The Things You Don't Type Into ChatGPT

There’s a small pause that happens a lot, and almost nobody talks about it.

You’re partway through typing something into an AI chatbot. It’s going fine. And then you get to the part that actually matters — the name, the number, the diagnosis, the thing your boss said — and you stop. You back up. You swap “my sister” for “a family member.” You round the salary. You write “a difficult colleague” instead of Dave.

Then you hit enter, and you get an answer to a slightly different question than the one you had.

Most people don’t think of this as anything. It just feels like normal caution, the same reflex that makes you lower your voice on a train. But it’s worth looking at directly, because it’s costing you something.

What people quietly leave out

Ask around and the same categories come up again and again. The things people almost type, and don’t:

• The letter from the doctor. You want help understanding it. You paste in the technical terms and leave out the fact that it’s about you.
• The actual numbers. You want help thinking through a budget, a debt, an offer, a divorce settlement. You make up round figures instead, and get advice built on fiction.
• The email you’re furious about. You want help writing a reply that doesn’t burn the relationship. But pasting the original means pasting your manager’s words into a company’s servers, so you paraphrase it, badly, and the reply comes back oddly generic.
• The family situation. Aging parents, a sibling who isn’t speaking to you, a decision about a kid. Enormously useful to think through out loud. Not something you want sitting in an account somewhere with your email address attached.
• The first draft of anything honest. The resignation letter you might not send. The difficult message to a friend. The version where you say what you actually think, before you sand it down.
• The question you feel dumb asking. Which is, of course, exactly the kind of question a machine is perfect for — no eye-rolling, no history, infinite patience. Right up until you remember it’s not really just a machine.

None of these are scandalous. That’s the point. This isn’t a list of secrets; it’s a list of ordinary life.

The hedged version gets you a worse answer

Here’s the practical cost, and it’s bigger than people expect: an AI can only work with what you actually gave it.

When you strip out the specifics, you don’t get a slightly less personal answer. You get a slightly wrong one. “Help me reply to a difficult email from a colleague” produces advice for a generic difficult email. The real message had a tone, a history, a sentence that stung for a reason. All the useful help lived in the details you deleted.

It’s the same problem as describing a weird noise in your car over the phone instead of driving it to the mechanic. You can do it. You’ll get an answer. It just won’t be the answer you’d have gotten if you’d brought the actual car.

So you end up doing the work twice: once badly with the AI, then again yourself, filling in the parts you couldn’t hand over. Or you skip the AI entirely for anything that matters — which, for a lot of people, is most of what matters.

The instinct isn’t paranoia

It’s tempting to tell yourself you’re being precious about this. You’re not. The hesitation is a good instinct, and it’s responding to something real.

When you type into a cloud chatbot, your words travel to a company’s computers and land in an account with your name on it. In most cases they’re kept — for a while, or a long while. Staff can sometimes review conversations. Policies get rewritten. Companies get bought, breached, or subpoenaed. And your chat history sits there being a surprisingly complete diary of what you were worried about, month by month.

That’s not a conspiracy; it’s just how cloud services work. And it means the pause before you hit enter is your brain correctly noticing that a permanent record is being created somewhere you don’t control. If you use AI at work, there’s a second layer to it — and if you handle other people’s confidential information, a third.

The strange part is how quickly we all just… absorbed this. We adapted around it, quietly editing ourselves, and mostly never mentioned it to anyone.

What changes when it’s just your computer

There’s another way to set this up, and it removes the problem rather than managing it.

An AI can run on the computer in front of you. Not a slimmed-down version — a real, capable assistant, the same kind of back-and-forth you’re used to, except the program lives on your laptop the way your photo app does. You type, your own machine does the thinking, and the answer appears. Nothing is uploaded, because there’s nothing to upload to. No account. No history on someone’s server. If you close the lid on a plane with no Wi-Fi, it still works, which is as clear a demonstration as you could ask for that your words aren’t going anywhere.

What that does to the experience is hard to appreciate until you’ve felt it. The pause goes away.

You paste the whole letter. The real numbers. Dave’s actual email, with Dave’s name in it. You ask the embarrassing question. You write the angry draft you’d never send and ask for help finding the reasonable version underneath it — and you don’t spend a second wondering where that draft ends up, because it ends up in a file on your own hard drive that you can delete whenever you like.

The answers get better, immediately and obviously, because for the first time the AI has the real thing to work with.

This isn’t about having something to hide

The most common shrug about privacy is “I’ve got nothing to hide,” and it slightly misses the shape of the thing.

You close the bathroom door. You don’t read your diary aloud at dinner. You tell a friend something in a kitchen at midnight that you wouldn’t put in the group chat. None of that is because you’ve done anything wrong. It’s because thinking honestly requires somewhere to do it — a place where you can be half-formed, unfair, frightened, or wrong out loud without it being filed anywhere.

For a long time, computers were that place. Your documents folder didn’t report to anyone. Then the useful tools moved to the cloud, and we quietly traded the drawer for the filing cabinet in someone else’s building.

Local AI is just that drawer, given back. It’s not a security product for people with secrets. It’s an ordinary tool that happens to work the way people already assumed their computer worked.

Try noticing the pause

Here’s the only homework: for the next week, notice the moment you edit yourself before hitting enter. Notice what you took out. That deleted phrase is a pretty good measure of how much the tool is actually helping you.

If it turns out you’re leaving out the important half most of the time, that’s not a reason to trust the AI more. It’s a reason to use one that doesn’t ask you to.

WorkInPrivate is a private AI assistant that runs entirely on your own computer — no cloud, no account, nothing sent anywhere. It sets itself up in a few clicks, and the free trial runs on the machine you already own. Type the whole thing.