0
I sometimes start a fresh Claude chat with the exact same prompt and get a noticeably different answer. The facts may be similar, but the structure, level of detail, or recommendation can change quite a bit.
Is this expected model variation, or could conversation settings, selected model versions, or hidden project instructions be influencing the result? I’m curious what people check when they want a more consistent Claude response without turning the prompt into a long list of constraints.
