Just Keep Talking: How to Tell if You Can Trust What AI Tells You

The simplest way for a non-technical beginner to check whether a ChatGPT or AI answer is any good.

You asked AI a question and it gave you an answer. A good one, by the look of it. Tidy, confident, well organised, the kind of thing you’d expect from someone who knows what they’re talking about. One small part made you pause for half a second, but it looked so sure of itself, and you were busy, so you let it go.

That half-second is what this post is about. Because the answer looked finished, you were about to treat it as finished. Sometimes that’s fine. Sometimes it isn’t, and the awkward part is you can’t always tell which.

How do you know if an AI answer is right?

Here’s the short version. You don’t need to know more than the AI to judge whether its answer is any good. You just need to stay in the conversation a little longer. The first answer isn’t the last word, it’s the first draft, and the quickest way to test it is to ask the AI to turn on its own answer and show you where it’s weak.

That’s the whole move. The rest is permission to do it, and the exact lines to use.

Why a polished answer feels finished

A finished-looking answer feels like a finished answer. That’s the trap. AI writes in the same calm, confident, well-structured voice whether it’s right or completely wrong, so the polish tells you nothing about whether the thinking underneath holds up.

The belief doing the damage is a simple one: “if it sounds sure, it probably knows.” But confidence isn’t a sign of correctness. It’s just the voice the machine always uses, for everything.

You’re allowed to question it

If a tradesperson handed you a tidy quote, you wouldn’t sign it on sight. You’d ask what’s included, what’s assumed, and what might change once they opened up the wall. That isn’t rude. It’s the ordinary thing a sensible adult does before relying on something.

You’re allowed to do exactly that with AI, and you have more standing than you think. The AI has read about your kind of situation in general. You’re the one actually living this one, with the real details and the real consequences. Being the person on the hook isn’t the junior position in the room. It’s the senior one.

Just keep talking

So here’s the principle, and like the first one it’s plain: just keep talking.

When an answer matters and you’re about to lean on it, the move is to make the AI inspect its own work, rather than asking it to reassure you. The single most useful thing you can say is some version of this:

Before I rely on this: what’s the weakest part, what should I check before I trust it, and what would you change if you argued against yourself?

Notice what that does. It doesn’t ask “are you sure?”, because a machine will often just apologise or dig in. It asks the AI to build the case against itself, so you can look at both sides and make the call. You’re still the judge. You’re simply better informed before you decide.

What this looks like in real life

Say you’re planning a small surprise party and you ask AI to map out the afternoon. It hands you a lovely, confident run sheet. Something about it nags at you, but you can’t put your finger on what, so you nearly let it slide.

Instead, you say the nag out loud:

Something about this plan feels off to me and I can’t say what. What might it be? Give me a few possibilities.

The AI comes back with three things it might be, and one of them lands: the cake pickup overlaps with the time you’ve told guests to arrive, so the room would be half empty at the surprise moment. You didn’t need to be an events planner to catch that. You just refused to treat the first answer as the last word, and you said the uneasy feeling out loud instead of swallowing it.

Lines you can use right now

Here are the ones worth keeping for the moment an answer is sitting there looking done.

The weak version, for comparison, is “Are you sure?” It’s better than nothing, but it tends to invite a reassuring yes. These do more:

What’s the weakest part of what you just told me?

Argue against your own answer, properly. Make the strongest case you can that it’s wrong.

And when something nags and you can’t name it:

Something about this feels off and I can’t say what. What might it be? Give me a few possibilities.

One thing matters here, so this doesn’t tip into worry. You don’t do this to every answer. Most of the time, a quick reply to a low-stakes question is fine exactly as it is. You keep talking when the answer genuinely matters, when being wrong would cost you something real. Check the part that would hurt to get wrong, and then let yourself stop.

Where this goes next

The answer is where the thinking starts, not where it stops. And the thinking is still yours.

If the blank box is still the harder problem for you, start with the first principle, just start talking, and come back to this once you’re getting answers you need to weigh up. This post picks up the moment after, once AI has replied and you’re deciding whether to believe it.

This is the heart of my second book, Just Keep Talking, the next step in The Simplest Beginner’s Guide to AI. The book takes this one habit and works it through the trickier cases: when the writing is so good it hides thin reasoning, when you can’t check the subject yourself, and when AI quietly talks you round to its way of thinking.

You can find Just Keep Talking on Kindle here: https://www.amazon.com/gp/product/B0H57YHPXZ