I’ve been glued to ChatGPT for six months now. I’ve pushed it, broken it, confused it, and even had another model—Grok 4—tap out and tell me it was “worn out” and wanted a different topic.
That’s when I realized something:
The most important rule I’ve learned is simple: new topic = new chat. Every time.
If you try to cram three different subjects into one long AI thread, it will bite you later. You’ll go back looking for “that one post about birthright citizenship” and find it buried between Rosie O’Donnell, Trump, and a WordPress troubleshooting session from three days earlier.
Starting a fresh chat for each subject does three things:
- Keeps the “memory” clean so the AI doesn’t drift from politics into golf into Philippine malls all in one breath.
- Makes it easy to find later when you want to write part two, update a rant, or pull a quote.
- Prevents mushy answers when the model is trying to juggle ten old contexts at once.
That’s lesson #1. But it’s not the only thing I’ve picked up in the bunker.
Give the AI a mission brief, not a vague order
Bad order:
“Write an article about AI.”
Good order:
“You’re writing in my Chatrodamus voice: USMC ’68–’72, conservative blogger, talking to normal people who don’t have time for academic mumbo-jumbo. Plain English, a little snark, no weasel words. Write about what I’ve learned after six months of using ChatGPT.”
Same machine. Completely different result.
When you give role + audience + tone up front, the output sounds a whole lot more like you and a whole lot less like a corporate PowerPoint.
Work in passes, not perfection
My workflow now looks like this:
- First pass: Rant. Just get the thoughts out.
- Second pass: Polish and tighten.
- Third pass: Meta description, tags, slug, and a featured image idea.
The machine doesn’t get tired of this. It doesn’t complain. It just keeps cranking. That beats staring at a blank screen waiting for “inspiration” while cable news rots your brain.
Use AI as a partner, not an oracle
AI is a tool, not a prophet.
It gets things wrong. It reflects the bias and blind spots baked into the data it was trained on. Sometimes it’ll sound confident and still be dead wrong.
Your job is to:
- Think.
- Check what matters.
- Decide what you actually believe.
AI can draft a post. It can’t tell you what you stand for. That part is on you.
Name and reuse your “voices”
Over time, I’ve accidentally built a little cast of characters:
- Chatrodamus – prophetic/snarky political voice.
- Joe Everyman – plain-language explain-it-like-I’m-busy voice.
- Digital Rangemaster / 2ndA – gun/Marine persona.
- My Philippines Retirement – friendly expat uncle mode.
Now, instead of re-explaining all that, I just say:
“Write this in my Joe Everyman style,”
or
“Give me full Chatrodamus snark.”
The AI remembers the tone, and we move faster.
When a chat starts feeling “off,” cut it loose
Long threads can get weird. The answers start to blur. It forgets what we were doing three twists ago. That’s not evil—it’s just how these systems work.
When that happens, I don’t argue with it. I just:
- Start a fresh chat.
- Paste in the latest version of the post or notes I care about.
- Give it a clear mission brief and keep moving.
New subject, new chat. Old subject with too much baggage? Also new chat.
Bottom line from the bunker
Six months in the AI trenches, this is the best advice I’ve got:
- New topic = new chat.
- Give the machine a clear mission.
- Work in passes.
- Treat AI like a tool, not a religion.
- Keep control of your own voice.
Do that, and you won’t just “use AI.” You’ll put it to work—on your terms, from your bunker, with your values at the center.
Bunker Notice
If you made it this far, you’re bunker material. Join the Bunker Briefing—my unfiltered monthly dispatch from Bunker #69.