Ch.4: Make ChatGPT Useful in Five Seconds

0:00 Welcome to Learning Podcasts. ChatGPT for Everyone: The Art of Asking. Today we look at the gap between a question that gets you a generic answer and a question that gets you something you can actually use. 5 seconds of typing. That is the whole chapter. Two people open ChatGPT a minute apart. The first one types, write a thank you note. They get back three paragraphs of greeting card prose. Generic warmth, generic gratitude, the words "from the bottom of my heart" in there somewhere. Yeah, pleasant. Useless. They are not going to send it.

0:32 The second person types, write a two-sentence thank you note for a neighbor who watched my dog last weekend, warm but not gushy. 5 extra seconds of typing. And the response comes back as two warm, specific sentences. Something the person could text immediately, without changing a word. Same model, same minute. Completely different output. That gap is the whole chapter. Small changes in how you phrase a request can change what you get back, dramatically. And the changes are not technical. They are just being a little more specific about what you actually want.

1:04 When you type something vague, ChatGPT does not refuse to answer. It just makes dozens of decisions on your behalf. How long should this be? What tone? Who is the audience? How formal is too formal? Should it use bullet points or flowing paragraphs? And every one of those quiet decisions is a place where ChatGPT might guess wrong. Yeah. The output drifts a little further from what you wanted with every guess. By the time you read the response, the gap can feel huge. But the gap was opened one tiny decision at a time.

1:39 Here is the part that surprises most people. Adding restrictions to a request actually makes the output better, not worse. Wait. You mean things like word-count limits, or more like tone? Right, both. Tell ChatGPT to keep it under one hundred words. Tell it not to use any technical jargon. Tell it to write for someone who has never bought a house before. Each of those tightens the box ChatGPT has to work in. Yeah. And I'll say, I was slow to trust that. My instinct was always that more rules meant a stiffer result.

2:12 I had to see it break the other way a few times before it clicked. A tighter box eliminates the vague, padded, one-size-fits-all writing that nobody wants to read. Well, hold on. Aren't you just describing less creativity? Like if I tell ChatGPT, no jargon, under a hundred words, write it like this, am I just micromanaging it into a stiffer version of what it would have given me anyway? That is the natural reaction, and it is wrong by about a full order of magnitude. Constraints do not remove options.

2:43 They remove the boring middle. Without constraints, ChatGPT defaults to the average of every possible answer that fits your prompt, and the average is always the most generic version. Constraints push it toward the corners. The output gets sharper, not stiffer. So it feels less creative, but it is more specific. Yeah. Specific is what you actually wanted. Think about how a photographer frames a shot. You stand in the middle of an open field, and you point the camera at nothing in particular. You get a boring photograph. No subject, no foreground, no focal point. Just a wide green nothing.

3:23 Now you pick something. A single flower, 6 feet in front of you. You walk closer. You get the camera low. Yeah. You blur the background so the field becomes a soft green wash. You crop tight on the petals. And now you have a photograph somebody might actually frame on a wall. And the constraints did not limit the photo. The constraints are the composition. Same field, same camera. Whole different photograph because you chose what was in it. So which things can you actually constrain. A few specific levers, and naming any one of them makes a difference.

3:59 Length. Keep it under three sentences. Tone. Make it sound like a friendly text, not a business email. Audience. Use language a high schooler would understand. Vocabulary. Avoid any technical terms. What about format. Right, format too. Bullet points or flowing prose, a list or a paragraph. And is there a place to start, if you can only name one of these. Audience. Name the audience first. Most of the others come along for the ride once you do. None of these need a special syntax, by the way. They are just the way you would tell a colleague what you want, said out loud to ChatGPT.

4:34 Another move that punches above its weight is asking ChatGPT to respond from a particular point of view. Explain the benefits of stretching as if you were a physical therapist talking to a patient. Different language, different emphasis. Explain how compound interest works as if you were talking to a 10-year-old. Strips away financial jargon, gives you something clear and visual. Or, review my resume as a hiring manager at a tech company. Shifts the lens from general writing feedback to career advice.

5:04 So you skip past the generic middle ground and get something tailored. Yeah. The generic middle ground is the boring photograph. You did not say what was in the frame. A nutritionist explaining meal prep will spend most of the answer on, you know, balance and portion sizes. A busy parent explaining meal prep will spend most of the answer on speed, on shortcuts, on what you can do on Sunday night so Tuesday morning is not a disaster. But couldn't the same shift happen if you just added more context to the request.

5:34 Why does calling it a perspective matter. Yeah, that is the right question. Adding context says, here are the facts of my situation. Setting a perspective says, here is the lens I want you to read those facts through. Same facts, different lens, very different answer. So neither one is wrong. They just do different jobs. Right. When you tell ChatGPT which lens to look through, you are not asking for a biased answer. You are asking for the answer that fits your actual situation. Sometimes the fastest way to communicate what you are after is to just show it.

6:09 If you want ChatGPT to write something in a particular style, paste in a sample. Say, write something similar to this, but change the topic. Yeah. Like, paste in an email you liked. Then say, follow this structure but make the topic annual reviews. Right. A concrete example anchors the output in a way that an abstract description, Often cannot. It is the same shape as asking a singer to cover a song. If you say, sing me a folk song, you get something. But the singer is making a hundred decisions in the dark.

6:43 Which folk song? Whose version? Tempo? Key? Hand them an actual recording and say, do it like this, and most of those decisions are made for them. Now they know the shape of what you want. And you can still ask for tweaks. Slow it down. Try it without the second verse. Make it feel a little sadder. Yeah. The example locked the shape. The follow-ups change the details. One of the most underused moves is telling ChatGPT what not to do. Do not use bullet points. Avoid jargon. Do not start the response with the word "certainly".

7:18 Skip the introduction and get straight to the advice. Well, those negatives are often more leverage than another paragraph of positive direction. Right. Because they close off the paths ChatGPT would have taken anyway. Without them, the model defaults to its most common patterns. With them, you have, basically, blocked the doors you did not want it to walk through. And ChatGPT has habits. Real verbal habits, the same way a writer has tics they fall back on under pressure. It will start with "certainly".

7:48 It will load up on bullet points even when you wanted a paragraph. It will give you a polite, formal preamble before getting to the actual content. And a single "do not" removes any one of them. Yeah. The first time I noticed that was kind of by accident. I put "do not start with the word certainly" in a request because it kept showing up and it was driving me a little crazy. The whole response changed register. I had not expected one negative instruction to do that much. So three of them stacked up?

8:19 Three short negative instructions will, Change how the response feels more than a whole new prompt would. One more move that saves rounds. Ask for variety in the same prompt. Give me three different versions of this. One casual, one formal, one funny. And then you compare instead of iterate. Right. You read three responses, you point at one and say, the closing of this one with the tone of that one. That is one prompt instead of three rounds. Often saves several rounds of revision. So a precise first message earns most of the answer up front.

8:54 The back-and-forth from any conversation becomes a polish pass instead of the whole job. Next, we take these moves into writing. Emails, cover letters, the post you have been staring at for 10 minutes. We look at how to turn ChatGPT into your writing partner. Thanks for listening to Learning Podcasts.