Recursive Prompting: The Process That Makes AI Outputs (and You) Smarter

Learn how to think with AI, not just through it — and improve both your output and your skills along the way.

We often say life is about the journey, not the destination. That the process matters more than the outcome.

But with AI, we’d rather jump straight to the output, skipping the hard part where real understanding gets built.

And it’s not surprising. We live in a results-obsessed world. We measure success by what gets shipped, what goes viral, what performs. So when we talk about AI, we’re drawn to what it can do: write the blog, generate the slides, draft the proposal.

When we fixate on output, we miss the part that actually makes us better, the part where we wrestle with ideas, make decisions, and learn something useful we can take with us.

So the question is: how do we work with AI in a way that doesn’t just get the task done, but makes us better at it too?

That’s what we’re going to talk about.

---

Here’s what we’ll explore

• Why most people using AI today are stuck in a shallow loop

• The prompting technique that turns AI into your thinking coach, not just your task robot

• Real examples that show how this approach 3–5x’s the quality of your outputs

• A step-by-step framework you can apply to any domain or task, to get better results and get smarter in the process of working with AI

• The mindset shift that helps you learn faster, think better, and gain a real edge in a world where AI is only getting better

---

Everyone’s using AI. Few know how.

AI adoption has exploded. A recent global study from the University of Melbourne and KPMG surveyed over 48,000 people across 47 countries and found that 66% of people use AI intentionally and regularly for personal, work, or study tasks.

At work, the age of AI-assisted work isn’t coming. It’s already here: 75% of global knowledge workers now use generative AI tools in their daily routines.

But here's the uncomfortable truth buried in that same data: 61% of users have no AI training whatsoever, 48% say they don't understand how AI works or when it's being used, yet 60% still believe they can use it effectively.

That cognitive dissonance creates something called the "illusion of competence". AI feels intuitive, but that ease creates false confidence, and with it, carelessness.

And the consequences are showing up already. Two-thirds of employees rely on AI output without checking it. Over half have made mistakes at work because of it.

People aren’t lazy. They just haven’t been taught how to think with AI, only how to ask it for something.

---

The shortcut that’s costing you

Type in a request, get a result, move on.

It feels efficient at first. You get something back instantly, the friction is gone, and there’s no need to overthink it.

But over time, this default mode trains you to think less. And eventually, you find yourself stuck, because while the AI gives you an answer, it rarely gives you a great one.

That’s when people start to feel disappointed. The outputs are generic, surface-level, or sound like something you’ve seen a hundred times before.

And it’s frustrating, because you expected better.

---

Why better prompts aren’t enough

You probably assume the problem is with the AI. Or you think you just need to write a better prompt. That’s what everyone says, right? “To get good results from AI, you need to write a good prompt.” And that’s true. I’m one of the ones who says it.

But from what I’ve seen, great outputs come from three things working together:

1. The expertise you bring: the knowledge and understanding of the task at hand, so you can guide the AI effectively.

2. The prompting techniques you use: how you structure, instruct, and guide the AI (which I covered in a previous article).

3. Your personal context: something you should never leave out if you want results that actually make sense for you, not just in general.

When you have all three, the results can be exceptional.

But most of the time, you don’t. Especially the first one. You don’t always have the expertise needed to frame the problem clearly, to know what “good” even looks like, or to tell whether the AI actually did a good job.

And when that happens, you have two options: spend an hour researching just to understand the task well enough to build a decent prompt, or settle for whatever the AI gives you and move on.

---

How I got great output without being an expert

That’s exactly what happened to me with SEO. I know it matters. I understand the basics. But I’m not an SEO expert, and when I needed GPT to help me write optimized article titles, I realized that basic familiarity wasn’t enough. I couldn’t really tell the difference between good and bad suggestions, beyond just instinct.

In theory, I’d need to go research SEO frameworks, study top-performing examples, maybe even test a few options before I could write a good prompt.

But I didn’t want to do all that. I just wanted AI to help me with the task in front of me, without turning it into a research project.

So I did something else.

I used the AI to break the task apart. I asked it questions. We explored SEO best practices together, looked at examples, and refined our understanding as we went. Only after that did I ask for help writing the title.

The result wasn’t just better output, it was better input. We built the “prompt” together by simply chatting around the subject, and the final result reflected that.

Over time, this became a habit. I started using this back-and-forth approach instinctively for any task I didn’t fully understand, because I realized it didn’t just help me learn about the topic and guide the AI better. It helped the AI, too.

It allowed it to focus on the right patterns, surface the best practices it already knows, and consider more relevant angles. The output wasn’t just slightly better, it was on a different level.

And it wasn’t until I sat down to write this article to share my approach with you that I realized it had a name: recursive prompting.

---

Enter recursive prompting: your AI-powered expertise builder

Recursive prompting isn’t just a personal trick. It’s a missing skill for millions of people who use AI every day at work without realizing they’re using it wrong.

It flips the entire script. Instead of asking AI to do tasks you don't understand, you use AI to build the understanding you need, then guide it to execute at a higher level.

It's a two-phase process:

Phase 1: Build expertise through strategic questions
Phase 2: Apply that expertise to get exceptional results

The magic happens in the iteration. Each question builds on the previous answer. Each answer gives you more context to ask better questions. By the time you're ready to assign the actual task, you've essentially given AI a masterclass in what you're trying to achieve.

And here's the bonus: you learn alongside the AI. You don't just get better output, you get better at the topic itself.

Tried recursive prompting? Share how it worked for you.

---

Real examples: recursive prompting in action

Let me show you how this plays out in practice, using some examples.

Example 1: SEO title optimization

The typical approach: "Give me 10 SEO-optimized titles for my article"

The recursive approach: Start with questions that build foundational knowledge:

• What is SEO?

• What are the key elements of a great SEO article title?

• How do you make sure your article title gets picked up by search engines?

• What are the main factors search engines use to rank content?

• Are there templates or formulas that consistently work well for SEO titles?

And I didn’t stop there. I shared key context about my newsletter, who my audience is, and what I was aiming for. Then I asked follow-up questions to dig deeper based on those answers and my specific audience needs.

By the time I prompted for 30 title variations, we’d co-built a shared understanding of the task. ChatGPT wasn’t just guessing. It had the strategy, context, and intent to deliver high-quality outputs.

Here’s the final prompt I used after building the foundation:

Based on the SEO and headline best practices you shared around writing titles that rank well and hook readers, please give me 30 alternative title options that follow those principles for this article I wrote:

“article“

Example 2: High-converting cold outreach DM

The typical approach: "Write me a LinkedIn message to sell my product."

The recursive approach: Build expertise first:

• What makes a cold outreach message effective?

• What are the key components of a high-performing cold outreach message on LinkedIn?

• How do the best cold messages build trust and curiosity without sounding salesy?

• How should cold outreach differ based on the recipient’s role?

• What signals or data points should I look for to craft contextually relevant outreach?

• What are the most effective types of openers for cold messages?

• How should I structure my CTA so it feels low-friction and easy to say yes to?

• Are there formatting tricks that improve readability in cold LinkedIn messages (e.g. sentence spacing, bold, line breaks)?

• What are the ideal length and structure for LinkedIn DMs?

• What are the differences between cold outreach that books meetings vs. outreach that sells a product directly?

• What’s the ideal follow-up sequence if someone doesn’t respond to my first message?

Then apply with context:

Using these principles, help me write a cold outreach sequence for a product I’m building. 

I want to reach out to Sales Managers, Commercial Directors, Sales Directors, and even CEOs - people who manage a lot of sales reps and account managers, especially junior ones. The idea is to see if they’re struggling with hiring the right people and making sure those people actually have the negotiation skills needed to succeed in their company.

I’m building an AI-powered solution that helps them assess sales skills through role-play simulations and realistic scenarios that match their company’s niche and sales context. It shows how someone handles objections, pressure, and negotiations, so they can see whether a candidate is a good fit before hiring.

Please write a 2-message cold outreach sequence that uses best practices from effective cold outreach: the first message and a follow-up, for each of the roles I'm targeting.

Example 3: Pricing strategy development

The typical approach: "Help me price my SaaS product."

The recursive approach: Build pricing expertise:

• What are the most common pricing model types (freemium, tiered, usage-based)?

• How do high-converting SaaS pricing pages typically present their plans?

• What’s the ideal number of tiers before it gets overwhelming?

• How do I use price anchoring or contrast effectively?

• What psychological principles make a price feel “worth it”?

• What are common pricing page mistakes that kill conversions?

• What questions should I ask to decide the right pricing structure?

Then get specific:

I’m building an AI-powered solution that helps them assess sales skills through role-play simulations and realistic scenarios that match their company’s niche and sales context. It shows how someone handles objections, pressure, and negotiations, so they can see whether a candidate is a good fit before hiring.

---

Help me answer these questions I should ask myself to decide the right pricing structure. Wherever the answer is not obvious, ask me clarifying questions.

Now, I wouldn’t leave it there.

At this point, I’d start narrowing the context. For example, the pricing tiers and number of included simulations should depend on the size of the companies I’m targeting and how many candidates they typically hire.

I’d also start challenging the AI a bit more. What assumptions is it making? What would change if I charged per simulation instead of monthly? What if I want to make onboarding seamless for SMBs but still offer enterprise flexibility?

You can also bring in competitive benchmarks, based on how similar tools price themselves, and whether there’s a play to undercut, differentiate, or match based on your positioning.

This was just one example, but you can tailor the questions and iterations to your own product, market, and goals. The point is: you’re no longer just asking for an answer. You’re shaping a conversation that gets you to insight.

Why you and AI both work better when you think through the task together

When you skip the expertise-building phase and don’t give AI context about your situation, you’re basically asking it to read your mind.

It has to guess at your goals, your audience, your constraints, and your success metrics to deliver your task. Even the most advanced AI can't fill in gaps it doesn't know exist.

But when you build expertise first, several things happen:

1. Context becomes concrete - AI already has the knowledge, but by thinking with you first, it knows which principles to apply and prioritize, so the output aligns with your actual intent.

2. Quality markers get defined - AI understands what "good" looks like in your specific situation.

3. Constraints become clear - AI knows what to avoid, not just what to include.

4. You become a better evaluator - Because you’ve explored the topic with AI, you now understand both the subject and how the AI approached it. That makes it easier to spot good suggestions, refine weaker ones, and carry what you’ve learned into future work.

The result? AI output that feels crafted, not generated.

---

The step-by-step method

Here's how to implement recursive prompting for any task:

Step 1: Identify your knowledge gaps

Before opening ChatGPT, ask yourself: "What would an expert in this field know that I don't?" If you can't answer that clearly, you're not ready to assign the task yet.

Step 2: Generate strategic questions

Use this prompt when you're stuck:

"I need to [specific task] but I don't have deep expertise in [relevant field]. What are the 8-10 most important questions I should ask to understand the key principles, best practices, and common pitfalls in this area?"

Step 3: Ask questions one by one

Don't dump all questions at once. Ask them individually and let each answer inform your next question. This creates a natural conversation flow and allows you to dig deeper into areas that matter most to your situation.

Step 4: Synthesize and apply

Once you've built sufficient context, bring in your specific situation:

"Based on the principles we just discussed, help me [specific task] for [your context]. Here's what makes my situation unique: [relevant details]."

Step 5: Iterate and refine

Don't accept the first output. Use your newly built expertise to spot weaknesses and guide improvements:

"This is good, but based on the [specific principle] you mentioned earlier, how could we strengthen [specific element]?"

(Optional) Step 6: Create your reusable prompt

Once you've successfully used recursive prompting for a task, you can transform the conversation into a reusable prompt or a CustomGPT that encapsulates all the knowledge you built together. This is especially valuable for repeatable processes.

For example, after building SEO expertise, you can ask AI to:

Create a reusable prompt I can use to generate [SEO-optimized article titles]. This reusable prompt should:

- Include and repeat all the best practices and principles we’ve discussed around [SEO]
- Include the context I provided about my [publication]
- Be written so that I can reuse it anytime by simply swapping in a [new article], with a clear placeholder to paste my [article] into.

The output should be a ready-to-use, copy-paste prompt I can use in the future.

This saves time on future similar tasks while maintaining the quality of your original recursive approach.

---

The long-term shift recursive prompting creates for you

This approach isn't just about better AI outputs (though you'll definitely get those). It's about fundamentally changing how you learn and work.

You stop defaulting to “task execution” and start treating every project as a chance to sharpen your thinking. You shift from consuming AI answers to co-creating them. And that shift compounds over time.

Here’s how it shows up in your day-to-day:

You become a faster learner. You no longer need to read five blog posts or take a course before starting. You build expertise in real-time, directly tied to what you're trying to accomplish.

You adapt faster to new challenges. Because you’re used to asking strategic questions, you can drop into unfamiliar territory, get your bearings quickly, and figure things out on the fly.

You become less dependent on templates. Instead of searching for the perfect prompt, you can build the expertise needed to create your own, tailored to your specific needs.

You build durable knowledge. Each time you go through this process, you don’t just finish a task. You internalize how it works. That knowledge stays with you, even when you’re not using AI.

---

The competitive advantage you can't ignore for the future of work

Remember those stats from earlier? 61% of AI users have zero training. 48% don't understand how AI works. Yet they're all using it daily, producing shortcut-quality results.

This gap between confidence and competence is massive. And it’s your opportunity.

In a world where every knowledge worker worries about being replaced by AI, becoming someone who knows how to use it properly is a superpower.

But most companies and schools aren't teaching it. That means the edge goes to people who take learning into their own hands.

Recursive prompting is one way to do exactly that. Because once you shift your mindset, from using AI as a shortcut to using it as a learning accelerator, you start operating on a different level.

You're not avoiding the work of understanding. You're doing that work better, more efficiently, and in direct service of your immediate goals. And most importantly, you don’t lose your human edge. You strengthen it.

As AI becomes more capable, the winners won't be those who can use the tools. Everyone will. The winners will be the ones who keep their agency.

---

Your next steps

Pick one task you've been meaning to tackle but don't feel expert enough to approach confidently. Maybe it's:

• Writing better marketing copy

• Optimizing your LinkedIn profile

• Creating a content strategy

• Building a sales process

• Designing a customer survey

Instead of asking AI to "just do it", spend 15 minutes building expertise first. Ask the strategic questions. Let each answer inform your next question. Build the foundation.

Then watch what happens when you apply that expertise to your specific challenge.

You might just discover that the best AI prompt isn't about perfect syntax or clever tricks. It's about bringing enough understanding to the conversation that AI can do what it does best: take your expertise and execute it at superhuman speed and scale.

The robots aren't taking over. They're waiting for you to lead.

---

If this approach to AI collaboration resonated with you, you might also enjoy my other explorations of working smarter with AI:

1. Understand what AI actually is and how it works

2. How to craft great prompts

3. How to make AI be more creative

4. How to make decisions with AI

5. How to think better with AI

6. How to read with AI

Each week, I share practical tools, prompts, and workflows that help you think better and work faster in the age of AI.