Key Features of the Framework

While in the early 2010s, pioneering LLMs like GPT-1 sparked the idea that we could “prompt” these models to generate output, Prompt Engineering really became a thing after the release of ChatGPT at the end of 2022, and was soon seen as an important business skill.

In early 2023, the Crafting AI Prompts Framework was founded, and you might wonder: “Is this still up to date given the speed at which Generative AI is developing”. Meanwhile, many Prompt Engineering Frameworks have been introduced.

So why still use a framework? Because a structured approach increases a model’s performance. By dividing prompts into clear sections, we reduce ambiguity and prevent the model from picking up unintended instructions. A framework also supports consistency across teams, makes it easier to share and test components, and serves as a memory aid, helping users remember what to include for better outputs. In ongoing chats, sections can be reused simply by referencing them, streamlining collaboration with the model even further.
Below is a breakdown why the Crafting AI Prompts Framework is a good approach and how this compares to other frameworks.

Component based

One of the standout features of the Crafting AI Prompts Framework is its adoption of practices from Software Engineering, notably the component-based concept. This approach brings significant advantages to organizations. It facilitates the sharing of components across different teams, promoting consistency and efficiency.
While prompts are generally tailored to address specific use cases, making them highly specific, they often contain elements that can be reused in various contexts. By deconstructing prompts into reusable components, organizations can maximize their utility and streamline the prompt engineering process. This modular approach not only enhances productivity but also ensures that high-quality elements are consistently applied across different prompts and projects.

A good practice, especially if you’re having difficulty outlining the context yourself, is to start by describing the task and then ask the model to generate clarifying questions. By answering those questions, you can incrementally build up the context window with relevant information, helping the model better understand your subject.

Register

Although training data has been collected from various countries, studies indicate that the majority is based on US styles. Providing clear guidance on output register, including writing style and tone, is important when producing content outside the US or for different cultural contexts in order to achieve appropriate results.
While most frameworks include something like a “tone of voice” or “writing style,” this is close to the register component of the Crafting AI Prompts Framework. However, register goes much further. Think of it in terms of the language you want the output to be in—or even the programming language, or the desired language level (for example, B1 English).

Non-Disclosure

Next to the elements in the Crafting Phase, the framework also advocates Non-Disclosure and making sure it is an interactive prompt (with references). This is highly overlooked, and having this strictly in place, will help and guide the organization and users to think before they prompt or even share confidential data to tools and models.

Examples

There are many frameworks that advocate to use “examples” as a main element of the prompt. While we agree this could enhance the output, and is therefore very important to consider, it should not be a standalone element. What kind of examples are we talking about? Mostly you can directly connect them to the Register or Format component, for example. It’s better practice to include them in these components, as guiding the model
effectively will be much easier and people do understand what the example is used for.

That doesn’t mean examples are not essential. Generative AI is trained on data, think of it as examples showing how it should format text, write in a specific register, and do specific tasks
based on these examples (training data). So, examples are the foundation of the models and for that reason important. Because the training data might not (think about cultural differences), reflect the way you want to have the output formatted. But connecting this clearly to one of the components described in the framework, you enhance the output even more.

Microsoft’s Prompt Engineering Framework

It’s valuable to understand the different types of frameworks that are available. We do not advocate that the Crafting AI Prompts Framework is a “must-implement”, we rather advise to use a framework, and keep consistency across your organization.

Microsoft in its prompt engineering framework, uses the elements “Goal”, “Context”, “Sources”, and “Expectations”.
Microsoft describes this as:
Goal: What do you want Copilot to do?
Context: Why do you need it, and who is involved?
Expectations: How should copilot respond to best fulfill your request?
Sources: What information or samples do you want Copilot to use?

Now that the prompt and the output are validated, we can move to the Enhancement Phase. During this phase there are two options:

Other large and smaller organizations have also introduced similar frameworks. We advise to use the Crafting AI Prompts framework as your frame of reference to see how any framework contributes to your skills in prompt engineering.