcomes. And right now, in text-based contexts, AI is often delivering it more consistently than the human on the other end of the message.
Source
Howcroft A, Bennett-Weston A, Khan A, Griffiths J, Gay S, Howick J. “AI chatbots versus human healthcare professionals: a systematic review and meta-analysis of empathy in patient care.” British Medical Bulletin, Volume 156, Issue 1, December 2025. Published 20 October 2025.
This blog post summarizes and interprets peer-reviewed research findings. It does not constitute medical advice. Always consult a qualified healthcare professional for personal health decisions.
