Stop Bloating Motherfucking Websites

Four one-page websites with a swear word in the domain have been making the case for plain HTML since 2013. Everyone links them. Nobody measures them. So I did, and then I measured the sites people point to as proof the argument works in the real world.

Before there was a name for any of this, there was motherfuckingwebsite.com.

One page. Black text on white. No framework, no build step, no cookie banner, no newsletter interstitial. The argument, delivered at a volume the domain prepares you for, is that a plain HTML document is already responsive, already accessible, already fast, and already works on every device anyone owns, and that most of what gets layered on top is the designer creating problems and then charging to solve them.

It's a joke that turned out to be right about almost everything. The interesting part is that people kept answering it, and the replies form an argument in order.

Four manifestos, each moving the line a few declarations

Read in order, they argue with each other, and the argument goes somewhere.

motherfuckingwebsite.com is the thesis. Use HTML. Your document is already fine.

bettermotherfuckingwebsite.com was the first rebuttal, and the most useful one. It's offline as I write this, so no link. It concedes that the original is unpleasant to read, and fixes it with a handful of CSS declarations: line height, a measure you can track across, some margin, a grey that isn't #000. "No framework" and "no typography" are different claims, and only one of them was ever defensible.

perfectmotherfuckingwebsite.com raises the bid to ten declarations, a media query, and one attribute. The attribute is lang on <html>, and hold that thought. It also picks up prefers-color-scheme, so it won't bleach your eyes at 1am, and it gets fussy about proper quotes.

thebestmotherfucking.website is the maximalist entry. It keeps a running weight budget in the copy and exists to make the point that you can have a real design, a webfont, an image, and still be an order of magnitude lighter than the median site.

There's also justfuckingusehtml.com, the same argument aimed at people reaching for a single-page app to render a brochure.

I ran every one of them through Lighthouse

Mobile emulation, simulated throttling, same engine as PageSpeed Insights. Here's the table.

Caption: Lighthouse 12, mobile, simulated throttling, run from one laptop on the day of writing; scores move a few points between runs.¹ The median row is the HTTP Archive's Web Almanac.²

Some of this is what you'd expect. The perfect one is perfect: three kilobytes, two requests, no script, two clean 100s. This site, which took the canon as its editorial policy, lands in the same row with eleven more kilobytes of CSS. And every page in the canon paints in about a second on a throttled phone, against a median page that is six hundred times heavier than the best of them.

Then there are the parts that are funnier than the pages.

The original is 98% Google Analytics by weight

Two kilobytes of HTML, 179 kilobytes of tracking. The manifesto against bloat ships more script than most of its readers.

motherfuckingwebsite.com is 181 KB on the wire. The document is 2.2 KB of it. The other 179 KB is analytics.js and Google Tag Manager's gtag/js, which between them are 81 times the size of the page they're measuring.³ It still scores 100, because a throttled phone can chew through that much JavaScript in under a second, and because Lighthouse grades the experience, not the irony.

It also loses twelve accessibility points for one reason: <html> has no lang attribute. Which is the exact attribute the perfect motherfucking website added, four years later, with a paragraph explaining why. The canon corrected itself and the original never took the note.

craigslist is not fast any more

The site everyone cites as proof that plain HTML wins scores 45 on mobile and ships 628 KB of JavaScript.

I expected craigslist to top this table. It's the canonical example of a business that never redesigned and never needed to. It scored 45. A city page is 737 KB, 628 of them JavaScript, and the largest paint lands at six seconds on a throttled phone. The <html> element has no lang, and there are links a screen reader announces as nothing.⁴

None of that is visible from a desk. The famous blue links are still there. What changed is everything underneath them, one dependency at a time, until the plain-HTML site was a 628 KB app wearing the plain-HTML site's clothes. That's the drift this whole canon is about, happening to its favourite example.

Legible News shows the argument works for a real product

A daily news site, 83 KB, one-second paint, and a page that ranks the competition.

Legible News is a real product with real readers, and it scores 100 at 83 KB, which is to say it's lighter than the original motherfucking website. Its 58 KB of script is for a reader, not a tracker.

It also publishes the comparison I'd have had to make myself. legiblenews.com/speed runs the major news outlets through PageSpeed Insights and ranks them: Legible News at 100, USA Today at 95, the Financial Times at 91, and then a long slide down through NPR at 39, the New York Times at 30, and CNN at 17.⁵ Same engine as my table, different neighbourhood. The median news site is a fine demonstration of what a decade of adding things to a document gets you.

The best case against the canon

Steelman first. The obvious objection is that these are toys.

"They're single pages with no product behind them. Of course they're 3 KB." True of the canon. Not true of Legible News, which publishes every day and sits in the same row. And the median page isn't 2.3 MB because it has a product; it's 2.3 MB because nobody looked.²

"Users don't notice a second." Google's own guidance draws the good line for largest paint at 2.5 seconds.⁶ The canon is at one. craigslist is at six. The gap isn't a second.

"Accessibility isn't free either. Half of these lose points." They do, and the losses are instructive: a missing lang, a low-contrast grey, a tabindex above zero. Each is one line. The 82 at the bottom of the table is missing lang and has unnamed links, on a page carrying 628 KB of script that could have added either.

"craigslist is fine. People use it every day." They do, and they'd use it on a worse connection with a six-second first paint too, because there's no alternative. That's not evidence the weight is free. It's evidence the users have nowhere else to go.

Nobody in the sequence is against CSS

What they're against is the default.

Read in order, they're not four people saying the same thing louder. They're a negotiation about where the line is, and each one moves it by a few declarations.

Nobody in that sequence is against CSS. Nobody's against design. What they're against is the default: the reflex where a page that could be a document becomes a build pipeline, a hydration step, and half a megabyte of JavaScript before anyone has written a sentence. The table above is that reflex, measured.

The through-line is just: start from the document. Add what the page needs, and be able to say what each addition bought you. Then, every so often, measure it, because the original stopped doing that and picked up 179 KB of analytics without noticing.

That's the whole editorial policy of this site, and those four pages got there first.