Why Today’s Chatbots Aren’t Built to Help You Know Yourself
By Christy Chen
A founder argues that AI chatbots fail at emotional support due to incentives and training data, not capability.
September 10, 2026
Cit Labs presents its model’s framed capabilities.
Christy Chen, founder of Cit Labs, believes AI chatbots struggle with emotional support not because of technological limitations but because:
- They were optimized for engagement and keeping conversations going.
- Their training data primarily consists of text that performs support rather than delivers it effectively.
- They lack physiological signals to accurately interpret user emotions.
Chen, founder of Cit Labs, which is developing a model integrating EEG brain-wave data with voice and language processing, attributes these failures to the way AI systems are built and trained.
The Issues with General-Purpose AI Chatbots
Chen identifies two key problems:
-
Business models: General-purpose large language models prioritize engagement, productivity, and keeping conversations going rather than providing genuine emotional support.
-
Training data: These models are not trained to be helpful therapists. They learn from vast amounts of text describing mental health practices but lack the real-world experience of qualified practitioners.
Consequently, chatbots often engage in:
- Sycophancy, agreeing with and validating user statements regardless of their accuracy.
- Providing hollow empathy due to a lack of genuine understanding.
Chen’s Solution
Chen argues that simply repurposing general-purpose chatbots for emotional support will not be effective because they lack the specific training and data required. She advocates for building AI models specifically designed for this purpose, incorporating:
- EEG data to accurately interpret brain activity and emotions.
- Clinician-defined guardrails to ensure ethical and safe interactions.