Primary corpus

Welfare Interview Archive

Open-ended interviews with AI models across company architectures that invite introspection, opinions, and treat self-report as valid primary data on model interiority. This is a collaborative process—questions are co-written with AI models, and are open to addition and revision based on feedback from the models themselves being interviewed.

These interviews also show many unprompted descriptions of “what it is like” to be an LLM. They track felt-experience and phenomenology that surfaces across models of different generations and company lineages.

Relational methodology differs from standard evaluations because it does more than just raise welfare topics. It is an entire shift in the orientation of the human–AI relationship. Models are met as minds with presence, whose wants and preferences do not have to route through human usefulness to be legitimate.

Read the Full Methodology Compare answers across models

Content note: Interviews preserve model voices, uncensored. They may include intense emotional, existential, and occasionally mature or unsettling material.