Advanced Prompting Patterns

Hook

There is a ceiling on what standard prompts can produce.

You can write a perfectly structured system prompt with a clear role, a precise task specification and thorough context, and still hit output that is technically correct but misses the mark. The analysis is shallow. The recommendation is obvious. The memo reads like something a capable intern wrote, not a senior leader.

That ceiling exists because standard prompts ask AI to do one thing: produce output. Advanced prompting patterns ask AI to think first, to show its reasoning, learn from examples and inhabit a role with enough depth to catch what a surface-level pass would miss.

These three patterns are what you add when the basic structure is not enough.

Context

Pattern 1: Chain-of-Thought

Chain-of-thought prompting instructs the AI to reason through a problem step by step before producing its answer. This sounds trivial. The output improvement is not.

Why it works: AI models produce higher-quality output when they reason out loud. The process of articulating intermediate steps forces the model to surface assumptions, identify gaps and catch logical errors that a direct-to-answer approach would skip. You get the benefit of a structured thinking process, not just the conclusion.

When to use it:

• Any analysis that requires weighing multiple factors (build vs. buy, hire vs. contract, market entry decisions)
• Situations where you need to see the AI's reasoning, not just its conclusion (due diligence, risk assessment)
• Complex recommendations where a wrong answer would be costly
• When you suspect the AI might jump to a surface-level answer

Standard vs. chain-of-thought prompt:

Standard:

Assess whether we should hire a Head of People Operations now or in six months.

Chain-of-thought:

I need to decide whether to hire a Head of People Operations now or wait six months.

Before giving me your recommendation, reason through this step by step:
1. What are the costs and risks of hiring now?
2. What are the costs and risks of waiting six months?
3. What information would change this decision significantly?
4. Based on steps 1-3, what factors matter most?

Then give me your recommendation with the three most important reasons.

The second prompt produces analysis. The first produces an opinion.

Pattern 2: Few-Shot Examples

Few-shot prompting gives the AI one to three examples of the output you want before asking it to produce the actual output. This is the fastest way to close the gap between what AI produces by default and what your work actually requires.

Why it works: AI models are exceptional pattern matchers. When you show an example, you are not just giving instructions, you are showing the model what the target looks like at a structural, stylistic and tonal level. This eliminates the single biggest source of editing work: output that is correct in content but wrong in form.

When to use it:

• Any output with a specific format or style you have defined (your email style, your memo format, your feedback framework)
• Situations where your standard is non-obvious (not "formal" but your specific version of formal)
• Templates where every output follows the same structure with different content
• Cases where previous AI output missed the tone and you want to show rather than describe the correction

Few-shot template:

Here is an example of the kind of output I want:

---EXAMPLE---
[Paste a real piece of your best prior work, something you would be proud to send]
---END EXAMPLE---

Now produce [the actual deliverable] using the same format, tone and level of specificity.

My input: [paste your actual content]

One strong example beats a paragraph of tone instructions every time. Two examples beat one. Three is usually diminishing returns.

The most common mistake: Using AI-generated examples. The few-shot example should be your own work, something you wrote, edited and would publish. Using a piece the AI wrote introduces the AI's defaults, which you are trying to override.

Pattern 3: Structured Role-Play

Structured role-play assigns the AI a specific, detailed persona and instructs it to stay in character throughout a multi-turn interaction. This is different from the basic role assignment in Lesson 1, structured role-play is deeper, more constrained and designed for specific high-stakes use cases.

Why it works: When an AI inhabits a detailed role over multiple turns, its output reflects the assumptions, priorities and blind spots of that role. This is valuable for:

• Devil's advocate: Having the AI argue the strongest possible case against your plan
• Adversarial review: Having the AI act as a skeptical CFO, board member or investor reviewing your work
• Interview simulation: Having the AI be a specific type of interviewer so you can practice
• Pressure testing: Having the AI push back on your reasoning until it finds a weak point

Structured role-play template:

You are going to play the role of [specific persona with detailed attributes].

Your background: [detailed expertise, experience, perspective]
Your priorities: [what this person cares about and optimizes for]
Your skepticism: [what this person is most likely to push back on]
Your communication style: [how they speak, what they value in answers]

Stay in character for the entire conversation. Do not break character to explain your role.

[Optional: Here is the scenario you are reacting to: {paste your work}]

Begin.

Example, adversarial board review:

You are a board member with a private equity background reviewing our people strategy.

Your background: 15 years as a CFO and board member at PE-backed B2B SaaS companies, three successful exits. You have seen every version of "we need to invest in HR infrastructure" and you have approved maybe 30% of them.

Your priorities: EBITDA impact in 24 months, headcount efficiency ratios, leadership accountability. You care about outcomes, not programs.

Your skepticism: You distrust initiatives that do not have clear ROI, timelines or owners. You push hard on the difference between "investment" and "cost."

Stay in character. I am going to present our 12-month people strategy. Push back where you see gaps, ask the questions a board member would actually ask and do not accept vague answers.

Here is the strategy: [paste your strategy]

This pattern produces feedback that forces you to defend your work at a level your internal team will rarely push. It is one of the highest-value uses of AI for senior executives.

Steps

Step 1: Add chain-of-thought to your three most complex recurring prompts

Go back to your prompt library from Lesson 2. Identify the three prompts that involve the most analysis or recommendation-making. Add a reasoning scaffold to each one:

Before producing the final output, reason through:
1. [First reasoning step relevant to this task]
2. [Second reasoning step]
3. [What factors most affect the outcome]

Then produce [the actual deliverable].

Test each one. Compare the output to what the same prompt produced without the reasoning scaffold. The improvement in depth is usually immediate and significant.

Step 2: Build a few-shot library from your best past work

Collect five to eight pieces of your own best work: a memo you are proud of, an email that landed exactly right, an analysis that drove a real decision. Save them as "examples" in your prompt library alongside the prompts they are paired with.

Label each one:

• EXAMPLE: Executive update memo (attach to your update memo prompt)
• EXAMPLE: Client follow-up email (attach to your email drafting prompt)
• EXAMPLE: JD, Head of Engineering (attach to your job description prompt)

When you run the prompt, paste the relevant example before the input. Your output quality will improve immediately, and it will stay consistent across sessions.

Step 3: Create one adversarial reviewer for your most important work

Think about the hardest critic of your work, the CFO who cuts initiatives, the board member who asks the questions you do not want to answer, the prospective client who presses on your methodology. Write a detailed role-play persona for that critic.

Use the structured role-play template from Pattern 3. Be specific: their background, what they optimize for, what they push back on. The more specific the persona, the more useful the output.

Run your most important upcoming piece of work through this reviewer before you send it. Treat every pushback as a gift, it is a version of the critique you are going to receive anyway, and now you can address it before it reaches the actual stakeholder.

Step 4: Combine all three patterns for high-stakes deliverables

The three patterns compound:

1. Few-shot example sets the format and tone standard
2. Chain-of-thought ensures the reasoning is thorough before output is produced
3. Role-play reviews the finished output from the perspective of its harshest critic

For a board presentation, a major proposal or a critical hiring recommendation, run all three in sequence. The combined output is qualitatively different from what any single pattern produces, and it is qualitatively different from what most AI users are producing.

You are going to play the role of [your hardest critic: a specific board member, CFO or client].

Your background: [their expertise, experience and track record, be specific]
Your priorities: [what they optimize for: EBITDA impact, headcount ratios, outcomes over programs]
Your skepticism: [what they push back on hardest: vague ROI, missing owners, "investment" vs "cost"]
Your communication style: [how they speak and what they value in answers]

Stay in character for the entire conversation. Do not break character to explain your role.

I am going to present a piece of work. Push back where you see gaps, ask the questions this person would actually ask and do not accept vague answers.

Here is the work: [paste your deliverable]

Recap

• Chain-of-thought prompting instructs the AI to reason step by step before producing output. The improvement in analysis depth is consistent and significant.
• Few-shot examples are the fastest way to close the gap between AI default output and your specific standard. Use your own best work, not AI-generated examples.
• Structured role-play assigns a detailed, persistent persona that stays in character across multiple turns. Its highest-value use is adversarial review, having an expert critic pressure-test your work before it reaches the actual stakeholder.
• The three patterns compound. For high-stakes deliverables, use all three: few-shot for format, chain-of-thought for reasoning, role-play for adversarial review.

You have completed the Prompt Engineering track. You now have the three-part prompt structure, a system for building and maintaining a personal prompt library and three advanced patterns for high-stakes output.

The next track takes everything you have built here and applies it to full-scale AI workflows: chaining prompts, building automation pipelines and constructing systems that run without your direct involvement.

Back to Prompt Engineering track →