The Better Way to Use ChatGPT Memory (From Someone Who’s Broken It… Twice)

Last year, I shared a simple idea: create a “Lightbulb List” in ChatGPT to catch all your best ideas.

Think of it as a digital spark-catcher. A running list of insights, metaphors, project hooks, and half-baked brilliance you don’t want to lose. A creative archive, powered by AI.

The post struck a chord.

Since then, OpenAI has expanded what ChatGPT memory can do—and I’ve broken, rebuilt, and refined my own setup quite a bit in the past few months. Take 2 covers:

What I’ve learned.
What to watch out for.
And how to really make memory work for you.

What’s New with ChatGPT Memory

If you haven’t used it recently, ChatGPT memory got a major upgrade:

Full visibility: You can now view, edit, and delete memory entries inside Settings > Personalization > Manage Memory.
Integrated context: ChatGPT remembers who you are and what you care about across chats—and starts suggesting accordingly.
Personalized search: Web results can reflect what ChatGPT has learned about you (your preferences, location, etc.).
Temporary chats: You can go “off the record” anytime by turning memory off for a conversation.

It’s not perfect. But it’s powerful. And if you use it intentionally, it becomes more than a notepad. It becomes a creative partner.

Lessons from the Field (a.k.a. Mistakes I Made So You Don’t Have To)

1. Custom GPTs don’t always remember your lists

If you use a custom GPT (like one tailored to writing or brainstorming), it might not have access to your main memory by default.

Workaround: Start from the main ChatGPT interface, then invite the custom GPT into that thread. As of April 2025, this preserves memory access and context.

2. Be very careful when deleting memories

Once, I told ChatGPT to “remove this list” and it overwrote far more than I expected.

Lesson learned: Be precise. Review what it stores. Don’t assume it knows what you mean.

3. Yes, it can reconstruct memory—to a degree

Surprisingly, after wiping my memory twice, I asked ChatGPT to reconstruct what it remembered from our past chats.

The results? Not perfect. But good enough to rebuild the essentials.

4. Treat memory like a shared workspace

ChatGPT is documenting your personality in real-time. You have the right to:

Audit what’s saved
Replace entries that feel off
Refine how it sees you

Pro tip: Don’t just delete something vague. Tell it exactly what you want it to remember instead.

5. Not all lists are created equal

OpenAI recommends memory entries between 1,000 and 4,000 characters. That means:

Great: Long-running ideas, frameworks, writing themes, core beliefs
Bad: Daily to-do lists, ultra-dynamic content

Think “creative trail markers,” not “scratchpad.”

How to Set Up Your Own Lightbulb List

Want to try it? Here’s a starter prompt to create your own memory-powered Lightbulb List:

[STARTER PROMPT]

Create a Lightbulb List using the following prompt:

# Lightbulb List 1 #

## Purpose: ##

The lightbulb list is used by the user as a conceptual and creative capture mechanism. Entries often originate from reflections, journal entries, or writing drafts and may be linked later to book sections, LinkedIn posts, or frameworks I may use. We call these “lightbulbs”.

## Lightbulbs ##

Lightbulb 1
Lightbulb 2
Lightbulb 3

# End #

Bonus: You can create a second memory entry with rules or instructions for how ChatGPT should manage the list. That prompt might look like this:

[RULES PROMPT]

# Lightbulb List Instructions #

##1 – When user provides a new lightbulb entry, always append it to the single unified “Lightbulb List” without interpretation, expansion, or rewording—unless explicitly requested.

## 2 – Warn user with then list risks exceeding or will exceed 4000 characters and should be broken into secondary list or cleaned up.

## 3 – Only expand to a secondary list after consulting with user and receiving permission. If additional lists are needed, each list will have a sequential numbering scheme such as “Lightbulb List 2”.

## 4 – Instructions to remove items from the “Lightbulb List” should not affect any other memory items being removed. All memory items should be managed separately from this list.

# End #

Final Thought: Your Brain, Backed Up

The real value of memory isn’t automation. It’s augmentation.

You get a second brain that grows with you. One that remembers what matters. One that helps you reconnect with the ideas you forgot you had.

Just make sure it’s remembering the right things.

What’s the smartest way you’ve used ChatGPT memory? Or the biggest mistake you wish others knew to avoid?

→ Drop your tip—or horror story—in the comments below.