HN Debrief

I Just Want to Search

  • Search
  • Business
  • Consumer Internet
  • AI
  • Developer Tools

The post is a frustrated tour of how everyday search broke. The author contrasts older, more literal search behavior with today’s systems that quietly reinterpret queries, ignore quotes and filters, and bury direct results under recommendations, ads, and AI-style guessing. The examples span web search, shopping, video, and internal product search, but the core complaint is simple: when someone asks for a precise thing, many platforms now treat that as a suggestion rather than a command.

If search is part of your product, exact-match behavior and honest zero-results handling are now a competitive feature, not a boring detail. If you rely on major platforms for discovery, assume their search experience is optimized for retention and monetization before relevance.

Discussion mood

Strongly negative and exasperated. People feel mainstream search products now override explicit intent in favor of engagement, ads, and recommendation surfaces, though a smaller group thinks the bigger issue is organizational neglect rather than deliberate sabotage.

Key insights

  1. 01

    Google snippets expose query rewriting

    Watching the bolded terms in Google’s result snippets shows that the engine often replaced or broadened the original query before ranking anything. That changes the complaint from vague “results feel worse” to something concrete and inspectable. The frustration is not just bad ranking. It is that the system silently stops performing the search the user asked for.

    When users complain that search ignores them, instrument query rewriting and surface it in logs or UI. Hidden reformulation is a product decision with trust costs, not just a ranking tweak.

      Attribution:
    • pooploop64 #1
    • kzalesak #1
  2. 02

    Classic search problems are mostly solved

    Accent handling, misspellings, morphology, transliteration, and ranking are real requirements, but they are not new and they are not unsolved. People were shipping workable indexing systems for these cases decades ago. That undercuts the easy excuse that modern search had to get fuzzy and unreliable because language is hard.

    Do not accept “search is inherently messy” as a blanket reason for bad product behavior. Separate hard retrieval problems from deliberate choices like ignoring quotes, filters, and zero-result states.

      Attribution:
    • zetanor #1
    • svachalek #1
    • pooploop64 #1
  3. 03

    Human-curated indexes would trade scale for sanity

    A curated directory model came up as a serious alternative, not nostalgia for Yahoo. The proposal was crawler-assisted discovery, lightweight automated triage, and human approval before inclusion. The key claim was that pre-inclusion moderation changes incentives. If low-quality pages, AI spam, and junk uploads never get into the searchable corpus, producers lose the payoff from flooding the system.

    If you operate in a bounded domain, consider editorially constrained search instead of trying to index everything. Smaller, trusted corpora can beat giant noisy ones when users care more about precision than breadth.

      Attribution:
    • gtowey #1 #2 #3
    • cosmic_cheese #1
  4. 04

    Third-party search used to bypass hostile sites

    Search originally won by crawling through sites that were hard to navigate and linking directly to the buried pages companies preferred users not reach. The same adversarial opportunity now exists inside login-walled services like YouTube transcripts, Facebook Marketplace, Slack, Discord, and Telegram. The argument is that useful search may need to move outside the platforms again, with service-specific connectors and archived public content.

    There may be room for products that index public content across walled gardens instead of waiting for each platform to fix its own search. If your business depends on discoverability inside closed ecosystems, treat external indexing as a strategic opening.

      Attribution:
    • derefr #1
  5. 05

    Search often decays because nobody owns it

    The non-conspiratorial explanation was organizational, and it rings true. Search usually loses priority unless the people setting roadmaps depend on it themselves or see metrics that prove its value. In big products, that means exact-match and power-user features quietly break as other teams ship changes around them. You do not need an evil plan to get a bad search box. You just need no accountable owner.

    Assign a real owner to search and measure success with task completion, not just engagement. Otherwise your product will drift from utility toward noisy discovery without anyone explicitly choosing it.

      Attribution:
    • bogdanoff_2 #1

Against the grain

  1. 01

    Kagi is not a universal fix

    Paid search alternatives are not automatically better just because they preserve old operator behavior. One commenter said Kagi no longer felt more useful than asking Gemini or Claude directly, which challenges the idea that a better ranking stack alone restores the old experience. For some tasks, people now judge search against answer engines, not other search engines.

    If you are building search, your benchmark is no longer just Google. Users will compare you to chat-based workflows that collapse searching and synthesis into one step.

      Attribution:
    • BrokenCogs #1
  2. 02

    Bad marketplace search can hurt advertisers too

    The cynical “more scrolling equals more money” frame misses a limit in commerce contexts. If Facebook Marketplace shows items that are too far away, the wrong size, or over budget, the user may just leave because the item is not purchasable. That means extra impressions do not necessarily translate into healthy ad economics or conversions.

    For transactional search, measure completed purchases and lead quality alongside time on site. Engagement tactics that work in feeds can backfire when the user has a narrow buying constraint.

      Attribution:
    • graypegg #1

In plain english

Claude ↩
A family of large language models and AI assistants made by Anthropic.
collation ↩
Rules a search system uses to compare and sort text, including differences in accents, case, and locale.
Gemini ↩
Google’s family of AI models, used here as a second model to rewrite Claude’s output.
Kagi ↩
A paid web search engine positioned as a cleaner alternative to ad-driven search.
stemming ↩
A search technique that reduces words to a shared root so related forms can match.
transliteration ↩
Converting words from one writing system into another, such as Japanese characters into Latin letters.

Reference links

Article and legal reference

  • I Just Want to Search
    The submitted blog post arguing that modern search increasingly overrides literal user intent.
  • HiQ Labs v. LinkedIn
    Cited to support the argument that indexing public content from platforms may violate platform rules without necessarily being illegal.