How to Save Context in ChatGPT
Custom instructions, memory, and portable context - what works

ChatGPT forgets everything when you close a session. New conversation, blank slate - it doesn't know your role, your projects, or how you communicate. You start from zero every time.
There are ways around this. Here's what actually works, what doesn't, and where the limits are.
Option 1: Custom Instructions
The highest-leverage change most ChatGPT users never make.
Custom Instructions let you define standing information that applies to every conversation - who you are, what you work on, how you want responses structured. Set it once, ChatGPT includes it automatically in every session.
How to set it up:
- Click your profile picture → Settings
- Personalization → Custom Instructions
- Fill in both fields: "What would you like ChatGPT to know about you?" and "How would you like ChatGPT to respond?"
- Save
What to put there:
- Your role, company, industry
- Current focus areas or active projects
- Communication style ("direct, no preamble," "numbered lists over prose")
- Standing constraints that matter for most tasks
The limit: static. Doesn't update automatically. Role changes, focus shifts - you update it manually. You get roughly 4,500 characters across both fields. Enough for the essentials, not enough for deep context.
Option 2: Saved Memories
ChatGPT stores specific facts across sessions. Unlike Custom Instructions, memories build over time - automatically from conversations, or when you explicitly ask it to remember something.
How it works:
- Tell it to remember something: "Remember that I'm building a B2B SaaS product targeting mid-market logistics companies." You'll see "Memory updated" below the response.
- ChatGPT also saves things it considers relevant from conversations - without asking.
- Memories persist until deleted.
How to manage them: Settings → Personalization → Manage Memories. Review, edit, delete. Worth doing periodically - they accumulate noise.
What to use it for: role and org context, active projects (top 2-3 with one-sentence descriptions), communication preferences, recurring constraints.
The limits: roughly 1,500-1,750 words of storage. ChatGPT decides what to save - which means it'll sometimes save your cat's name and forget your product strategy. Two documented memory-wipe incidents (February 2025 and November 2025) wiped memories without recovery options. Don't rely on it for anything critical.
Option 3: Projects
Projects group related conversations, files, and instructions into a workspace. Inside a project, ChatGPT can reference other chats and uploaded documents - useful for longer-running work.
When to use it: ongoing work with shared background - a product you're building, a client engagement, a research project. Create a project, add documents, set project-specific instructions.
The limit: ChatGPT-only. Context lives inside the workspace. Doesn't transfer anywhere.
What doesn't work
Chat history alone: ChatGPT references past conversations through summaries, not full transcripts. Details drop out as summaries update. Not reliable for anything that matters.
Pasting context into every session: works, but requires discipline. Only as current as the last time you updated and pasted it.
Custom GPTs: don't solve the memory problem. Context is static and manually maintained, still resets within context window limits.
The real problem
All three options store your context inside ChatGPT. Fine when ChatGPT is your only tool.
Most people use more than one. Claude for writing, Cursor for code, ChatGPT for research. Every tool starts from zero. The context you built in ChatGPT doesn't exist in Claude. The preferences you set in Cursor don't carry anywhere else.
That's not a configuration problem - it's structural. Each platform owns its memory.
The alternative: context that lives outside any platform. A structured profile you own, served to whichever tools you authorize. Connect a new tool and it already knows who you are.
→ How that works: What Is Personal Context for AI?
→ How to build a portable context profile: How to Build Personal Context for AI