ChatGPT Memory: The Complete User Manual

Memory is one of ChatGPT's most underrated features. Turn it on, and ChatGPT becomes more like your personal assistant over time.

1. Enabling Memory

1. Open ChatGPT (web or desktop)
2. Go to Settings → Personalization
3. Find the "Memory" toggle and switch it on
4. You should see "Memory is now enabled"

2. Two Sources of Memory

Automatic learning

The model writes entries it thinks are worth remembering, for example:

User: I'm Felix, a product manager who uses Next.js
(Model auto-saves: user is Felix, PM, tech stack Next.js)

Explicit additions

You tell ChatGPT to remember things directly:

"Remember that all my code examples use TypeScript, not JavaScript"
"Remember I'm building a knowledge platform called GPTMap"
"Remember my writing style is concise, technical, with code samples"

3. View and Manage Memory

Settings → Personalization → Manage Memory lists every entry. You can:

• Delete single entries
• Clear everything in one shot ("Clear all memories")
• Disable Memory without deleting (toggle off)

4. 12 High-value Use Cases

Use case 1 — Tech stack preferences

Tell ChatGPT your stack and every code sample comes out aligned.

"I use TypeScript + Next.js 15 + PostgreSQL + Tailwind + shadcn/ui. Keep code samples consistent."

Use case 2 — Writing style

For any long-running writing workflow.

"My writing style: concise, prefers short sentences, no filler, audience is technical practitioners."

Use case 3 — Professional identity

Helps the model produce output aligned with your work.

"I run an AI knowledge platform for technical developers and product managers."

Use case 4 — Project context

"I'm running an OpenAI knowledge site — the main brand is GPTMap; sister sites are OpenCode (AI programming) and SkillMap (learning paths)."

Use case 5 — Audience info

"My readers are 25-40 year-old AI practitioners in China — familiar with tech, not necessarily with math."

Use case 6 — Dietary restrictions

"I'm vegetarian — no meat."

Use case 7 — Family

"I have a 3-year-old daughter named Lily."

Use case 8 — Common tools

"I take notes in Obsidian, track tasks in Linear, design in Figma."

Use case 9 — Things to avoid

"Avoid jQuery; never mention Company X."

Use case 10 — Tone & culture

"Reply in Chinese; keep English technical terms untranslated."

Use case 11 — Decision style

"Be data-driven when making decisions — show numbers before conclusions."

Use case 12 — Time zone

"I'm in UTC+8; default all dates to Beijing time."

5. Memory vs Custom Instructions

Custom Instructions are general rules that apply to every chat. Memory holds concrete facts about you — facts you can delete one by one without breaking other sessions.

Best practice:

• Custom Instructions: rules that never change (output style, persona)
• Memory: facts that evolve (your projects, current identity)
• Memory can be cleared without affecting Custom Instructions

6. Pairing Memory with Projects

Projects are ChatGPT's "workspace" feature. Each Project can have its own Instructions, files, and conversation thread.

• Project Instructions: project-level rules ("This is a React component library project")
• Cross-project Memory: who you are ("I'm a front-end architect")
• Project Files: relevant docs (auto-RAG)

7. Privacy Notes

• Memory is stored on OpenAI servers, bound to your account
• Deleting your account deletes your Memory
• Memory is never used to train any model
• Team / Enterprise Memory is private per user
• Don't store highly sensitive data (passwords, ID numbers) in Memory

8. Troubleshooting

• ChatGPT doesn't remember → check Memory is enabled
• Want to forget a Memory → Settings → Manage Memory → delete it
• Want to wipe everything → Settings → Clear all memories
• Conflicting Memory entries → delete the older one