Glossary · Platforms
Constitutional AI
Anthropic's training method that shapes Claude's behavior using a written set of principles the model critiques and revises its own outputs against, rather than relying only on human feedback.
What it is
Constitutional AI (CAI) trains a model to evaluate and rewrite its own responses according to a "constitution" — a set of guiding principles — reducing reliance on large volumes of human-labeled preference data. Anthropic has published this as a core technique behind how Claude is trained to behave.
Why it matters for visibility
CAI shapes Claude's general caution and sourcing habits, including its tendency to hedge on unverified claims rather than assert something as fact without support. That indirectly affects how confidently Claude names or recommends a specific brand versus giving a more balanced, multi-option answer.
See more terms in the full glossary, or read the AI Visibility 101 guide for the full picture.