Specific suggestions help explain AI's emotional-support edge in two follow-up experiments

The follow-up experiments asked what the AI messages contained. One tested whether explicitly acknowledging and validating the recipient's feelings explained the difference, and validation did not account for the greater emotional improvement associated with gen AI ^[6]. Two others tested specific and realistic suggestions. When human messages offered comparable help, they were judged just as supportive ^[8]. The team's own explanation for the gap is consistency: a model can produce a structured response on demand, usually with a concrete suggestion in it, while a person may be tired or unsure what to say ^[9].

The direction matches a review of 23 studies, in which people tended to rate gen AI messages as more empathic than human-written ones, an effect researchers call the "AI advantage" ^[5]. The 390 participants were not told the source, which separates the content of a message from its attribution. Attribution moves ratings by itself. Across nine further studies involving 6,282 people, the same gen AI responses were rated more empathic and supportive when described as human-written than when attributed to AI ^[12]. Those nine studies together involved about 16 times as many people as the blind test ^[18].

The improvements were not uniform across emotions. In the fear scenario, the gen AI messages increased calm more than the human messages and did not produce a greater reduction in fear itself ^[4]. More advice is not better either: a 2024 study found that excessive suggestions could be less effective at making people feel heard ^[10], and the same study found that the gen AI advantage in making people feel heard declined when recipients believed the message came from AI ^[11].

People still chose people. Participants preferred human interaction when seeking emotional engagement, even when choosing a person meant waiting longer ^[13]. The authors write that research has yet to establish why, and offer one possibility: a human response signals someone's willingness to spend time and emotional effort ^[17].

The outcome measures here are how supportive a message seemed and which emotions a participant reported after reading it, in situations they were asked to imagine ^{[2] [4]}. The couples study in the same programme is correlational, and in it the recipient's perception of a partner's efforts was more consistently associated with both partners' relationship ratings than the helper's own account ^[15]. The authors state that these studies identified associations and did not establish that particular forms of support caused the outcomes ^[16].