Every time you start a new ChatGPT conversation without memory enabled, you are starting from zero. ChatGPT does not know your job, your communication style, your ongoing projects, the client you worked on last week, or the fact that you hate bullet points. You explain yourself again and again — spending the first exchange of every session re-establishing context that should already be there.
ChatGPT’s memory system and Custom Instructions exist to fix this. Memory stores what ChatGPT learns about you from conversations — your preferences, your context, your corrections. Custom Instructions let you front-load permanent context that applies to every conversation without waiting for memory to learn it. Together, they transform ChatGPT from a stateless question-answering system into something that actually knows who it is working with.
This guide covers both systems in full: how they work, how to set them up, how to build a personal AI profile that genuinely compounds across sessions, and the privacy controls that let you manage exactly what is retained.
🔗 This is Post #4 in the ChatGPT Unlocked series. Memory and Custom Instructions apply across all ChatGPT use cases. See ChatGPT for Writing and ChatGPT for Business for how these systems fit into specific professional workflows. Start with ChatGPT Masterclass 2026 if you are new.
The Two Systems: Memory vs. Custom Instructions
Before the setup, the distinction — because these are complementary, not redundant.
Custom Instructions are what YOU tell ChatGPT about yourself upfront. They are static, deliberate, and under your direct control. You write them once and they apply to every conversation. Think of them as your permanent briefing document to ChatGPT.
Memory is what ChatGPT learns about you through conversation. It is dynamic, accumulative, and happens automatically (when enabled). ChatGPT notices patterns, saves relevant facts, and uses them in future conversations. Think of it as ChatGPT building a profile of you from your interactions.
Neither replaces the other:
- Custom Instructions are under your control from day one — you do not have to wait for memory to develop
- Memory captures things you did not think to put in Custom Instructions — recurring preferences, corrections, evolving context
- Custom Instructions are for permanent, intentional setup; Memory is for organic learning
Both are available on Plus, Pro, Business, and Enterprise plans.
Custom Instructions: Your Permanent Briefing Document
How to Access and Set Custom Instructions
On desktop/browser:
- Click your profile icon (bottom left)
- Select Customize ChatGPT (or Custom Instructions depending on your app version)
- You see two text boxes
On mobile:
- Tap your profile icon
- Tap Customize ChatGPT
- Same two text boxes
The Two Custom Instruction Fields
Field 1: “What would you like ChatGPT to know about you to provide better responses?”
This is your permanent context. Everything you put here applies to every conversation, invisibly, before you type a single word.
Field 2: “How would you like ChatGPT to respond?”
This is your permanent behavior configuration. Format preferences, communication style, what to avoid, how to handle uncertainty.
Each field has a 1,500-character limit. Use them fully.
Building Your Custom Instructions: The Complete Template
Here is a complete custom instructions template you can adapt. Personalize every field — specificity is what makes this valuable.
Field 1 — Who you are and your context:
I am a [job title] at [company type, not name if sensitive].
I work primarily in [industry/domain].
My key responsibilities: [2-3 sentences].
I am currently working on: [current major projects or goals].
My technical level: [non-technical / mixed / technical in X domain].
My writing audience is usually: [who reads my work].
Tools I use: [key software you work in daily].
My time zone: [timezone] and I typically work [hours].
Key context: [anything else recurring that affects responses —
e.g., "I work with clients in Europe and Asia" or
"I manage a team of 5 and often need to communicate
decisions clearly to non-technical stakeholders."]
Field 2 — How you want responses:
Format: [Prose preferred / Use headers for long responses /
Tables only when comparing / etc.]
Length: [Match the question — short for quick tasks,
comprehensive for complex ones]
Tone: [Direct and professional / conversational / academic]
Avoid: [Bullet points for everything / excessive hedging /
generic openings like "Great question!" /
phrases I dislike: "leverage", "synergy", etc.]
When you don't know: Say so directly rather than guessing.
Feedback style: [When I share my work, be honest about
problems — don't soften critique.]
Iteration: [When I ask you to revise, make the changes
I specified — don't rewrite everything.]
Custom Instructions for Specific Roles
For writers and content creators:
Field 1: I write [type of content] for [audience].
My voice is [describe it]. I publish [frequency]
on [platforms]. I am currently working on [project].
Field 2: Match my voice — do not default to generic
professional. When I share drafts, critique the
specific weaknesses rather than praise first.
Never use em-dashes. Use sentence case for headings.
Maximum 150 words for first drafts unless specified.
For developers:
Field 1: I am a [backend/frontend/full-stack] developer
working primarily in [languages]. Current project: [brief].
Frameworks: [list]. My team's style: [coding conventions].
Field 2: Show code with inline comments explaining
non-obvious logic. Always include error handling.
When debugging, explain the root cause, not just the fix.
Flag security concerns explicitly. Prefer readable
over clever.
For managers and executives:
Field 1: I manage a team of [X] people in [function].
My company is [size/type]. I regularly communicate with
[C-suite / board / clients / etc.]. Key decisions I make
involve [areas].
Field 2: For communications, lead with the key point.
No fluff. When helping with people issues, give me
direct advice, not only "it depends." Flag if my
proposed approach has a common failure mode I should
anticipate.
Memory: The Learning System
Enabling and Configuring Memory
To enable memory:
- Click your profile → Settings (or Customize ChatGPT)
- Find Memory settings
- Toggle Memory on
Once enabled, ChatGPT will begin saving relevant information from conversations to a memory store. These memories surface automatically in future conversations where they are relevant.
Viewing Your Memories
To see exactly what ChatGPT has remembered:
- Profile → Settings → Personalization → Manage Memory OR
- In a conversation, type: “What do you remember about me?”
You will see a list of stored memories — facts, preferences, and context that ChatGPT has saved from your conversations.
Editing and Deleting Memories
In the Manage Memory view:
- Click any memory to see the full stored text
- Click the delete icon to remove specific memories
- Use Clear all memories to wipe the entire memory store
You are in full control of what persists. This is worth using actively — if ChatGPT has learned something incorrect or outdated, delete it explicitly.
Telling ChatGPT What to Remember
You can instruct ChatGPT to save specific things:
"Remember that I prefer responses without bullet points
for everyday questions."
"Remember that I'm launching a new product in Q3 and
everything I'm working on now relates to that."
"Please forget what you saved about my previous role —
I changed jobs last month."
ChatGPT will confirm what it has saved and what it has deleted.
Building Your Personal AI Profile Over Time
The most valuable state is when your Custom Instructions front-load the permanent context and Memory has accumulated the dynamic, evolving context. This is your personal AI profile — and it compounds over weeks and months of use.
Month 1: The Foundation
- Set up complete Custom Instructions using the template above
- Enable memory
- Begin telling ChatGPT to remember specific things that matter: recurring preferences, key project context, corrections to its default behavior
Month 2–3: The Refinement
- Review your memories regularly (weekly or bi-weekly)
- Delete outdated entries (completed projects, preferences that changed)
- Update Custom Instructions when your role or priorities shift
- Let memory accumulate without forcing it — the organic learning is the point
Month 3+: The Compound Effect
By month three, ChatGPT’s responses will be meaningfully different from a fresh-start conversation. It will know your communication style without being reminded. It will reference context from earlier work. It will avoid the things you have told it to avoid. The setup investment pays back across every future session.
The Memory Privacy Architecture
What memory stores
Memory saves text-based notes about you — facts, preferences, and context that ChatGPT judges relevant. It does not store entire conversations in memory — it extracts and saves the relevant elements.
Where memory is stored
Memory data is stored in your OpenAI account, associated with your user profile. It persists across devices when you use the same account.
What memory is NOT
Memory is not a verbatim transcript of your conversations. It is a curated set of notes that ChatGPT has generated from conversations. The underlying conversations themselves are stored separately (and subject to the data retention settings for your plan).
Privacy controls
- Disable memory entirely: Settings → Memory → Toggle off. Past memories are retained but no new memories are created.
- Use Temporary Chat: A temporary chat session is memory-exempt — nothing from it is saved to memory, it does not appear in conversation history, and it does not affect future sessions. Useful for sensitive queries.
- Delete individual memories: Manage Memory → Delete specific entries
- Wipe all memories: Manage Memory → Clear all memories
Memory on different plans
- Plus/Pro: Full memory features, data subject to standard OpenAI terms (opt out of training in Settings → Data Controls)
- Business/Enterprise: Memory features available, conversations not used for training by default
- Free/Go: Memory features limited or unavailable depending on current rollout
Memory vs. Custom GPTs for Persistent Context
A common question: if I need specific context for a recurring work type (client proposals, technical documentation, social media content), should I use memory + Custom Instructions or build a Custom GPT?
The answer depends on the use case:
Use memory + Custom Instructions when:
- The context applies broadly across all your ChatGPT use
- The context is personal and role-based (who you are, how you work)
- The context evolves organically over time
- You do not want to switch between tools for different tasks
Use a Custom GPT when:
- The context is highly specific to one type of task
- You want to share the configuration with teammates
- You want to add a specific knowledge base (documents) for that task type
- You want different behavior rules for this context vs. your general ChatGPT use
- The “tool” is something you want to return to as a named tool, not just an extension of your ChatGPT persona
The most powerful setup uses both: personal Custom Instructions + Memory for your general ChatGPT use, and targeted Custom GPTs for specific recurring task types. The Custom GPTs inherit your general context when they also have access to memory, but can layer additional task-specific instructions on top.
Advanced Techniques
The Memory Audit Prompt
Run this monthly:
Review everything you remember about me and tell me:
1. What is accurate and still current?
2. What seems outdated or may have changed?
3. What important context is missing that would
make your responses more useful?
This prompts ChatGPT to surface its memory store for your review and identify gaps — the context you have not told it that would change how it helps you.
The Context Document Method
For complex ongoing projects, create a context document in your conversation once and tell ChatGPT to remember it:
"Here is the complete context for Project X. Remember
all of this — I will reference this project frequently:
Project: [name]
Goal: [1 sentence]
Key stakeholders: [names and roles]
Current status: [where things are]
Key decisions made: [list]
Key constraints: [what we cannot do]
Next milestone: [date and deliverable]"
Reference it in future sessions: “This is for Project X — refer to your memory of the context.”
The Preference Training Loop
When ChatGPT produces something in a format or style you like:
"That response format is exactly right — remember this
is how I want [reports / analyses / emails / etc.]
structured."
When it produces something you do not like:
"Don't format it this way. [Explain what was wrong.]
Remember this preference."
Explicit preference training accelerates the memory learning significantly compared to waiting for ChatGPT to infer preferences organically.
Common Mistakes
Mistake 1: Setting Custom Instructions and never updating them
Your role, projects, and priorities change. Custom Instructions from six months ago reflect a past state. Review and update them every 1–3 months.
Mistake 2: Not auditing memory
Outdated memories create worse responses, not just neutral ones. A memory that says you are working on a project that ended three months ago actively misleads ChatGPT’s responses. Regular audits keep the memory store accurate.
Mistake 3: Over-specifying in Custom Instructions
Custom Instructions should capture what is persistently true and important — not every possible constraint for every possible task. Over-stuffed instructions conflict with each other and confuse more than they help. Keep them focused on the 10–15 most important, persistent facts.
Mistake 4: Not using Temporary Chat for sensitive queries
If you have a sensitive query — health, legal, financial, personal — that you do not want retained in memory or conversation history, use Temporary Chat mode. The session is ephemeral and leaves no memory trace.
Mistake 5: Treating memory as set-and-forget
Memory requires active management to stay accurate and useful. A five-minute monthly audit — review, delete outdated entries, update Custom Instructions — produces dramatically better results than passive accumulation.
Conclusion
Custom Instructions and Memory are the upgrade most ChatGPT users who do not have them do not know they are missing. The experience of using ChatGPT without them — re-explaining yourself every session, watching it revert to generic formats, getting responses calibrated to the wrong audience — is genuinely worse than the alternative.
The setup investment is small: 20 minutes to write thoughtful Custom Instructions, memory enabled, and a simple monthly audit habit. The compounding return is significant: every session benefits from context that was built in all previous sessions.
This is how ChatGPT stops being a tool you use and starts being a partner that works with you.
Your next step: Open ChatGPT → Customize ChatGPT → write your Custom Instructions using the template in this guide. Do not try to be perfect — write a first version in 15 minutes, then refine it over the next two weeks as you notice what is missing. Enable memory. Come back to it in a month and notice the difference.
📚 Continue the Series:
- ← Previous ChatGPT Advanced Voice Mode
- Next → ChatGPT for Writing: From Blog Posts to Books
- For teams ChatGPT for Business
- For specific workflows Custom GPTs: Build Your Own AI Tool Without Code
Last updated: May 2026. Memory features and Custom Instructions availability vary by plan and are updated by OpenAI. Verify current availability at help.openai.com.
