When to Start a New AI Chat
- Rich Washburn
- Jun 4
- 2 min read
The Magnetic Field Metaphor—and a Few Pro Tips You Didn’t Know You Needed
Imagine one of those glass tables filled with iron filings—the kind you played with as a kid, dragging a magnet underneath and watching the metal dance. That’s your AI conversation. Every prompt, every response, every idea is a tiny magnet shaping the terrain. Over time, those filings settle into peaks and valleys—patterns made from context, memory, and momentum.
Now, picture introducing a giant neodymium magnet a couple feet away. It’s not even touching the glass, but the field is strong enough to tug the whole thing in its direction. Your peaks are still there—but they’re leaning. Distorted. Nudged by this invisible force.
That big magnet?That’s the bias of the conversation itself.
The longer a thread runs, the more that invisible context starts pulling your outputs off-center. Prompts get interpreted through the lens of what’s already been said. Creativity narrows. The model stops exploring and starts echoing.
It’s the “snake-eating-its-own-tail” effect—Only now the snake is paraphrasing your last three prompts and suggesting bullet points.
And that’s when you bail.
When It’s Time to Whack the Chat
No red alert, no countdown. Just a vibe shift.When you’re noticing:
Looping answers
Responses that lean too hard on earlier context
Helpful suggestions that feel... weirdly familiar
That’s the signal. Save the gold. Start clean. Don’t look back.
But here’s the move most people miss:You don’t wait for it to get weird. You plan for it upfront.
Pro Tip (Actually... Pro Tip Squared): Set Up a Scratchpad
If you're working on anything more than a one-off prompt, start a scratchpad—Google Doc, Notion, whatever.
Drop in:
🔹 Killer prompts
🔹 Solid responses
🔹 Notes, insights, ideas worth keeping
🔹 Anything you’d hate to scroll back 200 turns to find again
That way, when the thread goes sideways (and it will), you’re not digging through digital spaghetti. You’ve already got the good stuff sorted and ready to go.
This one tip alone can save you hours—and make your outputs better.
FIELD MANUAL: HOW TO USE THE SCRATCHPAD STRATEGICALLY
Use it for:
Building clean context for a new thread
Prepping a custom GPT prompt
Documenting the AI’s best ideas for later use
Collaborating with teammates (especially on multi-phase projects)
💡 Bonus Move:
If the thread is salvageable but bloated, select the whole thing, paste it into Gemini or another large-context model, and say:
"Strip out the fluff. Summarize the core ideas. Give me a reusable prompt and a knowledge base."
Boom. You just turned 5,000 tokens of chaos into a streamlined prompt + training doc.
Then? Use that to spin up your own custom GPT. That scratchpad? It just became your blueprint.
🧠 TL;DR
Threads build gravity. Eventually, they warp your results.
If the conversation starts echoing, it’s time to bounce.
Set up a scratchpad early if you’re going deep.
Use it to reboot clean threads, build custom GPTs, and share insights.
Don’t just use AI—architect it.
Comments