Google Search Is Now a Giant Hallucination

Original Article ↗ Hacker News Discussion ↗

Overall sentiment on Google Search

  • Many report that Google search feels much worse than years ago: harder to find even known-existing pages, obscure facts, or exact lyrics.
  • Users complain about aggressive query “interpretation,” loss of true verbatim search, and cluttered, ad-heavy result pages.
  • Some still prefer Google over Bing/Copilot/Perplexity, but mostly as the least-bad option.

AI Overviews and hallucinations

  • Multiple examples of AI Overviews giving wrong or dangerous advice (e.g., gasoline for cooking, bad cleaning-chemical mixes, incorrect watch instructions, false claims about public figures, misleading interpretations of parachute studies).
  • Some viral screenshots (e.g., suicide recommendation) are acknowledged as fake, but others are real.
  • One side: these are fringe, meme-like queries among billions; kinks to be ironed out.
  • Other side: even rare but confident errors are dangerous because people tend to trust Google’s top answer, especially as a single authoritative summary.

Shift from search engine to answer engine

  • Users distinguish between:
    • Old Google: pointing to sources, where you compare sites and context.
    • New Google: asserting synthesized answers “in its own voice,” obscuring provenance and domain reputation.
  • Some argue AI could help with query understanding and keyword extraction rather than answer generation.
  • There’s nostalgia for when snippets and info boxes were simpler and clearly sourced.

Ads, incentives, and “enshittification”

  • Strong criticism of ad-driven incentives: spammy SEO content, misleading service-area info, overloaded pages (e.g., official park site) that become unusable without ad blockers.
  • Claims that Google has deliberately tolerated or induced lower-quality results for higher ad revenue and captured too much of the value chain.

User workarounds and alternatives

  • Heavy use of operators: quotes, -term, site:, before:YYYY, “Verbatim” mode, and the hidden &udm=14 “Web” view to get pure text links.
  • Several recommend Kagi (and to a lesser extent DDG) for cleaner, configurable, paid search; appreciation for domain ranking/blacklisting. Others find Kagi too expensive or hard to pay for outside the US.

Broader concerns

  • Worries about LLMs amplifying internet misinformation, including via Reddit training data.
  • Fear of recursive “AI training on AI sludge.”
  • Questions about legal liability once Google’s AI acts more like a publisher than a neutral indexer.