Transcription
Most people using AI are getting it wrong. That's why learning it the right way gives you an edge that 99% of it did.
I didn't come to AI from tech. I came to it after using it to uncover hidden leaks that added millions in value to my fitness and education businesses. And here's what I'm seeing right now. The gap between people who actually understand how AI works and people who just try prompts is widening fast.
In this video, I'm going to walk you through seven steps that close that gap. A clear path to using AI the way the top one person does. And here's the part that matters. You don't need a technical background. If you follow this the right way, you can start seeing real results in about 30 days, even from zero. Let's get into it.
This first step is about learning how AI is actually taking in what you say. Most people talk to AI like it's a person, and that's where things go wrong. Why? Because tools like Chad GPT or Claude or Gemini aren't understanding your words the way a human would. They're guessing what comes next.
Your brain does this too. If I start a sentence like once upon a, you already felt the ending forming. You didn't reason it out. You predicted it. That same idea shows up in search. You type a few words into Google and it starts finishing the rest for you. Why? because it has seen that pattern millions of times before. It learned what usually follows.
AI models like chat GPT or claude works in a similar way. But here's the key difference. Search engines point you to answer that already exist. AI doesn't. It builds the response in real time based on probability not memory.
So how does AI actually come up with the next word? At a high level it starts with breaking what you typed into small pieces. These are called tokens. Sometimes a token is a full word. Sometimes is just part of one. So in a phrase like once upon a once is a piece upon is another and or a becomes another.
Each of those pieces gets turned into numbers. Not a single number a whole bundle of them. Those numbers get dropped into a huge mathematical space. Think of it like a giant map. on that map. Ideas that belong together end up closer. Ideas that don't end up far apart. That spacing isn't random. It's learned from the patterns the system has seen over time. So words like once upon time, story and ferry end up living near each other. While words like engine, invoice or sandwich are nowhere close.
Now when it's time to respond, AI looks at the context you have given it and starts predicting what should come next. So when it sees once upon a it runs through the options like once upon a time once upon a mistake once upon a spreadsheet and based on probability and distance time wins. So the line gets completed not because air remembers a story not because it understands meaning but because that word fits best in that spot.
That's why AI can feel incredibly smart and also a little off because it isn't thinking it's predicting. I'm skipping a lot of mechanics here. You don't need all of that right now. Here's the part that actually matters. When you're unclear, guessing engines called chair GPT or Gemini has no choice but to guess. And vague questions create vague answers. But when you're precise, something changes. The guesses get tighter. The output gets sharper.
That's the difference most people miss. AI isn't trying to understand what you meant. It's trying to calculate what you want. That's what I mean when I talk about speaking AI language. You're not asking it to read your mind. You're giving it enough structure to compute your intent. And when you do that, the quality jump is immediate.
So what does a sharp prompt actually look like? I use a simple rule called act. First is actor. You tell the AI who it's stepping in as. Second is context. You give it the situation it needs to understand. Third is task. You tell it exactly what you want done. Most people skip this and type something like, "Help me write an email." That's why the response sounds generic.
Here's act in practice. You start with the actor. Act like someone who has handled difficult conversations hundreds of times and knows how to say things clearly without escalating the situation. Then you give the context. This is a follow-up email after a missed deadline. The relationship matters. I need to stay firm without sounding aggressive. Then you give the task. Write a short email that explains the issue, sets a clear expectation, and asks for a concrete next step.
Now the AI isn't guessing. It knows the role. It understand the situation and it knows what a good outcome looks like. That's act. When you use it like this, it turns a loose request into a structure the model that can work with. Now it can process the setup, follow the logic and respond with purpose. And you can use this structure almost anywhere like emails, writing, planning, learning, problem solving. And from now on, you will start seeing the results to be at least 10 times better than before. Because when you speak its language, AI finally starts pulling its weight.
Now that you know how to talk to AI, the next step is choosing your main tool. This is where most people mess up. They look up 50 best AI tools. They pick 10, bounce around, and never get good at any of them. That approach doesn't work. My advice is simple. Pick one, go deep.
Learning AI is a lot like learning a sport. People already play one sport don't start from zero when they try another. They already understand timing, practice, reading, movement. There's research that shows this. People with experience in one skill pick up related skill faster than complete beginners.
I have lived this myself. I have played cricket for as long as I can remember, like league games year round. When my school held baseball trials, I had never played before. Different rules, different grip, different movements, but I didn't feel foreign. I already knew how to track the ball, how to react, how to practice. So, the learning curve was shorter.
AI works the same way. When you go deep on one model, your brain learns how these systems behave. And once you feel that rhythm, every other model becomes easier to pick. Not because they're identical, but because you already know how to learn them.
So which one you should start with? If you want the most polished experience, start with Chad GPT. If you already live inside Google tools, Gemini will feel familiar. If you want something that's more structured for longer work, Claude is solid. But here's the truth. The choice doesn't matter as much as people think. What matters is this. For the first week, stick with one. Spend real time with it. Notice how it responds, where it's strong, where it breaks. You're not trying to master it yet. You're trying to feel its rhythm. And once you're comfortable, start using act structure we talked about. By the end of that first week, writing a structured prompt shouldn't feel deliberate anymore. It should feel automatic.
All right. Now that you're actually using it, let's talk about what really makes answers good. It's not the tool, it's the setup. Even the smartest AI sounds lost without it. Every response it gives depends on how it frames what you ask. And if the setup is weak, there's nothing for it to anchor to.
Remember, what's inside these systems? No understanding, no common sense, just a massive space filled with numbers. The setup is what guides it through that space. It tells AI where to focus and what actually matters. And the easiest way to do that is to think of your setup as a map you're handing it before it starts answering.
Here's the structure I use to build that map. I call it scope. S is for session. What's already happened in the conversation? Past messages, summaries, anything that gives continuity. If needed, you paste it back in or ask the model to recap before moving on. Then C is for content, the real material like files, notes, data, anything concrete you give it to work with. O is for operations, what the model is allowed to do. Search, analyze, write, organize. Those actions shape how it reasons. P is for prompt. This is the instruction itself. what you're asking for, how specific it is, what success looks like, and E is for emphasis, what matters most, what to prioritize, what can be ignored.
When you control these pieces, you're no longer throwing questions into the dark. You're guiding the model through the space. And the better you set up, the stronger the reasoning and the better the answer that comes back.
Once you start using frameworks like ACT or Scope, you're already ahead of 99%. But if you want to get genuinely good at this, there's one more skill you need. You have to learn how to fix your setup. Because when you're not getting the answers you want, most of the time it's not the AI, it's your setup, it's your instruction, it's what you didn't say.
I still remember the first time I tried using early models from OpenAI. I sat there for hours trying to make it work. And by the end, I was irritated because it felt random. Back then, nobody had language for it. Nobody was talking about prompt engineering like it was a skill. But here's the truth. Prompting isn't typing. It's iterating.
So, when the output is weak, I don't blame the model first. I check myself. Did I choose the right actor? Did I give enough setup? Did I make the task clear? Sometimes I'll even ask the model directly, "What did you assume? What made you answer it that way?" And when it explains, that's when it clicks. You're not just using AI anymore. You're starting to see how it thinks.
I use three quick loops for that. When things feel off, the first is the walkthrough loop. When an answer feels off, I tell it, "Go step by step, show how you got there, then give me the short final answer." The second is the clarifier loop. I say, "Ask me three questions so you understand what I mean one at a time. Then use my answers and try again." The third is the rewrite loop. Before it answers, I tell it, "Give me two cleaner versions of my questions." I'll pick one, then answer that.
And you keep cycling these loops because that back and forth teaches the model how you communicate. And it teaches how you to ask it in a way that gets clean results. You test, you adjust, you push again until you can tell exactly what's working and what's breaking. That's when it clicks. You're not tossing prompts at a machine anymore. You're in a real back and forth, and both sides gets better as you go.
But there's still a problem. Fixing how you ask isn't enough on its own. Because if what you get back sounds like the same polish fluff everyone else is posting on LinkedIn, then you haven't gone far enough. That's why the next step is learning how to point AI toward real expertise.
When you ask Chad DPT a question, it's not pulling a correct answer off a shelf. It's generating a response from a massive pool of likely ideas. Some of those ideas are solid, some are safe, some are invented, some are just wrong. And when your question is broad, the model usually lands in the middle. Comfortable, generic, forgettable. You read it and think yeah I have heard this before.
So instead of letting it sit in the middle you push it towards sharper thinking. Rather than asking something like how do teams improve performance you say explain performance improvement using ideas from elite sports coaching aviation safety systems and military afteraction reviews. Now you have given it direction, depth, edges to work with. And if you don't know who or what those experts are, that's fine. You ask AI first, who are the leading thinkers, researchers or fields that study this problem seriously. Then you take that list, feed it back in and say, use these perspective to build a clear framework. That's how you stop AI from repeating the average answer and start using it to explore real thinking instead of echoing the crowd.
One thing you can't skip, you have to check what you get back. That's step six. Because AI will tell you things like humans only use 10% of their brain. You have heard that before. It sounds right. It's completely false. And that's the scary part. AI sounds as sure when it's wrong as when it's right. You can warn it all you want to stop making things up. It won't. These systems are built to generate answers. That's the whole point.
So instead of trusting, you verify. Don't just read it, pressure it. I use five simple checks to separate real insight from confident nonsense. First is assumptions. Ask it what did you assume when you answered this. Then have it rank those assumptions by confidence. You'll often see the weak spots immediately. Second is source. Make it back up each major claim. Not vague references, actual sources you can look up yourself. That's how you see what's real and what's filler. Third is push back. Tell it to find a credible view that disagrees. Then explain why the disagreement exists. This is where shallow answers fall apart. Fourth is math. Anytime number shows up, slow it down. Have redo the calculation step by step. You'll be shocked how often the answer changes. And the last is cross check. Run the same questions through another model like Chad GPT, Gemini or Claude. Then ask one to critic the other or say, "Verify these claims." That's how you separate noise from knowledge.
By the end of week three, you'll feel it. You're not guessing anymore. You're in control. But here's the final shift. The best AI output doesn't sound clever. It sounds like you actually mean it. That's the goal. That's why the final step is about developing taste.
Because most people use AI like a shortcut. They take the first answer, clean it up a little, and move on. And it shows by now you passed that you have put in time. You have built control. So this is where the real work starts. You stop accepting answers and you start shaping them. You push back. You challenge the idea. You force clarity.
This is where the voice framework comes in. It's how you turn something generic into something that actually sounds like you. V is for view. Ask, is there an angle here I didn't already expect? If not, push it to explore. a few different perspective and pick the one that feels sharpest. O is for objects. Does the answer point to something real? You can picture a moment, a situation. If it's floating in abstractions, make it concrete. I is for insight. Can you follow the thinking behind the answer? If the logic feels rushed, slow it down. Ask it to show how it got there. C is for conviction. Does it take a position you could agree or disagree with? If it's hedging, force it to choose and explain why. E is for experience. Does it flow like something a human would actually say? If not, rewrite it as a short story, not a checklist.
So that's voice framework to add taste to your output. And once you start working this way for over 30 days, something shifts. Each time you adjust a prompt, each time you rethink an answer, each time you decide what's good enough, you're not only shaping the output, you're sharpening your judgment.
AI isn't something we get to opt out of. It's already here. And for a lot of people, that creates anxiety. I see it the opposite way. AI doesn't remove human value. It makes clear thinking more valuable than ever. It doesn't replace you. It exposes how you think. That's the opportunity.
If this clicked, next thing you should watch is this video. Because learning AI faster doesn't matter if your money system still forces bad decision under pressure. That video breaks down how the top tier design systems so one bad month never controls their focus, time or choices. Watch that next. It connects directly to what we just covered.
And quick favor, if this helped you think more clearly, subscribe. I promise not to waste your time or yell smash the like button at you. You'll get videos that actually build on each other. Also, if you want the thinking behind these frameworks in writing, sign up for the newsletter. That's where I share ideas that don't always make it into the videos.
And I'll leave you with this. AI doesn't reward speed. It rewards structure. And structure is a skill you can build. I am Farukq Zafar, the founder of Great Behind Growth, where we build real systems with grit, not gimmicks. Let's build something real together.