Prompt Engineering Cheat Sheet 2026

The foundational 4-part structure. Use this as your default starting point.

Role: You are a [expert role].
Task: [What you want the AI to do].
Context: [Relevant background the AI needs].
Action: Output [format] that [specific requirement].

Role: You are a senior UX copywriter.
Task: Rewrite the onboarding email below to increase activation.
Context: Our users are non-technical SaaS founders. Current click-through is 12%.
Action: Output 3 subject line variants + 1 revised email body (≤150 words). Use conversational tone.

Best for complex, multi-step tasks where you need the model to follow a process.

Role: Act as a [expert persona].
Instructions: [High-level task description].
Steps: 1) [step one] 2) [step two] 3) [step three]
End goal: The final output should [describe the ideal result].
Narrowing: Avoid [common pitfalls or constraints].

Role: Act as a data scientist and technical writer.
Instructions: Analyze this CSV of customer churn data and produce an executive summary.
Steps: 1) Identify top 3 churn predictors 2) Quantify each (% of churned users) 3) Recommend one intervention per predictor.
End goal: A 200-word summary a non-technical VP can present to the board.
Narrowing: Avoid statistical jargon. Do not re-print the raw data.

Great for content creation and writing tasks where tone and format matter most.

Context: [Background and audience].
Role: You are a [persona].
Action: Write [what].
Format: [Structure — bullet list, table, 3 paragraphs, etc.].
Tone: [Adjectives — professional, casual, persuasive, etc.].

Context: We're launching a B2B SaaS tool for HR teams. Target audience: HR directors at 200–2000 person companies.
Role: You are a B2B SaaS copywriter with 10 years experience.
Action: Write a LinkedIn post announcing our product launch.
Format: Hook line, 3-4 short paragraphs, bulleted benefit list, CTA.
Tone: Professional, warm, outcome-focused. No hype or buzzwords.

Forces the model to reason step-by-step before answering. Dramatically improves accuracy on logic and math tasks.

[Your question or problem].

Before giving your final answer, think step by step:
1. What do I know?
2. What are the key constraints?
3. What approach should I take?
4. Work through the solution.
Then provide your final answer.

A SaaS company has 1,200 customers. Monthly churn is 2.5%. They want to reach 2,000 customers in 12 months without increasing churn. How many new customers must they acquire each month?

Before answering, think step by step:
1. What do I know?
2. What are the key constraints?
3. What approach should I take?
4. Work through the math.
Then give your final answer with the monthly acquisition number.

Provide 2–3 examples before your real request. The model mirrors your pattern exactly.

Here are examples of [task]:

Input: [example 1 input]
Output: [example 1 output]

Input: [example 2 input]
Output: [example 2 output]

Now apply the same pattern:
Input: [your actual input]
Output:

Here are examples of converting plain feature descriptions into benefit-first copy:

Input: "Our software has automated invoice generation."
Output: "Stop spending Friday afternoons on invoices. Our software generates them automatically — so you close out the week in minutes, not hours."

Input: "The dashboard shows real-time analytics."
Output: "Know exactly what's working before you spend another dollar. The dashboard updates in real time so you can act on data, not hunches."

Now apply the same pattern:
Input: "Our tool integrates with 40+ apps."
Output:

Add explicit constraints to eliminate garbage outputs and force the model to respect your requirements.

[Your task description].

Hard constraints:
- Output ONLY [format] — no preamble, no explanation after
- Maximum [X] words / [X] bullet points / [X] items
- Never include [what to exclude]
- Always include [required element]
- Use [specific format, e.g. JSON with keys: name, description, price]

Suggest 5 names for a productivity SaaS targeting remote engineering teams.

Hard constraints:
- Output ONLY a JSON array of objects — no preamble, no explanation
- Each object must have: { "name": string, "tagline": string (max 8 words), "domain_available_likely": boolean }
- Names must be 1–2 words, pronounceable, not already a major brand
- Never suggest names with "AI", "Smart", or "Pro" in them