Study 001
The Web Was Never Supposed to Look Like This
Nobody ever decided what the web should look like. There was no committee, no house style, no founding document on layout. The conventions we treat as obvious were argued out in public, usually by people solving a different problem, and the arguments are still readable.
Before websites had a look
On the evening of 25 February 1993, Marc Andreessen sent a message to a mailing list called www-talk. The subject was “proposed new tag: IMG”. He wanted a way to put a picture inside a page: a standalone tag with a source address, which the browser would fetch and draw where the tag sat. He noted that it could go inside a link, so an image could be clickable, and he mentioned that the browser he was working on already did this.1
That last part matters more than the proposal. Tim Berners-Lee preferred a different approach, something more general that could embed any kind of object. On 12 March, Andreessen declined it and said they would most likely go with the form he had proposed.2 Mosaic shipped it. Everyone copied Mosaic.
Before that argument, pages had no look to speak of. They had headings, paragraphs and links, and the browser decided what those meant visually. The page described its structure; the reader’s software decided the typography. The idea that a website has an appearance that its author controls was not the starting point. It was a consequence.
The most consequential visual decision in the history of the web was settled by whoever had already built it. Not by a designer, and not on design grounds. That pattern repeats for the next thirty years.
When everyone had their own corner
GeoCities started in 1994 as Beverly Hills Internet and gave people free space to publish. Pages were sorted into themed neighbourhoods.3 By the time Yahoo agreed to buy it, in an all-stock deal dated 27 January 1999, the merger filing called it the world’s largest web-based community.4
What those pages looked like is a matter of record rather than nostalgia, because the record almost did not survive. In April 2009 Yahoo stopped new sign-ups, and the service closed on 26 October that year. Yahoo was explicit about what was happening: the service was not being decommissioned, it was being deleted.3 Jason Scott assembled a group of about twenty-five people who spent six months downloading whatever they could reach. Yahoo published no list of accounts, so they had to guess addresses. They saved roughly a hundred thousand of them.5
This era gets remembered as kitsch, and plenty of it was. The part worth keeping is structural: the person writing the page was the person deciding how it looked, with no layer in between. That is rarer now than the animated flames.
Style was meant to be negotiated
On 10 October 1994, Håkon Wium Lie published a proposal called “Cascading HTML style sheets”. It carried a version number, 0.92, and a note saying it was work in progress and not yet a basis for implementation. Its central idea was layering: the reader’s software sets its preferences first, and the document’s style sheet then has its say. The proposal explicitly covered non-visual media as well as screens.6
It is worth noticing what that document assumes. Style was meant to be negotiated between author and reader, not imposed. The version of CSS we actually use kept the cascade and quietly dropped the negotiation.
Everything starts looking familiar
Here is where the story usually turns into a complaint, so it is worth having a measurement instead of a feeling. In 2021 Sam Goree, Bardia Doosti, David Crandall and Norman Makoto Su published a study that did exactly this. They rebuilt 227,802 snapshots of 10,482 sites from the Internet Archive, covering 2003 to 2019, and compared them with computer vision. They also interviewed eleven web designers with at least fifteen years in the trade.7
Two findings matter. Sites have grown more alike since 2007, and the average difference between page layouts fell by 43 per cent between 2010 and 2019. So yes, measurably, sites converged.7
The second finding is the one that breaks the easy story. Over the same period the code of those sites became less similar, not more. What best predicted whether two layouts looked alike was the use of a small number of shared libraries, and Bootstrap is the one most strongly associated with layouts that resemble each other. The authors are careful about the limits: a representative sample of the web is hard to collect and hard even to define, and theirs leans towards corporate sites in the English-speaking web.7
The dates fit the tools. Ethan Marcotte’s article “Responsive Web Design” appeared in A List Apart on 25 May 2010, naming fluid grids, flexible images and media queries as its three ingredients.8 Bootstrap came out of a small group inside Twitter and was tagged 1.0.0 on 18 August 2011.9 It was not responsive. The responsive grid was the point of the next major version, tagged 2.0.0 on 1 February 2012.10
“Templates ruined the web” is the wrong sentence, and the data says so. Nobody was forced into a grid. What happened is that a handful of component libraries became the fastest way to get a page onto a phone, and speed won, one reasonable decision at a time. The convergence is real. The villain is not.
Three rooms
Scroll through the exhibit below. These are reconstructions built from scratch in the same stylesheet as this page, not historical screenshots. They are meant to show how the defaults of each period felt, not to reproduce any particular site.