Lesson 05 · ChatGPT Mastery Pro ~10 min read Instructions + Memory + pruning

Memory & instructions: teach ChatGPT to know you.

ChatGPT has two systems for remembering things about you: Custom Instructions (you write them) and Memory (it learns automatically). Used well, they make ChatGPT feel like it knows you. Used badly, they make it worse. This lesson covers what belongs where, the anti-patterns to avoid, and how to prune Memory when it goes wrong.

The mental model

Custom Instructions are the contract. Memory is the observation log.

Custom Instructions are what you tell ChatGPT explicitly about you and your preferences. Memory is what ChatGPT learns from your conversations and stores automatically. Both shape future responses — but they work differently and need different management.

Workflow 01 Write Custom Instructions that actually work

1

The 4-section template

Custom Instructions live in Settings → Personalization. They have two boxes: 'What would you like ChatGPT to know about you' and 'How would you like ChatGPT to respond.' Fill both.

The prompt that works

TemplateBOX 1 (about you): Role: [your job]. Industry: [your field]. Main use: [what you do with ChatGPT]. BOX 2 (how to respond): - Concise unless I ask for depth - Conversational, not stiff - No 'as an AI' disclaimers - When uncertain, say so - Never invent statistics - Use contractions

Best use cases

  • Anyone using ChatGPT weekly+
  • Power users with consistent style preferences
  • Professional users where consistency matters
Don't put confidential info (real customer names, financial details, secrets) in Custom Instructions — they're stored.
Time savings: Every conversation starts pre-tuned to your voice.

Workflow 02 Add to Memory deliberately

2

Tell ChatGPT to remember things explicitly

ChatGPT learns some things automatically (your name, work style). You can also tell it explicitly: 'Remember that...'

The prompt that works

Adding-to-memory promptsRemember that I'm a sales rep selling to mid-market healthcare companies. Remember that I prefer code examples in TypeScript, not JavaScript. Remember that I write a weekly newsletter, so when I brainstorm topics, they should be 800-1200 word essays.

Best use cases

  • Recurring context you'd otherwise re-explain
  • Specific preferences (tools, formats, conventions)
  • Long-term goals that should shape recommendations
Memory has a soft limit — when it fills up, old items get forgotten. Don't rely on it for critical context. Use Custom Instructions for that.
Time savings: Re-explaining recurring context: gone.

Workflow 03 Prune memory when it goes wrong

3

Check, edit, delete what's stored

Sometimes ChatGPT remembers things you didn't intend, or its summaries of you are wrong. Audit periodically.

The prompt that works

Pruning patternSettings → Personalization → Memory → review all entries → delete anything wrong, outdated, or you don't want stored. Do this monthly.

Best use cases

  • When ChatGPT's responses feel off-target
  • After a major life or job change
  • If you notice it making assumptions about you that aren't true
  • Privacy hygiene (regularly check what's stored)
Don't delete useful Memory items just because they feel personal — that's the point of Memory. Delete what's WRONG or OUTDATED, not just what's personal.
Time savings: Stale-Memory problems: prevented.

Workflow 04 When Memory should be OFF

4

The cases where you don't want it on

Memory is on by default. There are real cases where you should turn it off.

The prompt that works

When to disable1. Sharing your account (family, team) — memories get tangled. 2. Highly sensitive sessions (anything PHI, confidential financial, NDA-covered). 3. Testing prompts where you want a clean baseline. 4. If you find yourself spending more time pruning than benefiting.

Best use cases

  • Shared accounts
  • Sensitive work sessions
  • Prompt engineering tests
  • Anyone with privacy concerns
Memory off doesn't disable Custom Instructions. Those still apply. Memory only refers to the auto-learned items.
Time savings: Privacy and clarity when it matters.

Final challenge: audit and write tonight

Two tasks, 15 min total. (1) Open Settings → Personalization → Memory. Read everything stored about you. Delete what's wrong, redundant, or shouldn't be there. (2) Rewrite your Custom Instructions using the 4-section template above. Save. Test the difference on your next 3 prompts.

Back to ChatGPT Mastery → See pricing →

What you can do now

  • Write Custom Instructions using the structured 2-box template
  • Add to Memory deliberately with 'Remember that...' prompts
  • Audit and prune Memory monthly to keep it accurate
  • Know when to turn Memory off (shared accounts, sensitive work)
  • Recognize when ChatGPT's responses feel off — and fix it via these systems
Pro+
Up next in ChatGPT Mastery

Lesson 6 · GPT-5 agents: workflows that run themselves

Setting up agents to handle multi-step tasks. The use cases worth automating, the oversight patterns that prevent disasters, the cost-control playbook. See pricing →