Compare Google, Bing, Marginalia, Kagi, Mwmbl, and ChatGPT

Original Article ↗ Hacker News Discussion ↗

Search results for everyday queries like “YouTube downloader” or “ad blocker” are increasingly dominated by SEO spam, ads, and even malware, prompting comparisons between Google, Bing, smaller engines (Kagi, Marginalia, Mwmbl) and LLM-based tools like ChatGPT. Commenters debate how much of the decline is due to ad-driven business models versus the sheer volume of low-quality content, and whether user-side fixes like whitelists, uBlacklist, or paid engines meaningfully help non-expert users. There is broad agreement that search quality for technical and niche queries can still be good with the right tricks, but that “just type what you mean and get a safe, trustworthy answer” feels further away than it did a decade ago.

Overall view on search quality

  • Many feel Google and Bing have degraded: more ads, SEO spam, local junk, and even malware; straightforward queries now feel like a “minefield”.
  • Others report results are still fine, especially with ad‑blockers and for technical/programming queries; some say current results are better than a decade ago.
  • Disagreement whether the problem is:
    • The web becoming flooded with low‑effort content, or
    • Search engines optimizing for ads and engagement instead of quality (removing advanced features, tolerating spammy ad‑laden sites).

Subjectivity of “good” results

  • Strong debate over the article’s evaluation rubric, especially “youtube downloader”:
    • Some agree a CLI tool like yt‑dl/yt‑dlp is the “correct” safe answer.
    • Many argue normal users want a simple web page or app, not a command‑line tool, so ranking online downloaders first is reasonable.
  • Similar disagreement for “ad blocker”: some insist uBlock Origin must rank first; others consider AdBlock/ABP “good enough”.
  • Several argue judging engines by a handful of hand‑picked, nerd‑biased queries is not representative.

Kagi, Marginalia, and other alternatives

  • Kagi:
    • Fans praise ad‑free UI, per‑domain boosting/blacklisting, bangs, “Small Web” and FastGPT modes; some say it “feels like pre‑enshittification Google”.
    • Others find results not markedly better than Google/Bing, or weak for images/videos. Some cannot reproduce the article’s poor Kagi outputs (suggesting region, timing or customization differences).
  • Marginalia:
    • Lauded as an impressive one‑person engine focused on non‑commercial, “small web” sites and strong filtering.
    • Criticized as unsuitable for common navigational queries (e.g., Wikipedia/IMDB), which is partly by design.
  • Other tools mentioned: Brave Search, Searx, Qwant, Metaphor (embedding‑based), uBlacklist for per‑user blacklists, Google Programmable Search, and directory‑style approaches.

ChatGPT and LLM‑based search

  • Contention around using GPT‑3.5:
    • Some say any evaluation that doesn’t use GPT‑4 misrepresents “ChatGPT”.
    • Others counter that 3.5 is the free, widely used version, so testing it is valid if clearly labeled.
  • Reports that GPT‑4 can often give very good answers and relevant links but:
    • Hallucinates, varies from run to run, and may refuse or hedge on “gray” tasks (e.g., YouTube downloading).
  • Perplexity, Bing/Copilot, and Kagi’s FastGPT are cited as promising hybrids (RAG‑style summaries over web results).

Safety, spam, and censorship concerns

  • Worry about search engines funneling users to malware, deceptive downloaders, and content‑farm “blogspam”.
  • Some suspect under‑ranking of certain political/war‑related content; others attribute odd gaps to general ranking problems rather than deliberate censorship.
  • Broad agreement that incentives (ad money, SEO gaming) structurally bias search away from user‑centric quality.