Advanced techniques: step-by-step thinking, options and perspectives

0:00 Techniques for better thinking

Once you've mastered structure and refinement, there's a next level: getting the AI to think better, not just write better. Better campaign ideas. Sharper diagnoses of why a post flopped. Stronger arguments in a sales email. In this lecture you'll learn seven techniques, none of them technical, including reasoning before answering, generating many ideas then filtering, role playing your audience, and chaining big tasks. You'll also learn when to switch on reasoning and deep research modes. You'll see a Lahore finance creator's hooks and a Karachi streetwear brand's diagnosis, and watch me run a data based strategy prompt.

0:43 Why techniques matter

Why bother? Three problems. First, the first ideas a model produces tend to be the most predictable, the same ones everyone else is getting. Second, when a model jumps straight to a recommendation, you can't see or challenge its reasoning. And third, big one shot tasks, like write a full launch campaign, let errors compound silently: a wrong assumption in step one ruins everything after it. Each technique in this lecture fixes one of those problems. Together, they turn the assistant from a fast typist into a useful thinking partner.

1:22 Reason first, then answer

Technique one: ask for reasoning before the answer. Before recommending a campaign angle, analyze the audience's main motivations and objections, then recommend. You'll see the logic, and you can challenge it. Many assistants also have thinking or reasoning modes that spend more time working through a problem. Use them for positioning decisions, diagnosing underperformance with your real data, or planning a content series. Don't bother for captions. And when the thinking needs current external facts, like a competitor's positioning, switch to deep research instead. A longer answer isn't automatically a better one, so always read the reasoning critically.

2:05 Many, then filter

Technique two: generate many, then filter. Give me twenty five hook ideas, then pick the five strongest for a skeptical audience and explain why. Asking for more pushes the model past its first, most obvious ideas, and asking it to filter gives you a shortlist to challenge. Technique three: creative constraints. Write the hook as a question a customer would text a friend. No adjectives in the first line. Use a metaphor from cricket. Constraints spark originality. And technique four: ask it to ask you. Before you write anything, ask me up to five questions that would help you write a better sales email. Brilliant when you're not sure what context matters.

2:54 Perspectives and role-play

Technique five: role play the audience. You're a thirty two year old first time buyer in Manchester on a tight budget. Read this landing page and tell me honestly what confuses you, what you doubt, and what would make you click. It's a fast pre test that surfaces obvious gaps. Technique six: multiple perspectives. Three versions of this ad: one for a price sensitive buyer, one quality focused, one time poor. Or a debate: argue for and against running this campaign during exam season. One important limit. Simulated customers reflect average patterns, not your real audience. Treat role play as a pre test, then validate with real comments, surveys or sales data.

3:43 Chains with checkpoints

Technique seven: break big tasks into chains. Instead of write a full launch campaign, chain smaller prompts. One: summarize the audience and key message, and you review it. Two: generate ten campaign concepts, and you pick one. Three: draft the hero video script, and you edit it. Four: create cut downs, captions and emails from the approved script. Each step is easier to check, and errors don't compound silently. Here's the key idea. The value of a chain comes from the checkpoints between the links. A chain without checkpoints is just a long prompt with extra steps.

4:25 Example 1: the finance creator's hooks

A simple example. A personal finance creator in Lahore is making a video on budgeting in high inflation. She asks for twenty five hooks across named angles, then the five strongest for salaried professionals who feel they never save. The shortlist includes a contrarian hook, the fifty thirty twenty rule doesn't work in Pakistan, here's what does, a relatable pain hook, salary comes on the first, gone by the twentieth, and a mistake hook, you're saving what's left, that's the mistake. She picks the contrarian one and checks her video actually delivers on it. And her prompt included a rule that blocked hooks promising guaranteed returns, which would be misleading, and a regulatory risk for financial content.

5:16 Example 2: diagnosing a flat Reel

Now a business case. A Karachi streetwear brand's Reel got far fewer views than usual. The creator pastes the script, caption, posting time and simple metrics, like average watch time and where viewers dropped off. The prompt: analyze why this may have underperformed compared to our usual posts. List three hypotheses ranked by likelihood, and one test for each. Don't assume data I haven't given you. The model suggests the hook reveals the product too late, the audio trend was already fading, and the caption gives no reason to share. The drop off data supports the first one most. So next week, the creator tests a new hook that shows the product in the first second. Hypotheses to test, not proven causes.

6:09 Watch me do it

Let me run the strategy prompt. I switch on the assistant's thinking mode, because this is analysis, not copywriting. I paste the context, a home decor brand targeting renters in London, and a table of the last eight posts with reach, average watch time, saves and link clicks. Task: analyze why the last three underperformed, give three ranked hypotheses with evidence from my data only, one cheap test for each, and say so if the data can't distinguish between them. The answer ties each hypothesis to specific numbers. Longer videos had much lower average watch time. The three weak posts all opened on a wide room shot instead of a close up. And then an honest line: your data can't show whether posting time matters. That honesty is what makes the rest trustworthy.

7:07 Common mistakes

Three mistakes. Treating role play as research. The simulated customer is an average, not your audience, so validate with real data. Chains without checkpoints, which lose the whole benefit of chaining. And over analysis for simple tasks. You don't need a reasoning chain or a thinking mode for a birthday post. Match the technique to the task: the harder and more consequential the decision, the more you want visible reasoning, real data and a test.

7:40 Recap and try this now

Let's recap the seven techniques. Ask for reasoning before the answer. Generate many, then filter. Use creative constraints. Role play the audience. Ask for multiple perspectives. Chain big tasks with checkpoints. And ask the model to ask you questions. Switch on reasoning modes for hard analytical problems and deep research when you need current facts. Here's your try this now. Pick one post, ad or email that underperformed recently. Paste real numbers into the strategy prompt from the lesson text, get three ranked hypotheses, and run the cheapest test next week.