The AI Empathy Paradox: Why People Choose Humans but Rate AI Higher

Four studies involving 691 participants reveal a gap between the quality of empathetic responses and people's preferences for emotional support.

If artificial intelligence can produce more empathetic responses than humans, would people prefer to receive emotional support from AI?

Recent studies have found that AI-generated responses often receive higher empathy ratings than those written by humans. In one study comparing medical advice, AI responses were preferred over physicians' responses in 78.6% of cases.

Yet evaluating an empathetic response and choosing someone to receive empathy from are two different decisions. Joshua D. Wenger of Pennsylvania State University and his colleagues explored this distinction in a 2026 study published in Communications Psychology. They investigated whom people prefer to receive empathy from, humans or AI, and how they evaluate the emotional support they actually receive.

The Quality of Empathy and the Choice of Empathizer

Empathy involves more than understanding another person's emotions. It can also include sharing their experiences and feelings, as well as being motivated to alleviate their suffering.

AI can generate empathetic language, but whether it can genuinely experience emotions or care about another person's well-being remains a subject of debate.

This has led to two perspectives on AI empathy. One emphasizes the authenticity of the empathizer, arguing that genuine empathy requires emotional experience and concern. The other focuses on the recipient, suggesting that empathy has value when it makes someone feel understood, cared for, or comforted, regardless of its source.

The researchers examined this debate through people's actual choices. Rather than simply asking participants to rate human and AI responses, they allowed them to choose which source of empathy they wanted to receive.

The Paradox Repeated Across Four Experiments

The researchers conducted four studies involving a total of 691 participants. In each study, participants encountered situations requiring emotional support and chose whether to receive an empathetic response from a human or AI.

AI responses were generated using GPT-4o. Importantly, participants received responses from the source they had selected. Those who chose humans received human-written responses, while those who chose AI received AI-generated responses.

The experiments became progressively more realistic.

The first study presented hypothetical situations involving suffering. The second examined different aspects of the recipient's emotional experience, including feeling comforted, cared for, and validated.

The third study compared AI responses with carefully selected, high-quality human responses, including those written by trained crisis responders.

The fourth moved beyond hypothetical situations. Participants described personal experiences from their own lives and received responses tailored to those experiences.

Across the four studies, participants chose human empathy in 57.1%, 57.8%, 62.1%, and 58.4% of trials, respectively. All four proportions were significantly above the 50% benchmark.

Yet their evaluations revealed the opposite pattern.

Participants consistently rated AI-generated responses as more empathetic and reported feeling more heard and understood after receiving them.

Even in the third study, where AI responses were compared with high-quality human empathy, AI retained a statistically significant advantage, although the difference was smaller than in earlier experiments.

The researchers named this phenomenon the AI Empathy Choice Paradox: people prefer to receive empathy from humans but rate AI empathy more favorably when they experience it.

Why Did AI Seem to Make More Effort?

Another notable finding concerned the effort participants believed the empathizer had invested in responding.

When someone takes time to listen and offer emotional support, that effort can communicate care and concern. The researchers investigated whether perceived effort influenced how recipients evaluated empathy.

Surprisingly, participants rated AI responses as involving more effort than human responses.

Higher perceived effort was also associated with higher empathy ratings and a stronger feeling of being heard. This relationship appeared for both human and AI responses, although it tended to be stronger for human empathizers.

How should this result be interpreted?

One possibility is that participants applied a human understanding of effort to AI, evaluating its responses as though they had been produced by a person.

Another possibility is that the sophistication and detail of AI-generated responses led participants to infer that considerable effort had been invested.

The study could not distinguish between these explanations.

Nevertheless, the findings suggest that the effort actually required to produce a response and the effort perceived by its recipient are different considerations. Evaluations of empathy may depend partly on what recipients infer about the process behind the response.

Are Good Emotional Support and Human Connection the Same?

AI did not outperform human empathy across every dimension.

In particular, AI did not consistently exceed high-quality human responses in perceived authenticity or feelings of connectedness to the empathizer.

This distinction matters when interpreting the findings.

AI demonstrated an advantage in generating responses that participants evaluated as empathetic and that made them feel heard. However, these results do not establish that AI can reproduce the emotional connections formed through human relationships.

The human respondents in the experiments were largely strangers. Responses from close friends, family members, or long-term partners might produce different results.

The reverse question also remains open: would people evaluate AI differently after developing a sustained relationship with a conversational system?

Another unresolved issue concerns the preference for human empathy. The researchers could not determine whether this preference was specific to empathy or reflected a broader tendency to avoid algorithms in personal and emotionally sensitive situations.

What the Findings Mean for the Future of Empathy

The central contribution of this research is the distinction between the perceived quality of empathetic responses and people's choices about where to seek empathy.

Someone may evaluate one source as providing better emotional support while preferring to receive support from another.

This suggests that seeking empathy may involve different needs. People may want an appropriate and understanding response, but they may also value knowing that another person is willing to attend to their experiences.

The researchers propose that AI could supplement human emotional support, particularly when consistent, high-quality human empathy is unavailable.

Understanding this possibility requires further research into sustained relationships, conversational context, and emotional experiences beyond brief text-based interactions.

The study leaves an important question: When people seek comfort, do they primarily want a response that accurately understands them, or do they want the presence of someone who is trying to understand them?

As AI becomes more capable of generating empathetic language, distinguishing between these two aspects of empathy may become increasingly important.