The Best AI Prompt Framework For Work (2026)

Prompts for AI can come in many forms for different outcomes.

Looking for a quick answer to a simple question? Use a 1-2 sentence prompt. Want to unpack a complex work task with layers of actions? You need a different approach.

Whereas a one shot prompt like “What’s the weather in Ibiza on average in August?” is a simple string of words for any conversational AI tool to answer. Asking it to review, ideate and share how to create a learning strategy for your company is not so straightforward.

To get the best results from something like a ChatGPT, you need 3 things:

• Context
• Task outline
• Constraints

3 prompting techniques you need to know

Let’s take some of the techy terms you might have heard and translate for humans.

1️⃣ Zero-shot

Useful for: Simple Q&A inputs

Example: “On average, how long do Alaskan Huskies live?“

2️⃣ One-shot

Useful for: When your task requires a specific output such as a table or executive summary with one examples as guidance.

Example: “I want an executive summary on the current population of Alaskan huskies in North America. The summary needs to follow our company format, here’s an example:

‘Insert your example’

Now, let’s create a first draft”

3️⃣ Few-shot

Useful for: Working on highly complex tasks like data analysis or writing a report for senior leaders using 2 – 3 examples.

Example: “I need to produce an analysis of the most popular buying locations of Alaskan Huskies in North America from European origins. This will need to be in our preferred company format with supporting data visualisations.

Here’s a few examples to show what I want to achieve:

*Either insert the examples directly into the input bar or upload as images, Docs and PDFs

Note: No prompt is bulletproof. LLMs can behave in extraordinary and odd ways. If you find yourself hitting that brick wall, deploy these techniques.

How to write the best AI prompts for business tasks

This is a universal approach you can use with any tool.

Let’s unpack each of these:

Context

What does the LLM need to know to successfully support you?

Here’s some ideas:

• Your organisation
• Team
• Roles and work they do?
• Specifics on the task
• What have you done before?
• What is the role it’s playing? [If role-playing or coaching]

Task

• Outline the task
• What does success look like?
• What are the essential components of the task?
• Keep it clear and simple
• How should the output be structured? Bullets, sentences or paragraphs.

Constraints

• What should the LLM not focus on?
• What must it not consider?
• Should it only use its training data or connect to the internet or both?
• Should it only use the data you’ve provided?

Example of a powerful AI prompt for business tasks

If we put this all together, it can look something like this:

###Context###

I'm crafting my organisations L&D strategy for the year ahead. We’re a scale-up business with 800 employees in 5 global offices. We’re limited with our resource and budget to deliver. Our goal is not to do everything, but do the top 2-3 things that matter most.

Our strategy for the last few years has become stale and not working towards what we want to achieve.

Top things on our employees minds include:

- Having the right skills for the role

- Learning from their peers

- Manager support and coaching


###Task###

You will help me get clarity on how I can work with my team to build a relevant and meaningful strategy for our organisation.

To do this, you will ask me questions to help get clarity to build a better picture of where we can go.


###Constraints###

Keep questions short and relevant. They should be quick fire rather than in-depth. Let’s keep questions to a minimum of 3 at a time.

You might notice I’m using ‘markdown’ structure and delimiters in these prompts. This helps AI tools better understand the instructions you provide by using heading, bullets and general formatted structure etc.

This works with all popular AI tools including ChatGPT, Claude, Perplexity and all the ones you know.

Why you should use delimiters in AI prompts

Delimiters help AI models recognise that what follows or is enclosed within them is a directive or special instruction that should guide its response.

Here’s 6 ways they improve AI prompts

1. Clarifies Instructions: Separates commands or instructions from regular text, reducing ambiguity.
   
   
2. Enhances Accuracy: Helps LLMs focus on the specific parts of the prompt, improving response relevance.
   
   
3. Organises Complex Prompts: Structures detailed prompts into distinct sections for better comprehension.
   
   
4. Ensures Consistent Formatting: Maintains desired format in outputs, like code or structured data.
   
   
5. Highlights Key Elements: Marks important parts of the text, such as commands or keywords.
   
   
6. Reduces Misinterpretation: Provides clear boundaries to prevent confusion in interpreting different content types.

Common Delimiters to use with AI tools

Here is a list of common delimiters that are often used effectively with large language models (LLMs):

1. Triple Hash (###): Often used to separate sections or indicate instructions.
2. Angle Brackets (<...>): Used to highlight specific tags or commands.
3. Double Curly Braces ({{...}}): Can indicate placeholders or variables.
4. Backticks (’...‘): Commonly used to denote code or special terms.
5. Double Hyphens (- ... --): Sometimes used for comments or notes.
6. Pipe Symbols (|...|): Useful for delineating choices or options.
7. Square Brackets ([...]): Can denote optional elements or clarifications.
8. Double Quotes ("...): Used for direct quotes or exact text.
9. Colons and Semicolons (: ... ;): To separate elements in lists or statements.
10. Parentheses ((...)): Often used for additional information or clarification.

You can learn more about that in this video.

Reasoning Models vs GPT Models: What’s the difference?

Not all AI models work the same way.

Some are designed to think deeply before responding, while others prioritise speed and efficiency.

The two main types you’ll come across are Reasoning Models (like o1 and o3-mini) and GPT Models (like GPT-4o). They have different strengths and are suited to different tasks. 

Here’s how they compare:

Reasoning Models: The Deep Thinkers

These models are built to process complex problems step by step before responding. Think of them like a senior colleague who takes their time to analyse a situation before giving you a well-thought-out answer.

What they do well:

→ Handle ambiguity and figure out what you mean, even if your instructions are unclear

→ Solve complex problems by planning multiple steps ahead

→ Make reliable decisions based on lots of scattered information

→ Ask clarifying questions rather than guessing when details are missing

Best used for:

• Strategic problem-solving
• Sifting through huge datasets to find key insights
• Multi-step planning and execution
• Debugging code and reviewing AI-generated responses
• Visual reasoning tasks

GPT Models: Fast and Furious

GPT models, on the other hand, are built for speed and efficiency.

They work best when tasks are well-defined and don’t require deep reasoning.

What they do well:

→ Quickly generate content, summaries, and responses

→ Follow explicit instructions without overthinking

→ Handle repetitive tasks efficiently

Best used for:

• Writing, editing, and summarising information
• Answering straightforward questions
• Generating content when speed matters more than accuracy

Think of them like a junior co-worker: great when given clear instructions, but not someone you’d rely on to figure out a complex problem without guidance.

When should you use a reasoning model?

It depends on your needs:

• If accuracy and careful thinking are the priority → Use a Reasoning Model
• If speed and cost matter more than deep problem-solving → Use a GPT Model

In many cases, the best approach is a mix of both. A reasoning model can do the heavy lifting—like analysing and strategising—while a GPT model handles execution tasks quickly and efficiently.

Now you know the difference, you can use the right tool for the right job.

How to Prompt Reasoning Models (7 tips)

1. Keep prompts simple and direct

Reasoning models work best with brief, clear instructions. Avoid unnecessary complexity.

2. Avoid chain-of-thought prompts

Prompts like “think step by step” or “explain your reasoning” are unnecessary. These models already reason internally, and such instructions may even reduce performance.

3. Use delimiters for clarity

Use markdown, XML tags, or section titles to separate different parts of the input. This helps the model interpret sections correctly.

4. Start with zero-shot, then try few-shot if needed

Begin with prompts that don’t include examples—reasoning models often don’t need them. If your task is complex, add a few clear input-output examples that align closely with your instructions.

5. Provide specific guidelines

Clearly define any constraints, such as “propose a solution with a budget under $500”, to ensure precise responses.

6. Be explicit about your end goal

Set clear success criteria and encourage the model to iterate until the response meets your expectations.

7. High-level guidance works best

Reasoning models perform well when given a high-level goal rather than micromanaged steps—trust them to figure out the details.

Get more best practice in OpenAI’s official documentation.

What ‘good’ looks like

Here’s an example of a prompt with OpenAI’s o1 reasoning model by Greg Brockman, President of OpenAI.

Before you go… 👋

If you like my writing and think “Hey, I’d like to hear more of what this guy has to say” then you’re in luck.

You can join me every Tuesday morning for more tools, templates and insights for the modern L&D pro in my weekly newsletter.