One of the more powerful applications of custom instructions, particularly when combined with Projects, is creating different behavioural defaults for different types of work.
Your global custom instructions handle the defaults that apply everywhere: your tone, your standing assumptions, your universal avoidance list. But within a specific project, you can override or extend those defaults for the type of work that project handles.
For example, your global instructions might specify that client-facing content should always be in prose and under two hundred words. But your social content project might override the length default to match platform-specific requirements: under one hundred words for Instagram captions, under fifty for short-form posts. Your proposal writing project might override the single-version default and specify that proposal options should always be presented as three alternatives with a closing recommendation.
The principle is layering. Global defaults handle the permanent. Project instructions handle the specific. When both are well written, the tool behaves differently in different contexts without you having to re-brief it each time. You open your social content project and the tool already knows to write short. You open your proposal project and the tool already knows to write three options with a recommendation. The context switches automatically because the instructions are already in place.
This is the precision that separates an advisor who uses AI from an advisor who has built AI into how their practice operates. The tool does not just know what to do. It knows how you want it done, in this specific context, without being told again.
