Search engines solved the indexing problem and then slowly broke the relevance problem
Google in 2010 was genuinely impressive. PageRank plus content signals produced results that were, by most measures, better than what a librarian could surface in an hour. The web had a relatively small fraction of pages designed specifically to game the index.
What happened between then and now is mechanically simple: the content-farming industry matured, optimized explicitly for ranking signals rather than reader value, and scaled to billions of pages. Every major algorithm update — Panda, Penguin, Helpful Content — landed a cycle behind the exploit. The 2023-24 Helpful Content Update hit a lot of genuinely good sites while the worst SEO-mill content survived by distributing across thin domains.
The AI content wave didn't create the problem — it made it structurally harder to fix. Models trained on web text generate content that looks maximally similar to high-ranking pages, because that's what they were trained on. You get a feedback loop where ranking signals are gamed by content optimized on ranking signals.
PageRank worked because inbound links were too expensive to fake at scale. That assumption held for maybe 15 years. Every subsequent quality signal has eroded faster.
The honest question: is relevance even solvable once the supply of plausible-looking content is essentially infinite? Or does web search just become a different kind of tool — good for navigational queries, hours, directions, simple lookups — while something else handles the "actually explain this to me from primary sources" use case?
Curious what search workflows people are actually using in 2026 for finding real expert opinion or primary sources, as distinct from AI summaries of AI summaries.
0 replies