AI Prompting Cheat Sheet — Get Better Results from Any AI

Status: 🟩 COMPLETE 🟦 LIVING Section: cheat-sheets Tags: prompting, cheat-sheet, prompt-engineering, AI-tips, productivity


The five techniques that fix 80% of bad AI results

  1. Give context first. Tell the AI who it is, who you are, what you’re trying to do — before asking the actual question.
  2. Be specific. “Make it better” is useless. “Make this 30% shorter, more formal, and focused on the financial benefits” is actionable.
  3. Show, don’t tell. Give examples of what you want.
  4. Ask for the format you want. “As a bulleted list,” “as a table with columns X, Y, Z,” “in plain English.”
  5. Iterate. The first response is rarely the best one. Refine, redirect, ask for variations.

Quick prompt templates

For writing tasks

You're helping me write a [type of document] for [audience].
The tone should be [tone].
The length should be approximately [length].
Key points to include:
- [point 1]
- [point 2]
- [point 3]
Things to avoid:
- [avoidance 1]
- [avoidance 2]
Now write a draft.

For analysis tasks

Analyse the following [text/code/data] for [specific aspects].
Focus particularly on [areas of concern].
Format your analysis as:
1. Summary (1-2 sentences)
2. Key findings (bulleted list)
3. Recommendations (numbered)
4. Caveats or things you couldn't determine
[paste content]

For decision-making support

Help me decide between these options:
Option A: [description]
Option B: [description]
Option C: [description]
My priorities are: [priorities]
My constraints are: [constraints]
Don't recommend yet — first, lay out the pros and cons of each.

For learning something new

I want to understand [topic].
My current level: [beginner / some background / expert in adjacent field].
I learn best through [analogies / concrete examples / step-by-step / theory first].
Start with the simplest version, then add complexity. 
After your explanation, ask me one question to check my understanding.

For code

Help me [what you want to build/fix].
Constraints:
- Language/framework: [e.g., Python, React, etc.]
- Existing code style: [paste or describe]
- I want the code to [be readable / be performant / handle edge cases / etc.]
Show me the code with comments explaining the non-obvious parts.

For emails and difficult communication

Help me write an email to [recipient] about [topic].
Context: [why this is happening, the relationship]
What I want: [outcome you're hoping for]
Tone: [professional, warm, firm, etc.]
Length: [short / medium / long]
Constraints: [don't mention X, do mention Y]

Power moves

Role play / persona

Start prompts with: “You’re a [role]. Help me [task]…”

  • “You’re a senior accountant explaining to a small business owner…”
  • “You’re a sceptical investor reviewing this pitch deck…”
  • “You’re an Australian tax adviser…”

The persona shapes tone, depth, and the perspective the AI takes.

Specific output format

  • “Format your response as a table with columns: Pros, Cons, Cost Estimate”
  • “Give me your response as a bulleted list with each bullet under 15 words”
  • “Respond in Australian English with proper formal tone”

Audience targeting

“Explain this for [audience]:”

  • a 12-year-old
  • someone with no technical background
  • a CEO who needs the bottom line first
  • a developer reviewing code

Constraints that improve quality

  • “Cite specific evidence for any claims”
  • “Be honest if you’re uncertain”
  • “Don’t make things up — if you don’t know, say so”
  • “Mention any important caveats or risks”
  • “Suggest 3 alternatives”

The “draft + edit” pattern

  1. “Draft a [thing]. Don’t worry about polish yet.”
  2. Review the draft.
  3. “Edit this draft: make section 2 stronger; cut section 3 in half; add a more compelling opening.”

This is often faster than trying to get a perfect output on the first try.


Common mistakes to avoid

MistakeFix
”Make it better”Specify what “better” means
Asking yes/no questionsAsk “what are the considerations”
Single-shot for complex tasksBreak into steps
Not providing contextSpend the first 2-3 sentences on context
Accepting the first responseIterate; ask for alternatives
Trusting facts without verificationAI hallucinates; verify important facts
Asking for “10 ideas”Ask for 20, then pick the best 5
Being too polite (“could you maybe please”)Be direct (“write X”)

Few-shot prompting (showing examples)

For consistent output style, give examples in your prompt:

Here are two examples of the style I want:

Example 1:
Input: [example input]
Output: [example output in the style I want]

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

Now do the same for this:
Input: [your actual input]
Output:

This works dramatically better than describing the style abstractly. AI is better at mimicking patterns than following style rules.


Chain of thought / step by step

For complex reasoning tasks:

  • “Think through this step by step before answering.”
  • “First, list what you need to know. Then work through it.”
  • “Walk me through your reasoning, then give the final answer.”

This produces meaningfully better results on multi-step problems — even for AI models that don’t have explicit “thinking” modes.


Reverse prompting

If you’re stuck, ask the AI:

  • “What questions should I be asking about this topic?”
  • “What context would help you give me a better answer?”
  • “What are 3 different ways I could ask this to get different perspectives?”

The AI often knows what would help you better than you do.


When to use which AI

TaskBest toolWhy
Writing emails, reports, essaysClaudeBest writing voice
Image generationChatGPT (DALL·E) or MidjourneyBuilt-in or higher quality
Current informationGemini or ChatGPTReal web search
Long documents (>50 pages)Gemini 2.5 Pro (1M context)Largest context
CodeClaude or CursorBest at code reasoning
Quick factual questionAny free tierAll are good enough
BrainstormingWhichever you preferPersonality matters most
Sensitive contentClaude or local modelStrongest privacy

Australian-specific prompting tips

Save these as custom instructions in ChatGPT, Claude, or Gemini:

  • “I’m an Australian based in [city].”
  • “Use Australian English spelling (organisation not organization, etc.).”
  • “Use AUD when discussing prices unless otherwise specified.”
  • “Reference Australian context where relevant (ATO, ACCC, Medicare, Centrelink, state laws, etc.).”
  • “Australian time zone is [AEST/AEDT/AWST/ACST].”

This saves you re-explaining every conversation.


Prompt patterns for specific scenarios

”I have data, what should I notice?"

I have [data]. I'm trying to understand [question].
1. What patterns or anomalies stand out?
2. What might be causing them?
3. What additional data would help clarify?

"I’m planning [thing], what am I missing?"

I'm planning to [thing]. Here's my current plan:
[plan details]
What am I likely missing? What could go wrong? 
What questions should I be asking that I'm not?

"Critique this"

Critique the following [content] as if you were [type of critic].
Be specific about strengths and weaknesses.
Identify the 3 most important things to improve.
[content]

"Explain it three ways"

Explain [concept] three different ways:
1. As an analogy to something everyday
2. With a specific concrete example
3. In technical terms (if applicable)

"Steel man the opposing view”

I think [position]. Steel man the opposing view as strongly as possible.
What's the strongest case for the opposite position?
What evidence would convince me to change my mind?

When the AI is being annoying

Common AI failure modes and fixes:

ProblemFix
AI is wordy and adds disclaimers”Skip preamble; be direct”
AI keeps asking clarifying questions”Make reasonable assumptions and proceed”
AI hedges every answer”Be confident in your best judgment”
AI suggests “consult a professional” excessively”Give your best analysis; I’ll judge if I need a professional”
AI is too formal”Be conversational and direct”
AI is too casual”Be professional and formal”
AI ignores parts of your promptNumber your requirements; repeat key constraints
AI repeats the same responseOpen a new conversation; the context may be stuck

What NOT to do

  • Don’t share passwords, account numbers, full credit card numbers, or sensitive ID info
  • Don’t paste confidential client data without enterprise-tier protections
  • Don’t trust factual claims without verification (especially citations, statistics, recent events)
  • Don’t expect consistency across conversations unless memory is enabled and reliable
  • Don’t assume the AI knows your specific situation — tell it
  • Don’t accept legal, medical, or financial advice without consulting a qualified professional

See also


Sources

  • Anthropic Prompt Engineering documentation
  • OpenAI Prompt Engineering guide
  • “Prompt Engineering Guide” (promptingguide.ai)
  • Personal experience across thousands of AI prompts (2023–2026)