Choosing a search engine

The question almost nobody asks: who found these results?

Search engines are usually compared on what you can see: the look of the results page, the privacy policy, whether there is an AI summary, whether it plants trees. Those are real differences. None is the important one.

The important one is invisible from the results page: where did these links come from? Every search engine sits in one of three states, and which state it is in determines almost everything else about it.

• Own index. The engine runs its own crawler over the open web, stores its own copy, and ranks it. Google, Microsoft Bing, Yandex, Baidu, Mojeek, Brave Search and Marginalia all do this.
• Hybrid index. The engine crawls and indexes some of the web itself, and blends in results bought from a partner where its own coverage is thin. DuckDuckGo, Qwant, Ecosia and Kagi are all hybrids, in quite different proportions.
• Resold results. The engine has no crawler and no index. It sends your query to somebody else's engine and presents what comes back. Yahoo, Startpage, Swisscows, Microsoft Copilot, Dogpile and Kiddle all work this way.

This is a structural fact, not a quality judgement. Reselling is a legitimate product decision — Startpage does it deliberately and says so. But it means that if you leave Google for an engine that resells Bing, you have changed who is watching you and who ranks your results, while the underlying view of the web is still one of two American companies'. Whether that counts as switching depends on what you were trying to escape.

How few search engines actually have an index

There are hundreds of search sites. There are roughly a dozen web indexes. That gap is the whole story.

Crawling and ranking the open web at scale is done, worldwide, by Google and Microsoft in the United States; Yandex in Russia; Baidu in China, with Sogou as a hybrid alongside it; Naver in South Korea, using its Yeti crawler; Seznam in the Czech Republic; and Cốc Cốc in Vietnam, whose crawl is deliberately scoped to Vietnamese content and .vn domains — a real index, but a national one.

Independently in the West the list is shorter still. Mojeek, a small British company, has crawled and ranked its own index since 2004 and has never resold anyone's results; it reported 9 billion pages in 2025. Brave Search built its index from Tailcat technology acquired in 2021 and published its own progression away from dependence — about 87% of results from its own index at launch in June 2021, 93% in mid-2022, and removal of the last Bing API calls on 27 April 2023. Marginalia is one engineer running a documented crawler over a few hundred million documents on roughly $200 a month.

Two newer entrants built indexes for AI rather than for a results page: Perplexity, which described tracking over 200 billion unique URLs in September 2025, and OpenAI, whose OAI-SearchBot feeds an internal retrieval system. One genuinely new European index also exists — Staan, built by the Ecosia–Qwant joint venture European Search Perspective, live since August 2025.

That is the complete population. Everything else with a search box is a customer of one of them.

Why most "Google alternative" lists are misleading

The standard listicle of ten or fifteen alternatives to Google usually contains one or two engines with an index of their own and a long tail of Bing customers presented as independent competitors. Set out honestly, a representative list looks like this:

• Yahoo Search — Bing's results since the Search Alliance of 29 July 2009. Its AI answer engine Scout, launched 27 January 2026, is grounded on Bing's index with Anthropic's model on top.
• DuckDuckGo — runs DuckDuckBot and its own Instant Answer indexes, but says on its own help pages that it "largely" sources its web links and images from Bing.
• Startpage — no crawler at all. It relays your query anonymously to Google and Bing. It was Google-only from 7 July 2009 until March 2023, when Bing web results and Bing ads were added.
• Swisscows — states an exclusive partnership with Bing.
• Ecosia — blends Bing, Google and Staan, and does not publish the split.
• Microsoft Copilot — no crawler; it writes a search query you never see, sends it to Bing, and writes prose over the answer.
• Kiddle — a Google Programmable Search Engine instance with SafeSearch and Kiddle's own block lists.
• Dogpile, MetaCrawler and WebCrawler — System1 metasearch brands over bought feeds. WebCrawler, named for the first full-text crawler on the web, has not crawled anything since 2001.

There is hard evidence for how load-bearing the dependence is. During a Bing outage in 2024, DuckDuckGo stopped returning web results — and so did Ecosia and other Bing-fed engines, at the same moment, while engines with their own crawl kept working. No marketing page demonstrates an architecture as clearly as an outage does.

The second piece of evidence is commercial. Microsoft retired every public Bing Search API on 11 August 2025, directing customers to a more expensive AI-grounding product inside Azure. Anything reselling Bing through the public API had to move, pay considerably more, or stop — so any engine that "uses Bing" after that date deserves a follow-up question: under what contract?

Resold results are not a scandal — but know what you are buying

Startpage is the clearest case for taking reselling seriously as a product. It has no index, has never claimed one, and its whole pitch is Google-calibre results without Google building a profile of you. Its Anonymous View proxy, which loads the destination site through Startpage's own servers, is a feature nothing else offers as smoothly. If your objection to Google is surveillance rather than result quality, a proxy is the direct answer and an independent index is not.

What a proxy cannot do is change what you are shown; it changes who is watching. Startpage inherits whatever ranking decisions, omissions or filtering Google and Bing apply, and contributes nothing to search diversity, because every Startpage query is ultimately a Google or Bing query — with a dependency it cannot hedge, since there is no crawler to fall back on.

Two details are worth knowing rather than assuming. The operating company is Dutch and subject to the GDPR, but has been majority-owned since December 2018 by Privacy One Group, a subsidiary of the US advertising company System1 — undisclosed until October 2019. And its European Privacy Seal certifications are real but dated: 2011, recertified 2013 and 2015. Presenting a 2015 seal as current assurance for a 2026 product is the most misleading thing said about it.

Hybrids, and how to read the word

"Hybrid" covers a wide range, and the proportions matter more than the label.

DuckDuckGo is weighted heavily toward its partner: real crawlers, real Instant Answer indexes, but Bing supplies the ranked links. Both popular summaries — "it has its own index" and "it's just a Bing skin" — are wrong.

Qwant is the opposite balance, and carries a cautionary history. In early 2019 an investigation found that Qwant, then marketing itself as an independent European search engine, ran neither its own crawler nor its own indexer and was wholly reliant on Bing. It has since built one: roughly 20 billion indexed pages by its September 2023 statement, alongside a self-measured 51% overlap with Bing, meaning 49% of results differed. It still uses Bing where its own relevance is weak, and for images.

Ecosia resold Bing for most of its life and became a hybrid in August 2025 when Staan went live. It publishes monthly financial reports, which is unusually transparent, but not the Bing/Google/Staan split on any given query.

Kagi is the most misunderstood. It runs its own crawler, Kagibot, feeding its own indexes — Teclis for the web, TinyGem for news — but Teclis is a deliberately non-commercial "small web" index rather than a general one, and the bulk of a typical results page comes from anonymised API calls to third-party commercial indexes. Kagi's documentation named Google, Yandex, Mojeek and Brave until May 2024, when it was generalised to "all major search result providers worldwide". Its independence is real, but it sits at the ranking layer rather than the index layer — and that layer, with permanent per-domain raise, lower and block controls, is the actual product.

What you actually give up when you leave Google

Anyone presenting an independent engine as a straight Google replacement is setting a reader up to be disappointed, and it is worth being blunt about why.

Coverage. Mojeek's 9 billion pages and Marginalia's few hundred million are genuine web indexes, and both are one to several orders of magnitude smaller than Google's. On a large share of obscure or long-tail queries a small index simply does not hold the page. That is a coverage problem, not a ranking problem, and no algorithmic improvement fixes it.

Freshness, language and local. A small crawl budget cannot re-crawl news-velocity sites at Google's cadence — Marginalia's main refresh cycle runs eight to ten weeks. Coverage outside major Western languages is thin on every independent Western engine, and none has anything resembling Google's local pack or maps.

The platform web. The gap is increasingly maintained by contracts rather than technology. Reddit blocked non-Google crawlers in 2024, and an engine that cannot crawl the large platforms cannot show you what is on them, however good its crawler is.

The realistic framing is that an independent engine is an excellent second engine — the one you check to see the web without Google's hand on it, or when you need results reproducible because they are not personalised. Mojeek and Marginalia return the same page to everyone issuing the same query in the same region, which makes them useful in a way Google structurally is not.

Match the engine to the reason you are leaving

Most bad advice here comes from answering the wrong question. Four common reasons for leaving Google point at four different answers.

"I don't want to be profiled." A proxy or a non-logging engine solves this and the index behind it is beside the point. Startpage relays to Google and Bing without your identity; DuckDuckGo stores no search history against an identity, though it is a US company whose defence is data minimisation rather than jurisdiction. Choosing Mojeek here is not wrong, but it solves a different problem at a real cost in coverage.

"I want search that isn't controlled by Google and Microsoft." Only an own-index engine achieves this, and the list is short: Mojeek, Brave, Marginalia, or a regional engine such as Seznam or Naver in its home market. Picking DuckDuckGo or Startpage for this reason is a category error — both route through the duopoly by design.

"I want search without AI summaries." DuckDuckGo publishes an AI-free endpoint at noai.duckduckgo.com that turns every AI feature off with no settings; Mojeek declined to become an answer engine at all; Kagi's AI is opt-in per query. Google offers no account-level switch, only a per-search Web filter.

"I want better results, and I'll pay." Kagi is the only mainstream consumer engine you pay for, and what you buy is control over ranking, not an index independent of Google's ecosystem. Paying also creates a billing identity a free engine never collects — worth knowing before treating it as a privacy upgrade.

Some engines are poor choices for almost anyone. One that publishes no crawler documentation and names no result supplier should be read as an interface over an undisclosed feed until it says otherwise — Gibiru and Presearch both sit there, the latter having claimed its own index since September 2025 without publishing a crawler name, user-agent or robots.txt token.

How to check any engine for yourself

None of this has to be taken on trust. Three checks settle it in about five minutes.

1. Look for a crawler page. An engine that crawls publishes a named user-agent, a verification method by reverse DNS or published IP list, and blocking instructions. Mojeek, Marginalia, Brave, Google, Bing, Yandex, Seznam and Perplexity all have one; an engine with no such page almost certainly has no crawler.
2. Read its own "where do results come from" page. The honest ones say it plainly — DuckDuckGo's sources page names Bing outright. Vagueness such as "we aggregate the best results from around the web" is itself an answer.
3. Compare a narrow, non-commercial query against Bing. If the results are Bing's in a different typeface, you know the architecture.

And keep the scale in proportion. StatCounter put Google at 91.31% of worldwide search in July 2026, Bing at 4.47% and DuckDuckGo at 0.65% — figures measuring tracked page views rather than users or queries, and almost certainly understating engines that strip referrers by design. The point is not that the alternatives are winning. It is that a handful of organisations still crawl the web independently, and knowing which ones is what makes a choice a choice.

Frequently asked questions

Sources

• duckduckgo.com/duckduckgo-help-pages/results/sources
• brave.com/blog/search-independence/
• mojeek.com/about/
• help.kagi.com/kagi/search-details/search-sources.html
• support.ecosia.org/article/579-search-results-providers
• learn.microsoft.com/en-us/lifecycle/announcements/bing-search-api-retirement
• gs.statcounter.com/search-engine-market-share
• en.wikipedia.org/wiki/Startpage