We all probably think we know how to use AI. You ask for what you want and it does it, right? Well... not really. That approach won't always get what you want because AI needs to be prompted in specific detailed ways to get the best results.
To confuse matters, the way you prompt actually varies depending on what it is you're trying to achieve.
Fear not, this blog is here to help. We'll demystify the whole process and teach you how to build great prompts for every occasion using prompt frameworks as a checklist for each and every prompt you write. The key is to give the AI enough context, be specific about the task, and explain the format or outcome you want.
Let's get into it.
It's worth stating upfront that there are no universal industry accepted prompting frameworks. All the frameworks below are just helpful tools that you can use when learning how to get the best out of AI. You will see that they heavily overlap so the takeaway should generally be 'give the AI more, not less'.
Regardless of which you use, they will all help you craft better prompts. A quick example below shows you the difference between a basic poor prompt and well crafted one that follows a framework to provide a better result:
Weak prompt:
Write a customer email about downtime.
Better prompt:
Act as a customer success manager. Write a short customer email explaining that this morning’s service disruption has now been resolved. Keep it under 150 words, use plain English, apologise for the inconvenience, and avoid technical detail.
So now you know why you're using these frameworks and what you can expect as we go through, let's start with a simple comparison table to see, at a glance, which framework is right for you. Click the framework to jump down to the appropriate part of the blog.
|
Framework |
Best used for |
Why it helps |
Good example use |
|
Quick everyday prompts |
Keeps the request simple and focused. |
Summarising notes, drafting a short reply, or generating ideas. |
|
|
Precise, repeatable requests |
Adds useful context, boundaries, and examples. |
Client updates, incident summaries, or standard reports. |
|
|
Role-based thinking |
Helps the AI answer from the right professional perspective. |
Asking for advice as a finance lead, marketing manager, or security specialist. |
|
|
Flexible business prompts |
Balances context, role, action, format, and audience. |
Reports, proposals, internal guidance, or customer-facing content. |
|
|
Creative and marketing content |
Defines the purpose, audience, output, and style. |
Blog posts, social media, campaign copy, and web content. |
|
|
Complex workflows and AI agents |
Breaks the work into steps and sets clear limits. |
Triage processes, multi-step analysis, or structured automation. |
We'll look at each in turn, explaining their benefits and showing you a good example of each.
We've added tags through each example with [square brackets] around them just so you can see how they break down. You don't need to include these yourself when writing the prompt. It's just the overall structure that you need to think about.
TAG is a great starting point if you are new to AI prompting. It keeps things simple by making you explain what the task is that the AI is helping you with, the action you want the AI to take, and the goal you are trying to achieve.
This works well for everyday tasks where you do not need a long brief. Instead of asking a broad question, you give the AI a clear job to do and a clear outcome to aim for.
For example, a team member could use TAG to ask the AI to act as a content assistant, summarise a set of meeting notes, and turn them into five customer-friendly action points. It is quick, easy to remember, and ideal for building confidence across a team.
Example prompt: [Task] Act as a meeting assistant. [Action] Summarise these meeting notes into five clear action points. [Goal] Create a customer-friendly follow-up email.
GCSE is useful when you need a more accurate or repeatable answer. It works well for business tasks where tone, structure, or consistency matter because it gives the AI four clear building blocks: goal, context, scenario, and expectation.
The goal explains what you want to achieve. The context gives the background the AI needs to understand the wider situation. The scenario sets out the specific circumstances the response should address. The expectation defines what a good output looks like, including tone, length, structure, or level of detail.
This makes GCSE especially helpful for customer updates, incident summaries, internal processes, and any task where you want consistent results from different people across the business.
For example, you could ask the AI to write a short customer update about a resolved service issue, provide the business context, explain the situation, and set clear expectations for tone and length.
Example prompt: [Goal] Write a short customer update. [Context] This is for customers affected by this morning’s service disruption. [Scenario] The issue has now been resolved and service is back to normal. [Expectation] Keep it under 150 words, use plain English, avoid technical language, and match the tone of our previous service update.
RACE is best when you want the AI to think from a particular professional viewpoint. By giving it a role, an action, context, and an expectation, you can make the response more relevant to the person or team who will use it.
The role might be a finance director, customer success manager, cyber security consultant, or HR lead. This helps the AI focus on the right priorities and use the right language for that audience.
The context then gives the AI the background it needs, while the expectation explains what the finished answer should include and how it should be presented.
For example, you could ask the AI to act as a customer success manager, review feedback from a client meeting, identify the main risks and opportunities, and present the output as a short account update for an internal team.
Example prompt: [Role] Customer Success Manager. [Action] Review these meeting notes. [Context] The customer is considering adopting a new product. [Expectation] Summarise the key risks, opportunities, and recommended next steps for our internal team.
CRAFT is a strong all-round framework for business use. It gives the AI enough structure to be useful without becoming too complicated for everyday tasks.
The context explains the background. The role tells the AI how to approach the task. The action sets out what you want it to do. The format describes how the answer should be presented, and the target explains who the output is for.
This makes CRAFT especially helpful for reports, proposals, internal guidance, customer emails, and other content where the same information may need to be adapted for different audiences.
For example, you could ask the AI to use the context of your managed IT services business, act as a customer-facing consultant, write a short proposal section, format it with headings and bullet points, and target it at non-technical business owners.
Example prompt: [Context] We provide managed IT support to SMEs. [Role] You are a customer-facing IT consultant. [Action] Write a short proposal section about proactive cyber security monitoring. [Format] Use a heading and three bullet points. [Target] Non-technical business owners.
ROSES is most useful for creative work, especially marketing and communications. It helps the AI understand not just what to write, but why the content matters and how it should feel to the reader.
The role sets the creative perspective, such as copywriter or brand strategist. The objective explains the purpose. The scenario gives the audience context. The expected output defines what you need, and the style keeps the tone on brand.
This is helpful because creative work can be subjective. ROSES gives the AI enough direction to stay focused while still leaving room for fresh ideas and engaging wording.
For example, you could ask the AI to act as a B2B technology copywriter, write a blog introduction for business owners worried about cyber security, explain the scenario, produce 150 words, and use a friendly, expert tone in British English.
Example prompt: [Role] B2B technology copywriter. [Objective] Write an engaging blog introduction about cyber security for SMEs. [Scenario] The reader knows security matters but feels overwhelmed by the options. [Expected Output] 150 words. [Style] Friendly, expert, and written in British English.
RISEN is designed for more complex tasks, especially where you want the AI to follow a process rather than answer a single question. It is useful for AI agents, structured workflows, and anything that needs clear steps and boundaries.
The role tells the AI what kind of assistant it should be. The instructions explain how it should behave. The steps break the work into a sequence. The end goal defines success, and narrowing explains what the AI should avoid or escalate.
This matters because complex prompts can easily drift if the scope is not clear. RISEN helps keep the AI focused, safe, and consistent throughout a longer task.
It is particularly useful when you want the AI to triage requests, analyse information in stages, or produce a repeatable output with minimal supervision.
For example, you could ask the AI to act as a first-line support assistant, review incoming tickets, classify urgency, suggest standard responses where appropriate, escalate security incidents immediately, and avoid making changes that require administrator approval.
Example prompt: [Role] First-line support assistant. [Instructions] Review the ticket and suggest a response only if it matches a known issue. [Steps] Identify the issue, check urgency, suggest a standard reply, and flag anything security related. [End Goal] Help the team respond quickly. [Narrowing] Do not suggest administrator changes or security configuration changes.
There is no single best prompt framework. The right choice depends on what you are trying to do, how much detail the AI needs, and how important consistency is.
If you are getting started, use TAG. If you need more precision, use GCSE. If the answer needs a particular professional viewpoint, use RACE. If you are creating marketing or creative content, use ROSES. If the task is complex or process-led, use RISEN. If you are unsure which to use, fall back to CRAFT as it it the most flexible of them.
Many businesses will use more than one framework. A simple team guideline can help people choose the right approach for the task, rather than relying on trial and error.
The main thing is to move away from vague prompts and towards structured instructions. Once people understand the basic building blocks, AI becomes more reliable, more useful, and much easier to adopt across the organisation (although don't forget it still needs oversight as it can still be wrong sometimes!).
Start small, practise often, and save the prompts that work well. Over time, your team will build a practical library of prompts that support everyday work, improve consistency, and help everyone get more value from AI.
As always, bear in mind that even when using these frameworks, AI can (and often does) still hallucinate and make errors. Always check the veracity of what it produces before you go ahead and use it's output for anything in your business.
Additionally, it's worth building out an AI context. Both at the business level, and at the individual level as this greatly improves the quality of output from an AI.
If you’d like to power your business with AI then get in touch with our Edinburgh-based team of AI and IT experts who provide IT support, cyber security, consultancy, and even custom development work to connect your data and unlock the power of AI.