How to Write Prompts That Make ChatGPT Actually Useful

I want to learn the RISEN prompt framework by applying it to a real task. Help me construct a RISEN prompt for the following goal:

My goal: [describe what you want ChatGPT to help you with]

Walk me through building the prompt using each component:

**R - Role**: Assign ChatGPT a specific expert role. Ask me: "Who is the ideal expert to handle this task?" Then write the role assignment.

**I - Instructions**: Define what exactly I want done. Ask me: "What's the specific output you need?" Then write clear instructions.

**S - Steps**: Break the task into sequential steps. Ask me: "What's the logical order of operations?" Then write numbered steps.

**E - End Goal**: Define what success looks like. Ask me: "How will you know the output is good?" Then write the success criteria.

**N - Narrowing**: Add constraints that eliminate unwanted outputs. Ask me: "What do you NOT want?" Then write the constraints.

After building the prompt, show me the complete assembled RISEN prompt and explain why each component improves the output compared to a simple one-liner request.

Then have me test it and iterate based on the result.

Teach me the role-playing technique for prompt engineering by demonstrating with examples.

Show me how assigning a specific role changes ChatGPT's output by running this experiment:

**Task**: [describe a task, e.g., "review my business plan" or "explain blockchain" or "write a product description"]

Generate the same output for this task using 5 different role assignments:
1. No role (plain request)
2. "You are a [relevant expert with 20 years of experience]"
3. "You are a [skeptic/critic who challenges assumptions]"
4. "You are a [teacher explaining to a beginner]"
5. "You are a [relevant expert] AND a [contrasting perspective]" (dual role)

After generating all 5, analyze:
- How did the depth change?
- How did the vocabulary change?
- How did the structure change?
- Which role produced the most useful output for this specific task?
- When should I use each role type?

Then help me write the optimal role assignment for MY actual use case: [describe your real task]

The role should include:
- Their expertise and experience level
- Their communication style
- Their default assumptions about the audience
- What they prioritize in their work

Teach me chain-of-thought prompting — the technique of asking ChatGPT to show its reasoning step by step rather than jumping to an answer.

Demonstrate with this problem: [describe a complex problem or analysis you need help with]

First, show the basic prompt (just asking for the answer directly).
Then, show the chain-of-thought version with these additions:
- "Think through this step by step"
- "Show your reasoning at each stage"
- "Before giving your final answer, consider these perspectives: [list 2-3 angles]"
- "Flag any assumptions you're making"

Compare the two outputs and explain:
1. Why does chain-of-thought produce better results? (What's happening in the model?)
2. When is it most valuable? (complex reasoning, math, analysis, creative problem-solving)
3. When is it overkill? (simple factual lookups, formatting tasks)
4. How to use "Let's think about this..." as a trigger phrase
5. How to combine chain-of-thought with other techniques

Then help me apply chain-of-thought to my actual problem and see the difference in quality.

Teach me the few-shot prompting technique — giving ChatGPT examples of what I want before asking it to generate new output.

My task: [describe what you want ChatGPT to produce — emails, product descriptions, code, analysis, etc.]

Walk me through:

1. **Zero-shot** (no examples): Write the basic prompt.
2. **One-shot** (1 example): Add one example of the desired input → output format.
3. **Few-shot** (2-3 examples): Add multiple examples showing the pattern.

For each, generate the output and compare quality.

Then teach me:
- How to choose the RIGHT examples (what makes a good few-shot example?)
- How many examples is enough (and when more examples hurt rather than help)
- How to format examples consistently (the pattern matters)
- The "example + anti-example" technique (showing what NOT to do)
- How to use examples to control tone, length, and structure simultaneously

Template for building few-shot prompts:
---
Here are examples of [what I want]:

Example 1:
Input: [input]
Output: [ideal output]

Example 2:
Input: [input]
Output: [ideal output]

Now do the same for:
Input: [my actual input]
---

Help me build a few-shot prompt for my specific task.

Teach me how to control ChatGPT's output format precisely.

I often want outputs in specific formats but get rambling paragraphs instead. Show me how to fix this.

Demonstrate all formatting techniques using this sample task: [describe a task, e.g., "analyze my competitors" or "plan my content calendar"]

1. **Markdown formatting**: How to request headers, bullet points, numbered lists, bold/italic, tables
2. **Table format**: How to get clean, structured tables with specific columns
3. **JSON/structured data**: How to get machine-readable output
4. **Specific length constraints**: How to request "exactly 3 sentences" or "under 100 words" and have it respected
5. **Section structure**: How to define the exact sections and their order
6. **Template filling**: How to give a template with blanks and have ChatGPT fill them
7. **Multi-part responses**: How to request "Part 1: X, Part 2: Y" structures

For each technique, show:
- The prompt instruction that triggers it
- An example of the output
- Common failures and how to fix them

Then create a "formatting cheat sheet" I can reference:
| I want... | Add this to my prompt... |
|-----------|------------------------|
| A table | "Present this as a markdown table with columns: X, Y, Z" |
| Bullet points | ... |
| Specific length | ... |
| etc. | ... |

Teach me how to iteratively refine ChatGPT's output instead of trying to get a perfect result in one shot.

Starting task: [describe what you want to achieve]

Walk me through the refinement loop:

**Round 1 — Get the rough draft**: Write a basic prompt and generate initial output.

**Round 2 — Diagnose what's wrong**: Show me how to analyze the output:
- What's good? (keep this)
- What's mediocre? (improve this)
- What's missing? (add this)
- What's unwanted? (remove this)

**Round 3 — Targeted refinement prompts**: Show me how to write follow-up prompts that fix specific issues:
- "Keep everything above, but [specific change]"
- "Rewrite section 3 to be more [specific quality]"
- "Add [missing element] between sections 2 and 3"
- "Remove all instances of [unwanted pattern]"
- "Make the tone more [target tone] while keeping the content"

**Round 4 — Polish and finalize**: Final pass for consistency and quality.

Teach me:
- When to refine vs. when to start over (sunk cost trap)
- How to give feedback that ChatGPT actually responds to
- The "yes, and" technique (building on good output rather than replacing it)
- How to save successful prompts as templates for future use
- The optimal number of refinement rounds (hint: usually 2-3)

Teach me how to assign both a persona to ChatGPT AND define the target audience for more relevant outputs.

My content task: [describe what you're creating and who it's for]

Show me the difference between:
1. Plain prompt (no persona, no audience)
2. Persona only ("You are an expert copywriter")
3. Audience only ("Write this for small business owners")
4. Both ("You are a senior copywriter at a top agency, writing for first-time founders who are technical but not marketing-savvy")

For each, generate the same content and analyze the differences.

Then teach me the audience definition template:
"Your audience is:
- [Job title / role]
- [Experience level with this topic]
- [What they already know]
- [What they don't know]
- [What they care about most]
- [What would make them stop reading]
- [The action you want them to take after reading]"

Show me how this template changes output for:
- A technical tutorial
- A sales email
- A social media post
- A business report

Then help me define the persona + audience combination for my actual task.

Teach me how to use constraints and limitations to dramatically improve ChatGPT's output quality.

Show me how each type of constraint changes the result for this task: [describe your task]

**Length constraints**:
- "In exactly 3 bullet points"
- "In under 50 words"
- "In one sentence"
How brevity forces clarity.

**Format constraints**:
- "As a table with these columns"
- "As a step-by-step checklist"
- "In email format with subject line"

**Content constraints**:
- "Without using jargon"
- "Without using the word [overused word]"
- "Using only data from [specific source]"
- "Only include points you're highly confident about"

**Tone constraints**:
- "As if explaining to a smart 12-year-old"
- "In the style of [publication]"
- "Casual but not sloppy"

**Perspective constraints**:
- "From the customer's perspective, not the company's"
- "From a skeptic who needs convincing"
- "From someone who has tried everything else"

For each constraint type, show before/after examples and explain why the constrained version is better. Then help me design the optimal constraint set for my task.

Teach me negative prompting — the technique of explicitly telling ChatGPT what to avoid.

My task: [describe what you want help with]

Show me how adding "DO NOT" instructions eliminates common ChatGPT bad habits:

1. **Eliminate filler**: "Do NOT start with 'Certainly!' or 'Great question!' or 'Of course!' — begin directly with the content"
2. **Kill corporate speak**: "Do NOT use: leverage, synergy, robust, cutting-edge, game-changer, revolutionary, or seamlessly"
3. **Stop over-explaining**: "Do NOT explain concepts I already understand. Assume I know [X, Y, Z]"
4. **Prevent hedging**: "Do NOT say 'it depends' without immediately following with your best recommendation given the information available"
5. **Avoid repetition**: "Do NOT repeat the same point in different words. State it once, clearly"
6. **Stop padding**: "Do NOT add a summary section — the content should be clear enough without one"
7. **Prevent generic advice**: "Do NOT give advice that would apply to any business. Be specific to MY situation"

Generate the output with and without negative prompts and compare.

Then help me build a personal "banned list" — the specific bad habits that ChatGPT falls into most often for MY type of tasks.

I wrote a prompt that didn't give me the results I wanted. Help me debug it.

My original prompt:
[paste the prompt that didn't work]

What I expected:
[describe the output you wanted]

What I actually got:
[describe what went wrong — too generic, wrong format, too long, wrong tone, missing key info, etc.]

Diagnose the problem using this checklist:
1. **Ambiguity check**: Are there any words or phrases that could be interpreted multiple ways?
2. **Context gap**: What context did I assume ChatGPT would have but probably doesn't?
3. **Instruction conflict**: Are any of my instructions contradicting each other?
4. **Scope issue**: Is the task too broad (ChatGPT doesn't know what to prioritize) or too narrow (not enough room to be useful)?
5. **Missing constraints**: What constraints would have prevented the bad output?
6. **Format gap**: Did I specify the output format clearly enough?
7. **Role confusion**: Would a different role assignment fix the issue?
8. **Temperature mismatch**: Is the task creative (needs more variation) or precise (needs more determinism)?

For each issue found, rewrite that specific part of the prompt. Then show me the complete debugged prompt and explain why each change should fix the problem.