Icecat Streamlines Web Usability Getting Tips from AI

In a recent internal review, Icecat’s usability team turned to AI to identify and address key usability challenges across its three main web properties: Icecat.biz, Icecat.com and Iceclog.com. By using ChatGPT as a virtual analyst, we conducted a comprehensive audit. It uncovered a series of structural and performance issues affecting both user experience and accessibility. Nowadays a hot topic, as the EU Web Accessibility Directive and Act oblige improved website usability, especially accessibility.

Timely analysis

As classical web analytic tools, like Semrush, fail to monitor accessibility aspects, we turned to a general AI instead. So, our project began with a straightforward process: live website snapshots and developer-oriented prompts were fed into ChatGPT. In return, the AI provided detailed feedback on technical usability, highlighting areas such as semantic HTML, mobile responsiveness, performance optimization, and accessibility compliance. What would typically take days of manual inspection and documentation was condensed into a few interactive sessions.

Usability tips by AI

One of the findings concerns the outdated use of table-based layouts as product categories are sometimes presented in a rigid grid structure. This approach not only complicates responsive design on mobile devices but also limits semantic clarity for screen readers and search engines. ChatGPT recommends replacing this layout with a modern, CSS-based grid system to improve both accessibility and adaptability.

On the main homepage it reveals issues of a different kind. The site opens with a rotating carousel of brand logos—visually repetitive and devoid of meaningful user engagement. Compounding the problem is the absence of a clear headline or primary call-to-action. Creating friction for new visitors who may be unsure of where to start or how to interact with the site.

Multiple websites suffer from performance bottlenecks. This can easily be fixed with lazy-loading for off-screen images and the use of multiple render-blocking JavaScript files. On mobile devices, these inefficiencies contribute to slower load times and a clunky interface. Touch targets were also too small in many areas, particularly in language selectors and navigation links, violating basic usability guidelines for mobile interactions.

ARIA landmarks

Accessibility emerged as another critical theme. Neither site provided skip-navigation links, ARIA landmarks were inconsistently applied, and many interactive elements lacked descriptive labels for screen readers. Focus outlines, which help keyboard users track their position on the page, were either missing or poorly styled. An ARIA landmark is an HTML attribute that defines a region of a web page with a specific purpose or role. Such landmarks help assistive technologies – like screen readers – better understand the structure of the page and allow users to navigate by region. Why it matters: visually, users can scan and scroll pages. But for blind or low-vision users using screen readers, ARIA landmarks allow quick navigation between key sections like navigation menus, main content, search bars, footers, etc.

Efficient insights

Beyond identifying problems, ChatGPT offers concrete recommendations. It advises implementing JSON-LD structured data to improve search engine visibility and encouraged the adoption of lazy-loading and asset bundling tools like Webpack or Vite to enhance performance.

Perhaps the most unexpected takeaway was the efficiency of using AI as a web analyst. ChatGPT translated technical diagnostics into a prioritized to-do list, complete with explanations and rationale. Instead of trawling through raw audit data, Icecat’s team received a developer-friendly roadmap that’s now guiding frontend improvements.

With these insights, Icecat is preparing updates aimed at making the portals faster, more accessible, and more intuitive for its global user base. As the next phase begins, the team will re-engage the AIs it isusing, including Claude of Anthropic, to validate progress – turning this new AI experiment into a continuous improvement loop.