Just-in-Time Content Generation: Reimagining Website Creation
Time for an update on my AI experiments in web design...
As we all know, 2024 saw an obvious surge in AI assistants in traditional web design workflows - automating first-time user experiences, applying themes, and generating copy and images. Most implementations I've tested offer good levers to pull; however, real-world usefulness for established enterprises remains questionable - mostly because the results fail to keep a site on brand (assuming the brand is well defined and arrives to the product with its own assets and designs).
One thing I've also not seen in my research is what I think is still the fundamental shift - something I'm going to call "just-in-time page generation." Over the Christmas break I leaned more into these ideas and built some demos to test the concept.
The basic premise was relatively straightforward: document the brand, the content, and the design system. Push it through a handful of AI calls and generate a web page.
While the speed isn't viable (yet?) for a truly just-in-time approach, the experiments pushed me toward some interesting revelations:
1. Content is King (More Than Ever)
The success of generative website creation lies in content management, but it's deeper than just having good copy. It's about creating a rich, well-structured knowledge base that AI can understand and transform.
This means detailed brand guidelines, comprehensive content strategies, and clear editorial standards that AI can use to maintain consistency across all generated experiences. It includes research, design systems, media, icons, and fonts. And it continually evolves to ensure relevance and remain competitive.
Without this foundation, AI simply doesn't have the context needed to create something that feels authentic to your brand.
2. Quality Takes Time
Like tasking a human with work, generative AI requires time to produce quality results. I tried several LLMs, all with different nuances and challenges. Token limits, latency, and accuracy were the biggest issues, but also temperature and consistency in the results.
However, with the right breakdown of tasks and time (i.e., giving the AI the opportunity to self-correct), the results can be quite accurate. As such, the UX should adapt to this reality and avoid request/reply conversational experiences for consumer output.
This contradicts our common expectation of AI as "instant" - we need to design experiences that account for the time needed to generate high-quality results.
Rethinking the Generation Process
The most successful approach I found was to break down the generation into discrete steps:
- Analysis phase: AI analyses the brand guidelines, content strategy, and design system to understand the context
- Planning phase: AI creates a structured plan for the page, including content blocks, messaging hierarchy, and visual elements
- Generation phase: AI produces the actual content, with self-review built into the process
- Refinement phase: The final output is reviewed and refined before delivery
This multi-step approach consistently produced better results than trying to generate everything in one go, even when using the same models.
The Integration Challenge
The technical implementation isn't trivial. Current web platforms aren't designed for this kind of generation process. Most are built around the assumption that humans will manually create and arrange content.
To make just-in-time generation work, we need platforms that can:
- store and manage structured brand/content knowledge
- provide well-defined component libraries that AI can use as building blocks
- support asynchronous generation processes that may take seconds or minutes
- allow for human oversight and intervention where needed
Looking Ahead
While true just-in-time generation might not be feasible for production use today, we're closer than many people realise. The core technologies exist - they just need to be integrated in new ways.
For enterprises exploring this approach, focus first on building that comprehensive knowledge base and component library. These will be valuable regardless of how quickly full automation arrives, and they'll position you to take advantage of advances in AI capabilities as they emerge.
The future of web design isn't about faster manual creation - it's about creating the systems and structures that enable dynamic, intelligent content generation tailored to each user's needs.
