The Play-Doh Problem
When I was a kid, I had four colours of Play-Doh: blue, yellow, red, and white. For the first day or two, I could make anything. A yellow sun. A blue ocean. Red flowers with white centres. Infinite possibilities.
A few weeks later, it was all one brownish-grey lump.
Your chatbot works the same way. You start fresh. Clean slate. The responses feel sharp and on-target. Then you have a hundred conversations. You add memories. You tweak instructions. You work on five different projects across three months. And slowly, without noticing, everything starts feeling a little… off.
Here’s the uncomfortable truth: more context does not mean better context.
The Problem: Context Drift
Here’s the thing about chatbots: they don’t prioritize context. They accumulate it.
Every conversation you’ve ever had sits there in the background. Old project notes. Half-formed ideas. Instructions you wrote when you were working on something completely different. Preferences that made sense in September but are irrelevant in December. The chatbot doesn’t know which pieces matter right now. It just keeps mixing them all together.
You start noticing odd assumptions. It brings up topics you haven’t thought about in weeks. It repeats ideas you’ve already moved past. The responses aren’t wrong, exactly. They’re just slightly sideways.
It’s like having a conversation with someone who never forgets anything, but also never checks what’s still relevant. Polite. Earnest. Slightly off.
Come to think of it, most of my friend group passes this description.
The Hidden Cost: Mental Overhead
There’s another problem that’s harder to name.
You know there’s a great insight buried somewhere in your chat history. A breakthrough idea. A perfect phrasing of something you’ve been trying to explain. But you don’t remember which conversation it was in. Or which tool you were using. Or what week that was. Or where you left your glasses. Dammit.
So you don’t go looking for it. You just vaguely remember that it exists, and that makes you trust the tool a little bit less. The friction is small, but it’s constant. And over time, it quietly undermines the whole thing.
None of this is dramatic. It’s just drag.
The Fix: A Quarterly Chatbot Reset
This isn’t about optimization or productivity theatre. It’s digital hygiene.
Once a quarter, you’re going to reset your chatbot. Clean slate. Fresh colours. Twenty to thirty minutes. That’s it.
Think of it like cleaning your desk at the end of term. You’re not throwing away your brain. You’re clearing space so it can work again.
The Checklist
Before we start: nothing here is irreversible. You’re going to back things up before deleting anything. And you only need to do the parts that feel accessible.
If half of this feels advanced, that’s fine. Do the parts you can.
Export or Archive Chats
Best case: export everything. Most tools let you download your full chat history as a file. Do that.
If export isn’t an option, archive old conversations instead of deleting them outright. If neither of those work, skim your recent and high-value chats. Copy out anything that matters.
You’re not trying to save everything. Just the good stuff.
The point is this: your chat history is searchable and reusable once it’s out of the chatbot. Inside the chatbot, it’s just noise.
Delete the Conversations
Quick reminder: you already backed things up.
Now delete them.
Deleting chats isn’t about loss. It’s about clarity. You’re resetting the conversational context so the chatbot meets you where you are now, not where you were during midterms or that one weird committee sprint in October.
Less hoarding. More signal.
Clean Up Chatbot Memory
Most chatbots now have a memory feature. It stores facts about you so it doesn’t have to ask again.
That’s useful. But it also piles up fast.
Go through your memory settings. Remove anything that’s outdated. Old job titles. Preferences for projects that are finished. One-off requests that got permanently stored.
Scan it. Remove anything that no longer fits. Keep only what’s durable.
Refresh Custom Instructions
Custom instructions tell the chatbot how to act. Tone, style, format, approach.
They age quickly. Like milk.
Go read yours. Does it still match how you actually work? Are there instructions in there from a project you finished months ago?
Your needs change. Your tolerance for verbosity changes. Your workflow changes. That carefully crafted instruction from last year might now be part of the problem.
If it feels stale, rewrite it. And here’s a trick: use the chatbot itself to help. Paste in your current instructions and ask it to suggest a refresh based on how you’ve been using it lately.
Instructions vs Memory (Important)
This is where people get confused.
Instructions are about how you want the chatbot to act. Tone, structure, style.
Memory is about what you want the chatbot to know about you. Your role, your projects, your preferences.
Mixing these up is like telling someone your life story when they ask for directions. Keep them separate.
Get Important Insights Out of the Chatbot
Here’s the trap: it’s easy to do all your thinking inside the chat tool. Brainstorm there. Draft there. Refine there. It feels efficient.
Long-term, it’s a bad habit.
You’re building knowledge inside a tool you don’t control. Chats are great for thinking. Terrible for storage. And when you need that knowledge later, you have to dig through chat history to find it. That’s friction you don’t need.
When you find something good—an insight, a draft, a framework—move it out.
Apple Notes. OneNote. Google Docs. Notion. Wherever you actually keep things.
Personally, I use Markdown files in VS Code because it’s fun and it makes me feel like a hacker. Your setup doesn’t need to be complicated. It just needs to be yours.
Ask yourself: what would this look like if it was fun?
Because if it’s not at least a little fun, you won’t do it.
Make It a Habit
Do this once a quarter.
Tie it to something you already do. Term changes. Project resets. Budget cycles. Whatever marks time in your world.
Remember the Play-Doh. Fresh colours lead to better creations. Your chatbot works the same way.
Most major tools—ChatGPT, Copilot, Claude, Gemini—have settings for chat history, memory, and instructions. The exact location varies, but they’re all there. Poke around. You’ll find them.
Too much context turns everything brown.
A reset brings the colours back.
