Custom Instructions vs Memory vs Projects: What Goes Where
What it is
A routing guide for the three places ChatGPT stores context: global preferences go in custom instructions, project-specific rules and files go in a Project, and durable facts it learns go in Memory — with one-off detail staying in the prompt.
Why it works
These features overlap enough that people dump everything into custom instructions, which then bleeds the wrong context into unrelated chats. Sorting context by scope — global, per-project, or incidental — keeps each answer clean and stops one project's rules from contaminating another.
When to use it
When you're deciding where a piece of standing context should live, or cleaning up instructions that have grown into a junk drawer.
When not to use it
For truly one-time detail, none of these — just put it in the prompt. Don't create a permanent home for something you'll never reuse.
Prompt
Here's some context I want ChatGPT to keep in mind: <the fact/rule>. Does this belong in my global custom instructions, a specific Project, or Memory — and why? Assume I care about not leaking it into unrelated chats.Example
'Always use metric units' → custom instructions; 'this client's brand voice is playful' → that client's Project; 'my dog's name is Mango' → Memory; 'make this one reply formal' → the prompt.
Common mistakes
- Putting project-specific rules in global instructions, skewing every chat.
- Relying on Memory for structured reference material better held in a Project's files.
- Duplicating the same rule across all three, so edits get out of sync.