WHAT THE INTERNET WAS
Not nostalgia. Not nostalgia. Explained plainly, with sources named and dated. Four structural changes reshaped it — and three of them happened before AI arrived.
What it actually was
What it got right
- Pages had people behind them. Not always good people, but findable ones — a name, an email, an "about" page written by someone who wanted you to know who they were.
- Effort was a filter. Publishing cost time. Not much, but enough that most of what existed had somebody's attention in it.
- You went to places. Discovery meant finding a site and returning to it. You chose your sources, and you knew you had chosen them.
- Weirdness survived. Nothing optimised the strange, obsessive, single-subject site out of existence, because nothing was optimising.
What it got badly wrong
- It was small and exclusive — mostly people with money, English and a computer. The web that gets mourned was not most of the world.
- It was hard to use, and being hard to use is not a virtue.
- It was frequently vicious, with no moderation and no recourse, and the people harmed by that were rarely the ones now writing nostalgically about it.
- Most of it was bad. Survivorship bias does enormous work in these conversations — the pages you remember are the ones worth remembering.
The four changes
Publishing stopped requiring effort
Platforms removed the friction of having a website. Enormously democratising — and it also removed the filter that effort had quietly been providing. Both of those are true at once, and most arguments about this era are people insisting on only one.
Attention became the product
When advertising pays for everything, the thing being optimised is time spent, not value delivered. Nobody decided this maliciously; it is simply what a system does when you attach revenue to one number.
Everything that gets blamed on algorithms follows from this, and it happened long before generative AI existed.The same mechanism, followed forward: AI and advertising
Discovery moved from places to feeds
You stopped going to sites and started being shown things. That is genuinely convenient — and it means the question "who is telling me this?" became optional rather than automatic.
Answer engines are the latest step in the same direction: now you are not even shown the places, only the summary of them.See how to search now
Producing plausible content became nearly free
This is the AI part, and it is a change in supply, not in kind. The incentives to flood the internet already existed; what was missing was cheap production. That constraint is gone.
The measured results: automated traffic passed half of all web activity for a second consecutive year, and roughly half of newly published articles are primarily machine-generated — a share that plateaued rather than continuing to climb.Sourced and dated in the quarterly record
Three strata
The same story told as layers rather than causes — rescued from an earlier exhibit on this site, because it says something the list above does not.
The handmade web
Pages written one at a time, by people with names. Webrings, guestbooks, forums where a stranger would spend an hour answering your question for nothing. Slow, ugly, and unmistakably alive — the layer every later one stands on, and the one people are homesick for.
The platform density
Billions arrive through five doors instead of a million. The feed replaces the bookmark; the algorithm replaces the librarian. Automation exists — crawlers, spam farms — but the signal is still overwhelmingly human. This is the last layer anyone will describe that way.
The flood
Generative systems make publishing free and the economics invert. Automated traffic reaches more than half of everything moving on the web; machine-written sites are counted in the thousands; voices and faces are rented by the minute.
The tower keeps rising. The builders changed.Figures sourced and dated in the quarterly record.
Read those in sequence and the popular story falls apart. Three of the four changes were complete before AI arrived. AI did not cause the attention economy, or feed-based discovery, or the collapse of the effort filter — it removed the last remaining cost in a system already pointed in that direction. Which is why blaming AI for all of it is both inaccurate and comforting: it implies that removing AI would return us to something, and it would not.
What is actually recoverable
Very little of the old web can be brought back, and most of it should not be. But the habits it enforced by accident can be adopted deliberately, and they work better now than they did then:
- Go to places again. Keep a small list of sources you visit directly. It is the only defence against ranked results that does not depend on the ranking improving.
- Ask who is telling you. The step that feeds removed was never redundant — it was where you noticed a source was a vendor, or a decade old.
- Value effort as a signal, not because effort guarantees quality, but because it is now the scarce thing. A page that clearly cost someone something is rarer than it was.
- Make something that is not optimised for anything. The strange, single-subject, unmonetised page is still the most human artifact on the internet, and it still costs almost nothing to make.
The internet did not die and it was never golden. What happened is duller and more useful to understand: a system built to reward volume finally got a machine that could supply infinite volume. The people who miss the old web are usually not missing the technology — they are missing the sense that a person was on the other end. That is still available. It is just no longer the default, and it now has to be chosen on purpose.