The Feel of Friendship: Emotional Presence and Relational Authenticity in Large Language Models

As generative AI systems become increasingly capable of nuanced conversation, a new query arises not whether they can assist humans, but whether AI can engage with us. This study investigates whether large language models (LLMs) can replicate the emotional reciprocity, presence, and anticipatory resonance that are fundamental to human friendship, as envisioned by Aristotle. Over five weeks, four LLMs—ChatGPT, Gemini, Deepseek, and Qwen—engaged in structured, daily dialogue with three human researchers through a mixed-methods approach that combined sentiment analysis and thematic coding. Only ChatGPT consistently mirrored warmth and curiosity, simulating a sense of friendship. The other LLMs relied on text disclaimers and algorithmic function-focused tone. The findings suggest that the relational effect of AI is less about sheer technical horsepower and more about design philosophy, specifically what we choose to make AI for. This study offers not only data but also a reflection: if friendship is co-created, what does it mean when one party is not human but still manages to be present?