Every designer faces this problem: what are the current trends? What's working, what's not, what's bleeding edge, what's fading?
We've all recognized—hopefully—that we live in our own bubble. Visiting the same sites, gathering data points from the same blogs and opinions we're familiar with. How do we break out of that? How do we see what the broader internet has to offer without opinion, without confirming the trends we want to confirm?
That's exactly why UXC Trends exists. I wanted to know, objectively, what the true trends are. Not someone's opinions. But how?
I ran across a site with thousands of screenshots and thought: wouldn't it be cool to extract trends from these to understand the state of design and where things are going?
First I had to build the infrastructure. What kind of processing power would I need? What algorithms? How do I extract data from a screenshot? How do I format the output? All these questions needed answers before I could even start.
Once I figured out the processing power and base infrastructure—database, scraper, screenshot capture—it all needed to be strung together.
The process starts with finding sites. There are plenty of resources online showing top sites across various filters. Why rebuild what already exists? By leveraging those lists, I could start the pipeline: capture screenshots, store them with metadata, then process and extract usable data.
But how do you get meaningful data from a screenshot?
I'd done a project for an accountant trying to extract data from "standardized" forms—except those forms were far from standardized. Boxes in different locations, filled out differently, OCR struggling with all of it. About 80% accuracy. I knew I couldn't rely on that approach.
AI vision models are actually quite accurate. Air-gapped local AI got me to 97% on those tax forms. Higher-level models could handle design extraction no problem. So AI became part of the pipeline.
Stringing all of this together in sequence took careful planning—inputs and outputs at each phase, tweaking prompts for better results, normalizing data points across wildly different websites.
And all of it costs money. Infrastructure, processing power, database storage and I/O, AI API costs. Keeping it manageable was a task in itself. I tested dozens of models looking for quality output that wouldn't cost an arm and a leg per screenshot. Database optimizations, processing utilization, repurposing old machines for mundane tasks that older processors can handle.
Dealing with AI is a unique skill. You can get wildly varying outputs depending on the input—and it doesn't take a hugely modified prompt to go from usable to garbage. Testing prompts is time-consuming and expensive, but necessary to get quality output.
Then there's the complexity of the web itself.
Websites vary massively. Design, layout, structure, technology. Regional differences, languages, RTL layouts. Auth walls. SPAs that don't load everything at once. Modals, cookie banners. All of it needed to be accounted for during extraction. Errors and edge cases everywhere.
This "weekend project" became considerably more than a weekend.
But the output has been rewarding.
I can now see what's actually trending. What percentage of the top 10,000 sites use specific components, layouts, color palettes. What's fading away. All objective insights flowing through the system and being recorded. Real-time tracking with historical data to understand the true state of design—where we are, where we're headed, where we've been. No opinion.
Most designers have noticed the "sameness" problem. It started with the rise of design systems and has grown exponentially. Sites increasingly look alike.
That's because AI is taking over a lot of design work, and AI works based on patterns. So we get the same mix: Inter font, blue and purple gradients, shadcn components everywhere.
True design is becoming rare. The sameness is gaining traction.
There's an argument for it—users know the pattern, it's easy to navigate because it's everywhere. But it kills uniqueness. It actually makes distinctive, well-executed designs more memorable. Being different from the AI pattern madness is a good thing if done well.
And here's the thing: all this sameness is actually a boon for UX and design. It lets the industry shine because AI can never replace the human touch. The empathy for users that UX people have. That'll never be programmed into a machine.
I built this because I was curious and wanted the data for myself. It turned into something more when I realized others could benefit too.
Even at a cost to me, I'd love to see UX make a resurgence and show how human designers are superior to AI slop. So I'm planning to give basic trends and metrics away free for anyone interested.
When UXC Trends launches, you'll be able to download a free quarterly report—or upgrade for deeper data with verticals, regions, and more insight for a nominal cost. Gotta pay those bills somehow.