AI: Talking to it (Part 1)

Most people approach AI like it's either magic or a search engine. It's neither. It's more like a very capable, very literal assistant who takes you exactly at your word and has no idea what you actually meant unless you tell it.

There are two different ways to work with AI: talking to it, and handing it off. This is about the first one, the conversational mode, the one most people use every day without realizing there's a skill to it.

Say what you actually want, not just the topic

"Tell me about marketing" gets you a wall of generic information. "Give me three marketing ideas for a small coffee shop with a $500 budget" gets you something useful.

The plain version of a request, no examples, no extra context, is what's called a zero shot prompt. It's not wrong to start there, but it's why a first answer often feels generic. The model only has what you gave it to work with.

Think of it less like typing a search query and more like handing a task to a new coworker. The more context they have, the better the result.

A useful habit: before you hit enter, ask yourself what you'd tell a real person if you handed them this task. Include that.

Treat the first answer as a draft, not a final one

The first response is rarely the best one. That's normal, not a failure. Talking to AI well means staying involved instead of taking the first thing it gives you and running with it. There's an actual term for this: human in the loop. It means you keep reviewing, correcting, and redirecting instead of disengaging after the first request.

This matters more than it sounds like it should, because AI can sound completely confident while being wrong. That's often called a hallucination, a fluent, well written answer that just isn't accurate. If you've checked out after the first response, you won't catch it. If you're still in the loop, you will.

If the answer is close but not quite right, say so directly. "Make this shorter." "This is too formal, make it sound more casual." "I like the second idea, expand on that one." You're allowed to push back and redirect. It won't get annoyed, and it doesn't get worse the more you push.

Show it what you mean when words are hard to find

Sometimes it's hard to describe exactly what you want. That's fine. Instead of trying to explain it perfectly, show it.

This is few shot prompting, giving the model one or two examples of the kind of output you want, instead of describing it in the abstract. Paste in an example of a tone you like. Reference something specific: "write it like this email" or "make it sound like this paragraph." AI is very good at matching a pattern once it can see one, even if you can't fully put that pattern into words.

A useful habit: if you're stuck describing what you want, stop describing and start showing.

The bottom line

Talking to AI well isn't a technical skill. It's the same skill as giving clear instructions to any smart person you're delegating to: be specific, expect to go back and forth, use an example when words fall short, and stay in the loop instead of checking out after the first draft.

This is the mode most people default to, and it's the right one for most day to day tasks. But it's not the only mode. There's a second one, handing a task off entirely and letting AI work more independently, and that's a different skill with a different set of rules. More on that next.

Happy Learning,

KP