Website search hurts my feelings
Website search on e‑commerce and content sites is widely seen as frustrating and unreliable, from grocery chains and big-box retailers to Netflix and Amazon, often failing on even simple queries or faceted filters. Commenters argue this isn’t just technical incompetence: good search is genuinely hard (requiring clean metadata, tuning, and cross-team ownership) and often conflicts with business incentives that prioritize ad revenue, “searchandizing,” and short-term conversion over user satisfaction. Some see hope in better search tooling and LLM-based tagging or reranking, but note that without organizational will and metrics that value user experience, search quality is likely to remain a low priority.
Overall frustration with site search
- Many users find search on e‑commerce, media, and even browser history/settings unreliable or bizarrely behaved.
- Complaints include irrelevant results, missing obvious matches (like part numbers or titles), and facets that hide rather than reveal what’s wanted.
- Several posts note that even a simple full‑text search in a relational DB would outperform some current implementations.
Business incentives vs user needs
- Strong view that “bad” search is often deliberate: surfacing ads, promoted items, or high‑margin products can increase revenue even if it worsens UX.
- Analogies to physical retail (end caps, moving items around, milk in the back) and to “enshittification” of major platforms.
- Some argue most customers still buy despite poor UX, so companies prioritize acquisition and monetization over search quality.
Technical and data challenges
- Good relevance requires: clean metadata, tagging, synonyms, stemming, fuzzy matching, negative boosts, and domain understanding.
- Catalog data is often messy, inconsistently tagged, and owned by understaffed teams with no clear accountability.
- Faceted search is hard: facets must match how customers think, and combining filters (AND/OR, counts, constraints) is non‑trivial.
- Popular stack choice (e.g., “just slap Elasticsearch in”) is criticized as giving a false sense of completeness; tuning is ongoing work, not a sprint.
Examples: good, bad, and weird
- E‑commerce (Amazon, Walmart, grocery chains, office supply stores) widely criticized; results feel over‑fuzzy, ad‑driven, or inconsistent with filters.
- Specialized sites like Digikey, McMaster-Carr, RockAuto are praised for parametric, data‑driven search.
- Netflix search is controversially assessed: some like “similar titles” when a show isn’t available; others hate that it hides “we don’t have this” and lacks filters.
- OS/browser search (Windows Start, Firefox Android, iOS settings) singled out for non‑deterministic, counterintuitive autocomplete and ranking.
Org and product issues
- Debate: is search a “product” problem or cross‑departmental quagmire?
- One side: heads of product must own overall experience and push fixes across teams.
- Other side: they lack authority/resources; issues in data and integrations are politically hard to change.
- Some claim strong UX doesn’t sell; a failed startup example describes good search that lost out to competitors spending on marketing instead.
LLMs and future approaches
- Multiple commenters propose using LLMs for auto‑tagging and result re‑ranking; reported accuracy is decent but imperfect and still needs human oversight.
- Others counter that simply implementing conventional search well is cheaper and more transparent, though hard at large scale.