AI tools, such as GPT-4, now wield immense generative power. They are capable of accepting one type of information as input and respond with a different target type of information. As we steer this insight towards content management, we need to take into account several key aspects when considering how to use an AI tool to aid in crafting content-enabled digital experiences such as a marketing website.

Let's first look at what types of information we work with in the domain of content management:

1. Business requirements: In other words, what business goals are we aiming to achieve with our content?
2. Content schema: It provides the structural framework for our content, defining the shape of content we will be able to store.
3. Component design: When we consider how to visualize our content, we can typically identify a number of visual components such as a "hero" component, a slider component, a call to action (CTA) component, etc., and define an exact visual rendering of each such component. Designers often use tools such as Figma for this purpose and we should ideally see a very close match of visual components to a corresponding data structure in the content schema.
4. Content: Our actual marketing texts that hold the message we want to convey to recipients.
5. Frontend code: This is the software code implementation that makes the visuals from our component design step actually appear in the target device such as a web browser, for example using the React JavaScript framework. Often times, we can again observe a very close match between software code and the content structure in our content schema.

Leveraging AI tools in content management enables us to convert one type of domain information into another. For instance, a component schema could be taken up by an AI tool to generate a matching component design. Likewise, a visual component design could be transformed into corresponding React component code.

Content Schema + Prompt = Content

With Commitspark, we are enabling content editors to write a GPT-4 instruction (also known as "prompt") and generate high quality content with the press of a button.

Compared to other offerings, it is important to note here that with Commitspark, the full content schema is provided to GPT, therefore generated content conforms exactly to the expected content structure. By doing so, we can even offer context-specific instructions to be stored directly in the content schema, allowing for powerful fine-tuning of generated output. For example, we could indicate that content generated for a title field of a specific content component should never exceed a certain number of words or that a certain way of phrasing should be used when generating long-form text (see our documentation for details).

With Commitspark, we have chosen content schema to content generation over other possible transformations for our first GPT integration, as we believe this transformation offers the most value to content editors.

Stay tuned to our blog as we expand our integration further with additional transformations in the future.