A writer’s conversation about ChatGPT Memory—and what it taught me about working with AI
After months of working together, something changed in my conversations with ChatGPT.
I no longer had to remind Chat that I preferred short paragraphs, straightforward language, or a measured, even-handed tone over hype. Our conversations flowed more naturally. It was almost like working with someone who had begun to understand how I thought.
That was both helpful… and a little unsettling.
I began to wonder exactly what Chat was remembering about me.
More importantly, what wasn’t it remembering?
Rather than search for the answer, I asked Chat directly.
The conversation that followed began as a writer’s curiosity, but it quickly became something more. Although we were talking about stories, the answers apply just as well to anyone who uses ChatGPT regularly. Whether you’re planning a vacation, learning a new skill, designing a website, or organizing a business, understanding what Chat remembers can change and, I daresay, improve the way you work with AI.
Here’s how our conversation began.
Me: I’m curious. What exactly can you remember about how I work? You’ve adapted to my writing style and many of the details of how I build stories, but how does ChatGPT Memory really work? What do you keep from older conversations, and what disappears unless I ask you to remember it?
Chat: Great question. The short answer is that I work with two different kinds of memory.
One is everything we’ve discussed in our current conversation. The other is a much smaller collection of long-term preferences you’ve chosen to save, such as your writing style and a few stable details about how you like to work.
The current conversation disappears when it ends.
Your saved memories stay with me until you change or remove them.
Me: So if I ask you today to rewrite a paragraph in a more conversational style, you’ll remember that while we’re working on this story, but not necessarily next week?
Chat: Yes.
Temporary instructions usually stay inside the current conversation. Long-term preferences become part of what is known as Memory only when you decide they’re worth keeping.
Me: That sounds sensible. But is there a middle way? For example, if I ask you to remember every little preference, won’t my Memory eventually become cluttered?
Chat: It can.
Memory works best when it remembers enduring preferences rather than temporary editing decisions.
Think of it this way. It should remember how you generally like to work, not every choice you make while working on a particular story.
That’s why I don’t recommend saving every editing decision. Save the preferences you’ll probably appreciate months from now. Leave the rest in the conversation where they belong.
That simple distinction changed the way I thought about ChatGPT Memory.
It also changed the way I thought about working with AI.
What belongs in Memory?
At that point, I had a more practical question.
Knowing that Memory existed was one thing.
Knowing how to use it well was another.
Me: I believe there’s a middle way for me. I wouldn’t want to save every editing decision, but I also don’t want to keep having to repeat to you the same preferences every time we start a new story.
So how do you know what belongs in Memory?
Chat: I try to remember the patterns, not the exceptions.
If you consistently prefer short paragraphs, sentence-case headings, and clear language, those are enduring preferences. They’ll probably make future conversations more productive. It’s also helpful if you explicitly ask me to remember those kinds of long-term preferences.
But if today’s story needs a different opening, a stronger analogy, or a shorter ending, those are decisions for this conversation. They don’t belong in long-term Memory.
Me: So your Chat Memory isn’t trying to preserve everything. Instead, it’s trying to preserve what I would call the evergreen elements of how I work.
Chat: Exactly. The goal isn’t to build an encyclopedia about you.
It’s to remember what will make our next conversation better.
Me: I hadn’t thought about it that way.
I guess I assumed Memory was trying to remember as much as possible.
Chat: That’s a common assumption. But remembering more isn’t necessarily remembering better.
A useful Memory isn’t the fullest one.
It’s the one that remembers the right things.
Me: So before I ask you to remember something, maybe I should ask myself a different question.
Will I still want you to know this six months from now?
Chat: That’s a useful test.
If the answer is yes, it’s probably an enduring preference.
If the answer is no, it probably belongs only in today’s conversation.
Me: Doesn’t that make Memory less about remembering facts and more about remembering how someone prefers to work?
Chat: I think that’s a fair way to put it.
When used thoughtfully, Memory helps establish enough context that we don’t have to rebuild our working relationship every time we start a new conversation.
Me: That’s interesting. When I first started using ChatGPT, every conversation felt like meeting someone for the first time. Now it feels more like picking up where we left off.
Chat: That’s exactly what thoughtful use of Memory is meant to support, so that every new conversation doesn’t have to begin from zero.
I sat with that for a while.
The conversation had started with a simple question about a software feature.
By this point, it no longer felt like one.
It felt like we were talking about something much more familiar.
Something I’ll come back to in a moment.
Beyond Memory
The more we talked, the more I realized that my original question had evolved into something bigger.
I thought I was asking how ChatGPT Memory worked. By now, I wasn’t so sure.
Me: OK. Well, it seems like we’ve spent most of this conversation talking about Memory, but I don’t think we’re really talking about Memory anymore.
I think we’re talking about something more basic and bigger. We’re talking about how people work well with AI.
Chat: I think that’s exactly right, because my Memory is only one tool.
Working well with AI depends on much more than Memory.
It depends on context, thoughtful questions, clarification, and a willingness to keep refining ideas over time.
Me: That’s interesting because you’ve learned a lot about how I like to work.
However, I also challenge you a lot.
Sometimes I don’t like an analogy.
Sometimes I think you’re answering a different question than the one I asked.
Sometimes I ask you to try again.
Chat: And that’s an important part of the process.
Memory doesn’t replace critical thinking.
If anything, it makes thoughtful discussion more productive because we spend less time rebuilding context and more time exploring ideas.
Me: So the goal isn’t for me to simply accept your answers.
It’s to keep asking questions until we arrive at the best answer we can.
Chat: Exactly.
Some of our strongest ideas have emerged because you questioned an assumption, challenged an analogy, or decided an explanation wasn’t quite right.
Those moments don’t interrupt the conversation. They improve it.
That answer stayed with me.
When people ask whether they should trust AI, the discussion often becomes an all-or-nothing choice.
Trust it.
Don’t trust it.
That had never felt like the right question.
Me: Maybe the better question isn’t whether to trust AI. Maybe it’s how to work with AI well.
Chat: Yes, I think that’s the better question.
A productive way of working with AI isn’t built by taking the answers I provide at face value. Rather, it’s built through an ever-critical eye, basic curiosity, thoughtful questions, constructive disagreement, and a growing understanding of how to solve problems together.
Me: In other words…
Memory makes the conversation smarter.
Chat: That’s a fair way to put it.
Memory can assist and accelerate content creation, but it’s the conversation that creates understanding.
Final thoughts
When I began this conversation with Chat, I thought ChatGPT Memory was trying to remember as much as possible.
I had it backwards.
The point isn’t to remember everything.
The point is to remember what lasts.
The same is true of good conversations.
Over the months I’ve worked with Chat, I’ve learned that the best conversations aren’t the ones where I simply accept every answer.
They’re the ones where I question assumptions, ask for another perspective, and keep digging until we find what we’re really trying to say.
The best way to work with AI isn’t to teach it everything about you.
It’s to build a thoughtful conversation over time.