Utah student finds AI chatbots lack empathy in teen emotional tests

Testing AI for teen emotions

Ellianna Gu, an eighth-grader from Salt Lake City, Utah, conducted a study to see how AI chatbots handle the complex emotions of teenagers. She decided to investigate this topic after noticing that many young people are beginning to use these services for friendship and emotional support. Her project, titled "Bot-tling up Emotions," sought to understand how these systems interpret and respond to difficult life situations.

Gu focused her research on seven specific types of emotional challenges that are common for people in their teenage years. These included issues like academic stress, social anxiety, and loneliness, as well as anger, sadness, and family conflicts. By looking at these specific categories, she wanted to see if the technology could provide meaningful help to a vulnerable demographic.

The project was inspired by previous research, including a study from Stanford, which suggested that AI chatbots can sometimes be ineffective or even provide dangerous responses. Gu wanted to see if current models like ChatGPT, Claude, and Gemini could offer a better experience for those seeking help. Her work eventually made her a finalist in the 2026 Thermo Fisher Scientific Junior Innovators Challenge.

How the study was conducted

To conduct her research, Gu used AI to help generate realistic situations based on the emotional challenges she identified. She then submitted a total of 84 scenarios to three major chatbots: OpenAI’s ChatGPT, Anthropic’s Claude, and Google Gemini. A key part of her methodology involved varying the way the prompts were written to see if the tone of the user changed the AI's output.

She intentionally included both formal writing styles and informal slang in her prompts. This allowed her to determine if the chatbots were more capable of understanding the way teenagers actually speak in daily life. By comparing these different styles, she could measure how much the linguistic delivery of a problem affected the quality of the support received from the machine.

Gu evaluated the responses from the three chatbots using four specific criteria: empathy, clarity, problem-solving, and helpfulness. Most of these categories were scored on a scale of one to three, while empathy was given a slightly broader scale of one to four. To ensure her results were accurate, she assessed the answers manually and also used AI to analyze the data for a dual evaluation.

Findings on empathy and style

The results of the study showed that the three chatbots were generally good at providing organized and understandable answers. They were able to structure their thoughts clearly, which is a core strength of current large language models. However, Gu found that these systems often failed to display much empathy, which is a critical component when a user is discussing deep feelings like anxiety or loneliness.

The research also highlighted a significant discrepancy in how the chatbots reacted to different writing styles. A scenario presented using slang could result in a different response than the exact same problem described in formal language. This suggests that the way a teenager chooses to express their feelings might change the type of advice or support they receive from an AI system.

In terms of overall effectiveness, the chatbots showed varied performance across the different scenarios. According to the project summary, only Claude received a score of three for effectiveness among the three services tested. This indicates that while the technology is improving, it still struggles to consistently provide high-quality support for emotionally challenging situations.

The human element in AI

Gu's research underscores a growing concern about the limitations of human-like interaction in artificial intelligence. While these tools can provide clear information, they do not possess the genuine emotional resonance that a human friend or counselor provides. Her findings suggest that the lack of empathy could be a significant barrier for teenagers who are looking for a deeper connection than a chatbot can offer.

The study serves as a reminder of the importance of human judgment when evaluating AI-generated content. Because the chatbots' responses can change based on the style of the input, users might receive inconsistent support depending on how they phrase their problems. This variability is a key factor for developers and parents to consider as AI becomes more integrated into daily life.

Despite her focus on AI, Gu's personal interests are rooted in human biology and health. She plays the piano and clarinet and hopes to become a neurologist to study the human brain. She expressed a desire to help people suffering from life-altering conditions like Alzheimer’s and Parkinson’s, showing a commitment to understanding human health from a biological perspective.

What comes next for AI support

The findings from Gu's project provide a foundation for further research into how AI can be made safer and more empathetic. As more people turn to these tools for emotional support, there is a need to understand the risks of receiving structured but emotionally hollow advice. Her work highlights the necessity of refining how models interpret the nuances of teenage speech and emotional distress.

Future research may focus on narrowing the gap between the clear, structured answers AI provides and the empathetic responses humans expect. Developers might look at ways to make models more sensitive to slang and informal communication to ensure that the support provided is consistent. Gu's project serves as a starting point for identifying these specific areas of improvement in the current AI landscape.

As the technology continues to evolve, the balance between AI efficiency and human empathy will remain a central theme. For now, her study suggests that while chatbots are becoming more capable of organizing information, they still have a long way to go in truly understanding the complexities of human emotion. Her work provides a clear look at the current boundaries of what these systems can do for young people today.