Consider this: are there more good web pages today than there were in 1999?
I don’t have any evidence (yet) but I’m pretty confident that the answer is yes. While the ratio of good to bad may have changed, there’s still an enormous amount of good stuff on the web. What has changed is how we find it, and I have a theory that the feeling that the problem is the content isn’t something that happened by accident.
There’s always been junk web pages (I’ve created my share of them) but so long as you could sift the wheat from the chaff the bad ones had little to no impact on you; they might as well not exist. In the early days we made pages of links to the good pages and linked to each others pages of good links and you could spend an afternoon “surfing” from link list to link list and discover some amazing websites. This was great for surfing, but if you were looking for something specific you needed a way to search for that thing and so we used existing text searching capabilities of the various operating systems these sites ran on to provide a way to search-through these lists of websites. At some point someone realized that maintaining these lists was tedious and that the linked-list nature of the web was a natural fit for a program that went from site-to-side indexing each site that was linked to create the index automatically.
This worked pretty well for awhile, but as the web grew it was hard to keep that index up-to-date and the results of searching it using basic text search tools other approaches were experimented with. Some were more popular than others but they all provided a way to search the pages that were indexed, they only indexed the content, they didn’t alter or attempt to create it.
It wasn’t long after search engines became automatic that the automatic generation of websites began. This was long before “AI” was a generic term for predatory Large Language Models and the earliest generated sites used the same sort of existing text manipulation tools that were used to build the first search engines. This evolved to use techniques that are “classic AI” like machine learning and such but the quality of the results was low enough that it was pretty easy to spot (like the first-generation Terminators). These techniques also used far fewer resources so while these sites were a nuisance, they were still fairly benign. The next generation of LLM-powered slop websites were a different matter and over time they became harder and harder to detect while at the same time producing worse anti-information (or in other words, Bullshit). Searching for a web page with factual information about a subject became incredibly difficult and it was very easy to grab something from the first or second hit in the search results that was absolutely incorrect.
Around the same time the major search engines started to provide “summaries” to help with this. Instead of trying to find an accurate website, the search engine would analyze the content in the index and synthesize a response for you that was correct. The only problem was, they use the same LLM technique to synthesize this anti-information and as a result you simple get the wrong answer faster, but with one key difference:
You never visit another website.
Never visiting another website is good business for the one website you visit, and while it may seem that the search engine is being helpful by summarizing the web for you (to protect you from the garbage web!), it’s also helping itself. The reason the garbage web exists at all is because for most people, the first page of search results is the web. By promoting garbage websites through prioritizing them in search results is the primary evidence most people have that the web sucks!. Rewarding the enshittification of the web through preferred position in search results laid the foundation for making LLM-generated summaries acceptable just like crushing the complexity of human communication into a single binary value (Like!) laid the foundation for believing that LLM’s are intelligent. I don’t know if this long-play strategy is something Google did on purpose or if they were just lucky but the result is the same: most people thing the web sucks.
This is what I’ve been thinking about this morning and it’s yet another reason I believe that Personal Search Engines (PSE’s) are worth a try. A stand-alone PSE shouldn’t be capable of returning a bad website in the search results (why would you add one in the first place?) and a federated PSE would only return a bad site if the people you follow have bad taste (or are pranking you). In either case, you can immediately and permanently evict any trash you come-across yourself and be done with it forever. Maybe PSE’s will never index the entire web, but I’d call that an asset.
I apologize if this feels like an ad for PSE’s but I keep finding ways that they could help restore the experience of the web many of us miss from the early days. That web is still out there, it’s just a lot harder to find and the people who built it, at least the ones I know really want to build it again but at the moment it’s virtually invisible and encouragement to feel like it’s worth the work is hard to come by. PSE’s might be a way to change that, so if nothing else writing posts like this about them help encourage me to push that work forward because I love the web for what it was and a grieve daily for what it appears to have become.
Jason J. Gullickson, 2026