Current Issue
This Month's Print Issue

Follow Fast Company

We’ll come to you.

2 minute read


What If There Was A Bot That Could Detect Ugly Websites?

German firm EyeQuant is using machine learning to rank the world's aesthetically pleasing (and displeasing) web pages.

What If There Was A Bot That Could Detect Ugly Websites?

[Image via Shutterstock]

EyeQuant is a startup which isn’t too unusual in the fact that it deals with machine learning and artificial intelligence and has big-name clients like Google and Spotify. But the company’s current fascination—using machine learning to train their AI to recognize bad aesthetics and poor website design—takes it into uncharted waters. "We use machine learning and computational neuroscience to build predictive models of how humans look at web ites," founder Fabian Stelzer told Co.Labs. "We focused on attention before but we are now branching out to more general things like why people prefer one image instead of another or what are the factors that drive trustworthiness of image." And he’s betting that machines can be trained to detect web pages that most of us think are ugly.

EyeQuant, whose web eyetracking has previously been featured in Fast Company, is betting on one simple fact: Once a sample size reaches a certain size, Internet users in English-speaking countries tend to identify the same sorts of websites as ugly or aesthetically cluttered. Stelzer told me that the company based their new machine learning platform on a sample of several thousand Internet users in the United States, Canada, Great Britain, Australia, and elsewhere. As Stelzer, whose company is based in Berlin, freely admits, the AI ranks sites according to the aesthetics of English-speaking desktop Internet users. Sites are than given a score along a scale of 0-100 depending on how cluttered they are, with 0 being the worst.

Click to enlarge

"If we used a different data sample, say 3,000 people from Shanghai, it might lead to a completely different AI because their perception of clutter is completely different," Stelzer addded. "But if you look at the Chinese or Japanese web, it all looks very cluttered which compares to what we found in our data." Because Chinese and Japanese are written differently than Latin-based languages, Asian web aesthetics differ significantly. But Berlin-based EyeQuant says their AI ranked Scandinavian websites, rather than German or American sites, as the most aesthetically pleasing.

According to an analysis of Alexa’s top 10,000 websites, EyeQuant’s analysis found that Scandinavian websites were the cleanest, followed by British and German sites. Among major commercial websites, Apple’s pre-iOS 8 page received a 94% score, while poor Best Buy only received 33%.

Click to enlarge

EyeQuant claims their AI can rank a website’ design clarity at 95% the accuracy of a human observer, and are using the algorithm-generated scores (referred to as Clarity scores) as part of an online service designed to help commercial websites and designers tweak their web presences to maximize conversion rates.

Next up for the company is the release of a Clarity-score product for mobile sites; Stelzer says the mobile edition should be launched later in 2014.