By skyforbes I use detailed custom instructions for style and structure. They include: very simple language, short sentences, no fluff, and a 3-part layout by default. GPT-5 Thinking follows these instructions well, but GPT-5.1 Thinking often does not.
Anyone experiencing this?