How to make ChatGPT remember your project

ChatGPT can remember some things, but it does not keep a full picture of your project. Turn memory on and review it, put stable instructions in custom instructions, keep related work in a project, and paste a short project brief into a new thread. For work that spans weeks or several AIs, keep the brief in a document you own.

· 6 min read

You can tell ChatGPT the same thing three times and still get an answer that ignores it. The fix is not a better sentence. It is a system for keeping the context in front of the AI.

1. Turn memory on, then read it

ChatGPT has a memory feature that saves facts and preferences and brings them into later chats. Turn it on in Settings, then open the memory list and read it. Delete anything that is wrong, stale, or too personal. A bad memory is worse than no memory because the app applies it confidently.

Once a month, spend ten minutes on this. It is the difference between an assistant that knows you and one that repeats an old version of you.

2. Put the stable stuff in custom instructions

Custom instructions are for things that do not change: who you are, what you are working on, how you like answers, and what the AI should not do. Keep it short. Three or four sentences are enough. Do not paste your whole life story; the point is to stop repeating the basics.

3. Keep related work in one project

A project keeps chats and files together, so a new thread starts with the right context. Use one project per real thing: a job search, a course, a move, a client. Do not put everything in one project; the context gets noisy and the AI starts mixing up the details.

4. Paste a project brief into a new thread

This is the fix that works even when memory fails. Write a short brief in a document you own, then paste it at the start of a new thread. Three paragraphs:

  • What this is: the project, the goal, and the deadline.
  • What is already true: decisions, constraints, names, and numbers.
  • What I need now: the specific next step, and what not to do.

A good brief is 150 to 250 words. Longer is not better; it buries the point. Update it when something changes, and keep it in a place you control, not only inside one chat app.

Why it still forgets

  • Memory is a short list of facts, not a full transcript.
  • A new chat starts without the old thread unless memory or a project brings it in.
  • A long thread gets compacted, and the details are often what gets lost.
  • Switching to another AI or app starts from zero, because each product keeps its own memory.

If your work spans several AIs, keep the brief in a document. That is the only portable memory. A chat app that stores memory across AIs can carry it for you, but the document is still the backup.

Questions

Does ChatGPT memory work across chats?

Yes, when memory is on and the fact was saved. It is not a full transcript; it stores a short list of facts and preferences, so details that were never saved can still be missing.

Why does it forget within a long conversation?

The thread has outgrown the context window, so older messages are dropped or summarised. Paste a short project brief at the start, or start a new thread with the brief, to keep the important details in front of it.

Can I move ChatGPT memory to another app?

Not automatically. Keep a portable project brief in a document and paste it into the new app. Some apps store memory across AIs, but the document is still the backup.

Keep reading

· 5 min read

What is an AI chat app with memory?

An AI chat app with memory keeps context between conversations instead of starting from zero. What that changes, and what to check before you rely on it.

← All writing