What We Lost When Search Stopped Making Us Think
I want to talk about something that's been on my mind for a while: search has quietly gotten worse over the past several years, and I think it's worth being honest about why, and what it's actually costing us.
Anyone who's pasted a specific error message into a search box recently knows the experience. The first several results are usually SEO content, the same underlying answer reworded a dozen different ways, padded with filler paragraphs before it even addresses the actual problem, because ranking algorithms have historically rewarded length and keyword density over direct usefulness. Mixed in increasingly are AI-generated pages that read confidently but occasionally get the actual technical details wrong, with no obvious signal to the reader that anything's off. Somewhere further down, if you're persistent, is often the original, genuinely useful answer, sometimes from a forum post years old, occasionally scraped and republished elsewhere with the context stripped out.
This isn't really anyone's fault in a simple sense. It's what happens when the economics of the web reward getting in front of an algorithm rather than being correct. That gap between "ranks well" and "is actually right" has always existed to some degree, but AI-generated content has widened it considerably, since it's now possible to produce large volumes of plausible-sounding text far faster than anyone can fact-check it, and ranking systems haven't fully caught up to distinguishing genuinely useful content from confident-sounding filler.
The response from several major search products has been to layer AI-generated summaries directly into results, effectively synthesizing an answer from whatever's been indexed, including the very content I just described. I understand the appeal from a product standpoint, it reduces the number of clicks needed to get an answer, which reads well in almost any metric. But it also means users are increasingly being handed a confident paraphrase instead of a source, with no easy way to verify whether that paraphrase is accurate unless they already know enough about the topic to catch an error. That's a strange thing to optimize for, since the people who most need a reliable answer are often the ones least equipped to spot when they've been given a wrong one.
What actually concerns me most isn't the quality of individual search results, it's what this shift is doing to the underlying skill of research itself. Holding a real question in your head, forming a hypothesis, checking it against multiple sources, and noticing when two sources disagree is a skill that gets sharper with practice and duller with disuse, the same as anything else. Every time someone accepts a generated summary at face value instead of following through to an actual source, that's a small rep they didn't do. It's not dramatic in the moment. It adds up slowly, the same way any skill quietly erodes when you stop exercising it, and you tend not to notice until you're actually asked to do the work yourself and find it harder than it used to be.
This pattern isn't limited to search either. It shows up anywhere people reach for a generative tool to skip the effortful part of producing something, an essay, a difficult message, a first draft of anything. The output is often fine, sometimes genuinely good, which is part of what makes this hard to talk about honestly. But the actual struggle of starting from a blank page is a large part of how people get better at generating ideas in the first place, and skipping that struggle repeatedly doesn't make someone faster at it over time, it tends to make the underlying skill weaker from lack of use.
I want to be clear that I'm not against these tools generally. I use them myself, including for mundane things like tightening up a resume, and there are plenty of contexts where they add real value. The distinction I care about is between a tool that extends what someone is capable of doing and a tool that quietly does the thinking for them while preserving the feeling that they're still in control. Search, at its best, used to require a small amount of genuine cognitive effort: comparing sources, weighing credibility, forming your own synthesis. It was never perfect, but that friction served a purpose. A lot of recent product decisions across the industry, not maliciously, but as a natural consequence of optimizing for engagement, have been quietly removing that friction, and I think it's worth pausing to ask what we're trading away in the process.
I don't have a tidy conclusion here, and I'm skeptical of anyone who claims to. This isn't a problem with an obvious fix, and I don't think one side project changes much about how the incentives above actually work. Mostly I just think it's worth naming what's happening plainly, because the trade is easy to miss when it happens one convenience at a time, and a lot easier to notice once you say it out loud.