The article's premise, which it doesn't bother defending, is that Google has stopped returning the page you wanted. It surveys six alternatives spanning paid (Kagi), independent crawlers (Marginalia, Mojeek), LLM answer engines (Perplexity, You.com), and Bing skins (DuckDuckGo, Brave, Startpage) as viable exit ramps.
Submitted the TechCrunch piece to HN where it hit 333 points, signaling broad practitioner agreement. The community resonance around the framing — that Google 'isn't really Google anymore' — reads less like debate than acknowledgment of a shared reality.
Practitioners in the 265-comment thread cite exact, testable degradations: pasted error strings no longer surface the Stack Overflow page containing that literal substring, site: operators silently fall back to whole-web results past page two, AI Overviews eat the viewport with wrong citations, and personalization makes two engineers' top-10s materially diverge on identical queries. These are diagnostic complaints, not nostalgia.
Argues the interesting point isn't that Google got worse but that none of the six alternatives are competing on Google's terms. Kagi sells ad-absence and per-domain ranking controls; Marginalia indexes the non-commercial web; Perplexity replaces the result list with a synthesized answer — each is a deliberately narrower product, not a clone.
TechCrunch published a roundup on May 21 — "Six search engines worth trying now that Google isn't really Google anymore" — and it punched through to 333 points on Hacker News, with the comment thread reading less like a debate and more like a wake. The premise the piece doesn't bother defending: Google search, as the thing that returned the page you wanted, is functionally over. The replacements it surveys span four distinct strategies: pay-to-play (Kagi), independent crawl (Marginalia, Mojeek), LLM-mediated answer engines (Perplexity, You.com), and the now-decade-old Bing skin Google's tail still props up (DuckDuckGo, Brave, Startpage).
The HN thread surfaces the specific failure modes practitioners cite. Pasting an exact error string into Google increasingly returns SEO-stuffed listicles instead of the Stack Overflow thread that contains the literal substring. Site-restricted searches (`site:github.com`) silently fall back to whole-web results past the second page. The "AI Overview" eats the top viewport with a paraphrase of the first result, often with the citation wrong. Personalization injects results from your recent browsing in ways that hurt reproducibility — two engineers searching the same query on the same day get materially different top-10s.
None of this is news to anyone who's been searching for technical content since roughly 2022. What's new is that the alternative tier has matured enough to actually rely on. Kagi reports ~50k paying subscribers as of late 2025 and runs its own crawler (Teclis) alongside Google/Bing/Mojeek/Marginalia anchors. Marginalia indexes ~80M pages of the non-commercial web with a hand-tuned ranker. Perplexity reportedly hit $100M ARR in Q1 2026. The exit ramps from Google are paved.
The interesting structural point is that none of the six engines in the TechCrunch list are trying to be Google. Kagi is selling the absence of ads and the presence of per-domain ranking controls — you can permanently downrank Pinterest and lift personal blogs, and the setting persists. That's not a search engine feature, it's a search engine *configuration interface*, which Google has never offered because their incentive is to optimize the click, not the answer. Marginalia explicitly excludes commercial content. Perplexity is barely a search engine at all — it's an LLM with citations, where the underlying retrieval is a means, not the product.
This fragmentation matters because the dominant-engine era taught a whole generation of developers a single skill: "Google it." That skill is now leaky in the way `cd` is leaky on Windows. The replacement skill is query routing: knowing which engine's index and ranking function suits which kind of question. Exact-string error lookup? Kagi or DuckDuckGo (Bing's index is genuinely better than Google's at literal substring match in 2026). Recent paper or framework changelog? Perplexity, because the LLM disambiguates ambiguous version queries. Hobbyist write-ups on niche hardware? Marginalia, full stop — it's the only engine that hasn't algorithmically buried the personal web.
The HN thread's most-upvoted concrete complaint isn't ranking quality, though — it's that Google's index appears to be shrinking. Multiple commenters cite the same observation: documents that returned a result in 2022 now return nothing, even with quoted exact-phrase queries. Whether this is index pruning, demotion of "low-quality" content under the Helpful Content Update, or a deliberate shift toward freshness-weighted recall isn't clear, but the practitioner-facing symptom is the same. If your debugging workflow assumes "if it was ever written down, Google will find it," that assumption no longer holds — and probably hasn't held for two years.
The pricing question deserves a beat. Kagi is $10/month for the standard tier (300 searches), $25/month for unlimited. The HN response to that pricing in 2023 was indignation; in 2026 it's a shrug. A senior engineer running 50 searches a day is paying roughly $0.017 per search at the unlimited tier — less than a single LLM call against a frontier model, and the search is arguably the more leveraged operation. The thing that actually delayed adoption wasn't the price; it was the friction of changing the default search engine in three browsers across two laptops and a phone, which is real but one-time.
The concrete move is to stop having a default search engine and start having a router. The lowest-effort version: install Kagi as your browser default, set up a `!g` bang for the cases where you specifically want Google's index, and bookmark Perplexity and Marginalia for the query shapes they're better at. The slightly-more-effort version: a local hotkey or Raycast/Alfred snippet that routes by prefix — `gh ` to GitHub code search, `so ` to Stack Overflow's native search (which is also better than Google's site-restricted version), `mdn ` to MDN. Most of what you used Google for, in 2026, has a purpose-built engine that's faster and more accurate.
For anyone building developer tools, the implication runs deeper. If your product's onboarding or docs assume users will arrive via a Google search for a relevant term, that funnel is degrading. Documentation that ranks today on a long-tail technical query may not rank in six months — not because Google penalized you, but because the AI Overview ate the click. The defensive move is to get your content into the indices that aren't collapsing: Mojeek and Marginalia crawl most public sites by default; Kagi's Teclis includes you if you're linked from the Kagi-curated set; Perplexity and the LLM-as-search vendors pull from whatever's in their training cut plus a live retrieval layer that increasingly favors a small set of canonical domains. Cross-posting canonical docs to a Hashnode or Dev.to mirror is no longer just an SEO play, it's an indexability hedge.
The ten-year monoculture in search ended quietly, without a single replacement winning. The future of search for developers looks like the future of databases: pick the engine that matches the query shape, accept that no one engine wins everything, and treat the routing layer as part of your tooling. The TechCrunch roundup is worth the click not because the six engines listed are surprising, but because the framing finally is — "now that Google isn't really Google anymore" is a sentence that wouldn't have parsed in 2020, and reads as obvious in 2026. The interesting question for the next year is whether any of the alternative engines can sustain an independent crawl at scale, or whether the entire long-tail web ends up indexed only by whoever's training the next frontier model.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.