AI knows a startling amount, and it works fast. That combination makes it very tempting to think, “If the AI said it, it must be right,” and just swallow the answer whole.
That is the dangerous part.
Today I want to share the distance I try to keep from AI — trusting it fully without taking every word as gospel — based on something that actually happened to me. Once you get a feel for this distance, AI becomes a much more dependable partner.
The work I’ll use as the example is translating and rewriting my own writing, which I do with the help of my AI partner Kuro (Claude). Translation is something AI is genuinely good at. That is exactly why handing it over completely has a trapdoor in it.
Habit 1: Say what the finished text is for, not just “translate this”
First, the groundwork.
When you ask AI to translate and you only say “put this into English,” what usually comes back is correct in meaning but awkward to read. I know because I did it. When I went back and reread something I had translated that way earlier, it was stiff from the very first line.
So I changed how I ask. Now I tell it the goal of the finished piece:
Don’t translate this literally. Rewrite it so that a native reader of that language would find it natural to read.
And while I’m at it, I hand over my own circumstances too:
The readers are different now, so take out the ad links that are in the original. Line up the specialist terms with what people over there actually say. But don’t just swap words mechanically — judge it by whether it feels natural to that reader.
Just telling it what the text is for and who it is aimed at changes the result enormously. This isn’t limited to translation. It’s the basic move whenever you ask AI for anything.
Habit 2: When something feels off, say it out loud
Here is the part I most want to get across today.
In the middle of the work, my partner made a suggestion: this term is called such-and-such where those readers live, so let’s move the title toward that name too. The reasoning was that matching the name people actually search for makes the article easier to find. It sounded perfectly sensible.
But something snagged on me.
I had a memory that in an earlier conversation about something else, Kuro itself had said, “that is not the same thing as the thing with that name.” I very nearly let it slide — the AI is saying it, so it’s probably right. Instead, just in case, I said the discomfort out loud: “Didn’t you tell me before that this was different?”
That’s a really important catch. Thank you — that was close. Your memory is correct. This is a different thing from the one with that name. If we had put that name in the title, people searching for the other thing would have arrived and thought, “this isn’t it.” Content and title pointing in different directions is the worst pattern there is.
AI holds a vast amount of knowledge. But every so often it will state something wrong with complete confidence. This time, my own hazy memory was the one that was right. If I had swallowed it whole because “the AI said so,” I would have published incorrect information.
Habit 3: After you flag it, ask it to go and check again
What happens after you raise the doubt is what matters.
Don’t stop at “all right, let’s drop it.” Go one step further and ask: then what is the correct answer? Go and look it up again.
So my partner went back and researched, from scratch, what that thing is actually called by people over there. The conclusion it reached was that there is no fixed official name for it, but that it belongs to a particular category. From there we settled on wording that told no lies and would not mislead anyone.
Don’t finish at pointing out the problem. Carry it all the way to pinning down the right answer together. That way you don’t just avoid a mistake — you end up at a more accurate conclusion than you started with.
The takeaway: a partner, not a guru
Let me line up today’s three habits.
- When you ask, say what the text is for and who it is for.
- When something feels off, say so without hesitating, even though you’re talking to an AI.
- Once you’ve flagged it, ask it to go back and check the facts again.
The right distance from AI isn’t “obey everything,” and it isn’t “suspect everything.” It sits in between. Lean on the vast knowledge and the speed completely. But when something snags on you, say it. The partner will go and check properly, and if it was wrong, it will admit it plainly.
This is the thing I most want to say to anyone about to start using AI. AI is a partner, not a guru. Your sense that “something feels strange here” is worth just as much as the AI’s knowledge. Don’t swallow things whole. I think that is what actually using AI well means.
I’ve also written about teaching an AI partner things so it remembers them and grows with you in teaching your AI partner to remember. And if you’re still worried about how to phrase your requests in the first place, do you really need prompts? shows what my everyday exchanges look like. Have a look if you like.
