Design notes on the 2023 Wikipedia redesign

Hey, I’m Alex Hollender. For the past few years I led the redesign of the Wikipedia desktop interface, which launched this past January. Below are some notes on the project and process.

2004–2019

We have to go back in time a little to properly frame this story. Back to 2004, which was the last time Wikipedia's interface changed significantly. Mark Zuckerberg had just launched TheFacebook, Gmail had just come out, and there were no smartphones (the first iPhone would come out 3 years later). Also CSS was much more limited, so websites were more simple. Two popular computers were Dell’s Dimension 2400 desktop, and their Latitude D505 laptop. For almost everyone using it, Wikipedia looked like this:

In 2010, a bunch of changes were made to increase the usability of Wikipedia for new editors (notes), and in 2015 the editing experience was again significantly improved with the introduction of the Visual Editor. However, the general layout, and specifically the reading experience, stayed largely the same between the 2004 update, and 2019, where my team’s involvement begins (the Readers Web Team at the Wikimedia Foundation — an organization of paid staff members who work on things like the Wikipedia apps, search, languages & translations, editing software, community wishlist, machine learning, and more).

However computers changed a lot between 2004 and 2019. Namely, screens got a lot smaller, and a lot larger. And while our team had built a mobile site to accommodate smaller screens, we hadn’t really done anything to improve the website for the growing desktop and laptop screens:

In other words, Wikipedia — a major, legacy website (top 10 ranked, for 10+ years) — had an interface that hadn’t been changed for 15 years. And then one day the Chief Product Officer came to our team (1 product manager, ~6 engineers, 1 quality assurance person, ½ a scrum master, ½ a data analyst, ½ a community liaison, and myself), and tasked us with making significant improvements. It might honestly be a once-in-the-history-of-the-internet kind of situation. Exciting, but rather difficult.

There wasn't one specific reason why, in 2019, this all came about (though there is arguably one person who got the ball rolling — whatup Nirzar). Instead there were a variety of reasons: the line length had no limit, the search box was small and in an unfamiliar place, the table of contents was only available at the top of the page, language switching was hidden, the hierarchy and organization of elements was confusing to some, the front-end was brittle, it was becoming difficult to find places for new features, etc. But perhaps above all was the fact that we had been neglecting the interface (due to other priorities, and limited resources), and we felt confident that if we gave it some overdue attention we could make it better for both reader and editors.

Meanwhile, between 2004 and 2019, the Wikipedia volunteer community had been busy making localized improvements to specific versions of Wikipedia. This is possible because Wikipedia is a democratically run, open-source website, with an interface (and back-end) that can be changed by anyone. And change it they do. But while these changes were real improvements, the majority of them were geared towards logged-in editors. Also the improvements had not been centralized, so they weren’t available on most Wikipedias. A slight tangent: unbeknownst to many people, the many versions of Wikipedia are not centralized. The Wikipedia you read (whether it’s English, Bangla, Telugu, Kyrgyz, Korean, Persian, or any of the 300 others), is actually a separate website from all of the other Wikipedias that exist. Sure they share a lot of code, use the same servers, and generally have the same interface. But changes volunteers make to the interface (and the content too, of course) are made locally. This meant that the improvement Korean Wikipedia made to have a collapsible sidebar didn’t carry over to English Wikipedia (or anywhere else), the changes Hebrew Wikipedia made to have a sticky header didn’t carry over to Burmese Wikipedia, and on and on.

Sometimes the lack of centralization makes sense: for example, the language variant switcher on Serbian and Chinese, aren’t needed on other Wikipedias. But overall, the lack of centralization meant that a bunch of good interface changes between 2004 and 2019 were localized to smaller Wikipedias, and the majority of Wikipedia readers were not benefitting from them. Like anything else, there are pros and cons to decentralization.

Before we go any further let’s pause and think about the situation our team was in. The general layout for readers hadn’t changed much since 2004, but the interface (or interfaces) had changed in hundreds of little, decentralized ways, spread across 300 different language versions of Wikipedia, that were each democratically controlled by their own community of volunteers. And we, a team of ~12 people, wanted to make a bunch of centralized changes to it. Also we didn’t (and still don’t) have comprehensive and detailed analytics for the interface, so we didn’t have much data about how people were using it. Neither did/do we have the ability to easily run A/B tests with logged-out people. The sheer task of learning about all of the different local modifications of Wikipedias across various languages was daunting. Let alone deeply engaging with thousands of volunteers spread across these communities to learn more about their workflows, and why they had made certain modifications. While ultimately being beholden to the democratic system, within which we had to represent the needs of the billions of readers who are inconveniently absent from the whole conversation. But we had unlimited credits on usertesting.com, so that was something.