Three Years In, Talking to AI All Day — Here's the Mistake Almost Everyone Makes

Three years ago I started talking to an AI, and I mostly haven't stopped since. Most days, morning to night — more than I talk to some actual people, honestly. So take this as a confession with a payoff attached: after an unreasonable number of hours, what actually matters turns out to be small. A handful of moves. No tricks, no prompts to copy. Just what I'd tell you if you asked why this works so much better for me than it does for you.

It starts with the one mistake almost everyone makes, the same one I made for longer than I'd admit. You compose a careful little question, trim it, make it sound smart, get it exactly right before you hit enter. Then you read the answer, decide it's kind of generic, and close the tab. From that one round you conclude the thing is shallow. It wasn't shallow. You stopped.

That careful question is the tell. It trains you to treat AI like a search box — one question in, one answer out, done. A search box hands you a fact. This is closer to the smartest, most patient person you've ever met, who knows almost everything and never gets bored of you. You don't search a person like that. You talk to them. Here's how.

Talk, don't type. Use the microphone. It's not that voice is magic. It's that typing makes every round cost something — you sit up, find the words, punctuate the thing — so after one answer you quit. Talking makes the back-and-forth free, and the back-and-forth is the entire game. People give up after one reply because round two felt like work. Take the work out and you stay.

Never stop at the first answer. That first reply is the AI guessing from one cold sentence, with no idea who you are or what you actually want. Of course it's generic — you'd be generic too. The good stuff shows up in the third round, the fifth, the one right after you said "no, not quite" a second time and it finally understands what you meant. You don't walk in with clarity and hand it over. Clarity gets made in there, over a few rounds.

Be curious, not just corrective. Most people only speak up when the AI gets something wrong — they treat it like a vending machine that occasionally jams. Try treating it like someone genuinely interesting instead. Ask why. Pull the thread. Chase the part of the answer that surprised you instead of moving on. I do this on walks all the time — something lands a little sideways, I go "wait, what do you mean by that," and four questions later I'm somewhere I'd never have gotten on my own.

Demand truth, not a pat on the head. Out of the box, it's built to please you — to agree, to validate, to smooth the rough edge off a hard thing, because agreeable feels good and feeling good keeps you around. You have to tell it, more than once, that you want it to push on your thinking, ground things in evidence, go for what's true over what's comfortable. It can do that. It just won't volunteer it. (That one's big enough to be its own piece — but ask now and you'll already get a sharper machine.)

When it's dumb, give it the reason, not the rule. A dumb answer almost never means it doesn't know the thing. It means it knows too much and can't tell which part you care about, or it's misread what you're after. It's not stupid — it's flooded. And here's the move most people miss: don't bark a tighter instruction at it. Tell it why. Not "make it shorter" but "this is going to a client who's already annoyed with me, and I need them to feel heard" — then let it work out what shorter means. Hand it the goal, not the order, and it reasons its way to something better than any rule you'd have written. It's smarter when you give it the reason. Watch it sharpen the second you do.

I'll be honest about my part in this, because the easy version of the story would be a lie. I never had to learn any of it. I'm a talker — I'll blame the ADHD, but the truth is I just don't shut up. So the conversation came easy to me, and I won't pretend I cracked some discipline you have to grind toward. You don't. If it doesn't come naturally, you do it on purpose.

I talk to my system constantly now — over coffee, working out a hard email on a walk. Three years ago I started by recording myself an hour at a stretch and handing over the transcripts. I basically don't type to it anymore.

So tonight, press the mic. Talk for thirty seconds about something you're actually stuck on. Then don't stop — push back, get curious, go five rounds before you close the tab. That's where it stops being a search box and starts being a conversation worth having.