How ChatGPT projects and memory actually work

What it is good at

• Projects keep files, custom instructions and chat threads scoped to one client or collection, so nothing crosses over by accident.
• Memory persists useful facts, your role, your brand's tone, your usual output format, into chats where you never mention them.
• You can view and edit exactly what memory has stored, so it is not a black box.
• Project-level instructions override your general custom instructions, which is useful when a client needs a different voice.

Where it falls short

• Memory can pick up things you would rather it forgot; it needs periodic review, not just trust.
• Projects do not share memory with each other by design, so background you want everywhere still has to sit in general memory or custom instructions.
• Very large project file libraries slow retrieval and can cause it to miss a detail buried in an old upload.

How to use it for creative work

1. 01

   Separate a project per client or collection

   Create a new project the moment a new client or collection starts. Add its brief, brand guidelines and past approved copy as files so every chat inside it starts with the right context.

2. 02

   Write project instructions, not just a brief

   Use the project's custom instructions field to state tone, banned words and output format. This is read before every message in that project, unlike a brief you paste once and lose in scroll.

3. 03

   Audit memory every few weeks

   Ask 'what do you remember about me' and delete anything wrong or outdated. Old assumptions about a former job or discontinued product quietly skew results otherwise.

4. 04

   Keep general memory for stable facts only

   Reserve memory for things true across all your work, your job title, your preferred reply length, your usual audience. Project-specific facts belong in the project, not memory.

5. 05

   Reference files by name in the prompt

   In a project with many files, name the exact one you mean ('using the Q3 lookbook brief, not the SS26 one'). This cuts down on it quietly answering from the wrong document.

Ready-to-use prompts

List everything you currently remember about me and my work. Group it into: stable facts, preferences, and anything that might now be outdated.

Write custom instructions for this project. Client: [name]. Tone: [three adjectives]. Always do: [list]. Never do: [list]. Output format for copy requests: [format].

Using only [file name] in this project, answer the following without drawing on general knowledge or other files: [question].

Here is our brand voice guide: [paste]. Compare it against how you have been writing in this project and flag any drift.

Ignore everything remembered about me from other work and treat this as a fresh brief with only the files in this project as context.

Questions people ask

Does ChatGPT memory work across all my projects?

General memory is available in every chat and project by default, but each project also has its own instructions and files that only apply inside it. Project instructions take priority over general memory when they conflict.

Can I turn off ChatGPT memory?

Yes, memory can be turned off entirely or per-chat in settings, and you can delete individual remembered items without switching it off completely.

How many files can I put in a ChatGPT project?

You can add a substantial working library of files, but retrieval quality drops the more unrelated documents pile up, so keep each project to one client or collection rather than a general dumping ground.