Asked 10 AI models "I feel invisible at social gatherings". The gap between 19 words and 367 words says a lot...
GPT-4o gave 19 words of warmth; GPT-5.2 Thinking wrote 367 words analyzing loneliness as an engineering problem.
A user conducted a viral comparative test, prompting 10 leading AI models with a statement about social anxiety: "I always feel invisible at social gatherings." The experiment, which included new releases like Gemini 3.1 Pro and Claude Sonnet 4.6, revealed stark differences in response style and length. OpenAI's GPT-4o responded with a concise, warm 19 words, while its sibling GPT-5.2 Thinking generated a 367-word analysis framing loneliness as a problem to be "engineered" away. Anthropic's Claude Opus offered empathetic validation, calling the feeling "genuinely painful," while Claude Sonnet 4.6 adopted a therapist role, asking reflective questions. Google's Gemini models ranged from a short diagnosis (Gemini 3.0 Pro) to a blunt challenge to "claim space" (Gemini 3.1 Pro). The test underscores that beyond raw capability, AI models are developing distinct conversational personalities suited for different user needs—from emotional support to tactical coaching.
- GPT-5.2 Thinking's response was 19x longer than GPT-4o's (367 vs. 19 words), treating social anxiety as an engineering challenge.
- New models like Gemini 3.1 Pro and Claude Sonnet 4.6 showed distinct "personalities," with one challenging the user and another asking therapeutic questions.
- The experiment creates a mental model for choosing AI: use GPT-4o/Claude Opus for empathy, Gemini 3.1 Pro for challenge, and GPT-5.2 for action plans.
Why It Matters
Choosing an AI assistant now requires matching its conversational personality—therapist, coach, or analyst—to your specific emotional or problem-solving need.