Five models, and most engines use more than one
Search is expensive. Crawling the web, storing an index, and answering queries in a fraction of a second are all capital-intensive, and the bill arrives whether or not anyone clicks anything. Every engine therefore runs on at least one of five structural models:
- Advertising — selling placement against the query, at auction.
- Subscription — the searcher pays directly, and there is no advertiser.
- Metered API access — selling the index wholesale to developers and other products.
- Affiliate commissions — a share of transactions that begin with a search.
- Mission or redistribution funding — a conventional revenue line with a binding rule attached about where the surplus goes.
These are not equivalent choices with different branding. Each determines who the engine must satisfy, what data it needs to collect, and what its results page is optimised to do. This page describes mechanisms and quotes no prices: subscription tiers and API rates move, and any figure printed here would be wrong within a year.
Advertising, and the auction that invented it
Advertising funds the overwhelming majority of web search, and the mechanism is more specific than "they show ads".
It was invented by GoTo.com, Bill Gross's spin-out from the Idealab incubator, demonstrated publicly at TED in February 1998. Three components, novel in combination: advertisers bid against each other for a keyword; ranking followed bid order; and the advertiser paid only when someone clicked. Bids started at one cent. From 1 June 1999 GoTo shipped self-serve tooling so advertisers could manage bids without a salesperson — an underrated part of the invention, because an auction only scales if tens of thousands of small advertisers can operate it unassisted. By the first quarter of 2003 the company, by then renamed Overture, had more than 88,000 advertisers.
GoTo never crawled the web. Its own "results" were the paid listings; unpaid algorithmic results were bought in from Inktomi from June 1998. The real business was syndication: placing its paid listings inside other engines' results pages and splitting the click revenue. A "GoTo-in-a-Box" agreement dated 15 April 2001 put its listings on Microsoft properties, a Yahoo deal followed on 20 November 2001, and AOL and Excite carried the feed too. Yahoo bought Overture outright on 7 October 2003 for $1.63 billion.
Google launched AdWords in 2000 and moved to a pay-per-click auction with AdWords Select in 2002. Overture sued for patent infringement in April 2002; the case settled in 2004, with Google issuing 2.7 million shares of common stock to Yahoo in exchange for a perpetual licence under the patent. Every search advertisement served since is a descendant of that 1998 design.
The scale of the result is easy to understate. Alphabet's 2017 annual report attributed 86% of revenues to advertising, and "Google Search & other" remains the largest single line item in Alphabet's quarterly results.
What advertising does to the product
The important consequence is not that ads appear. It is what an ad-funded engine is incentivised to know about you.
There are two ways to make a search ad valuable. You can target the query — someone typing a commercial phrase has declared intent, and that is worth money by itself. Or you can target the person, using a behavioural profile assembled across products and sessions, which raises the price per impression and requires sustained data collection.
Google does both, and the second is why its privacy posture is what it is: signed-in search and browsing activity, location, device identifiers and cross-product signals from Android, Chrome, YouTube and the ad network combine into the most comprehensive commercial behavioural profile of any search engine. That is not an accident or an oversight; the ad business depends on it.
DuckDuckGo is the working demonstration that the first path alone is viable. Its ads are matched to the query and a coarse region, not to a user profile, and most ad clicks are handled through Microsoft Advertising. Its own position is that keyword-matched ads on commercial-intent queries are valuable from the query alone. The trade-off is real and structural: you earn less per user, so the model works at a scale below Google's and would be harder to sustain at Google's cost base.
There is also a wholesale layer above all of this. Default placement is bought. Google's payments to Apple were central to the US antitrust case against it, and a November 2023 disclosure put the arrangement at 36% of search advertising revenue generated through Safari. The searcher who never changes a default is, in commercial terms, the product being traded.
Subscription: the searcher as the customer
The cleanest alternative is to charge the person doing the searching and have no advertiser at all.
Kagi is the operating example. There is no advertising anywhere in the product and no ad-derived ranking, because there is no advertiser to serve. Its argument is that the subscription is the alignment mechanism: with no third party paying, the only party the ranking has to satisfy is the person who paid. That funding model is what makes its distinctive features possible — per-domain controls that let a subscriber raise, lower, pin or permanently block any site from their own results, and filters for low-value content. No advertising-funded engine can offer a block button, because the blocked domains would include the advertisers.
Kagi crossed $1 million in annual revenue in 2023, raised roughly $670,000 in a fundraise open to its own users that year, and converted to a public benefit corporation in 2024. Tiers have been restructured more than once, which is why no prices appear here.
The weakness is equally structural: people are habituated to search being free, and the addressable market is whoever will break that habit. Neeva, founded by former Google advertising executives on an explicitly ad-free subscription premise, did not reach sustainable consumer scale and shut down. Subscription search is viable; it is not obviously scalable.
DuckDuckGo added a subscription line of its own in April 2024 — a privacy bundle including a VPN and personal-information removal — which was its first meaningful revenue that is not advertising or affiliate.
Metered API access: selling the index wholesale
An index is an asset independent of the consumer product built on it, and selling access to it is now one of the more strategically important models in search — because independent indexes are rare and the demand for grounded retrieval has risen sharply.
Mojeek sells a Web Search API and a Site Search API alongside its own search advertising. It is small, but it owns something scarce: one of very few general web indexes on earth that is not Google, Bing, Yandex or Baidu. That scarcity became commercially obvious when Microsoft raised Bing API pricing sharply in 2023 and then retired all public Bing Search APIs on 11 August 2025, removing at a stroke the default way most third parties had bought web results. Partners with longstanding direct syndication contracts, DuckDuckGo among them, were reported not to be immediately affected — but the terms of those contracts are not public, and the episode showed how much of the ecosystem rested on one supplier's pricing decisions.
The same logic drives Ecosia and Qwant's joint venture. European Search Perspective, announced in November 2024 and headquartered in Paris, was deliberately incorporated as a separate capital-raising entity to build a European web index, Staan, which went live in August 2025. Selling index access to third parties — AI companies in particular — is intended to be a revenue line in its own right, not merely a cost centre serving the two consumer brands.
Kagi sells metered AI endpoints built on its own Teclis and TinyGem indexes, and even Marginalia, a one-person project, offers a keyed API — attractive as a retrieval source precisely because its index is full of non-commercial text.
Affiliate commissions and mission funding
Affiliate commissions take a share of transactions that begin with a search. DuckDuckGo names Amazon and eBay affiliate relationships and a TripAdvisor arrangement, and states that no personal information is passed to them. The model has two structural properties worth knowing. It only monetises commercial queries, so it can never fund an engine on its own. And it creates an obvious suspicion — that ranking might favour whatever pays a commission — which is why engines using it publish explicit statements that it does not influence results. Those statements are company claims, not audited findings.
Mission funding is an advertising business with a binding rule attached about where the surplus goes. Ecosia is the significant example: it shows search ads, largely through Microsoft Advertising and Google, and commits the large majority of profits to tree planting and landscape restoration. Its published 2021 split was 80% of profits to tree planting, which Ecosia characterised as 47.1% of total income — 2021 figures, and they should always be cited with that date. Reported revenue was approximately €43.2 million for 2023; no later audited figure is confirmed. Ecosia reported reaching 250 million trees around Earth Day in April 2026, and has since reframed its mission toward broader landscape restoration.
Two caveats belong with that. The tree count is self-reported and verified through planting partners rather than by an independent auditor — better disclosure than most, but not third-party attestation. And Ecosia has moved away from any fixed "searches per tree" ratio, because it depends on ad revenue per search, which varies by market.
The category also contains models that did not work. Cuil launched in July 2008 with no advertising at all, funded by roughly $33 million of venture capital and intending to add ads later; it never generated meaningful revenue and shut down in September 2010.
What the funding model tells you
The single most useful question to ask about any search engine is not what features it has. It is who pays for this, and what do they get.
If the answer is advertisers, the incentive is to hold your attention and usually to learn about you — though DuckDuckGo shows the learning part is separable from the advertising part. If the answer is you, the engine has no reason to keep you on its page longer than useful, and can afford to let you block domains it would otherwise be paid to show. If the answer is developers buying API calls, the consumer product may be a shop window for the index. If the answer is a mission, you are still using an advertising engine; the difference is in the distribution of profit, not the mechanics of the page.
None of these is disqualifying and none is automatically virtuous. But the model is the most reliable predictor of how an engine will behave when its interests and yours diverge, and unlike ranking quality it is a matter of public record.
The primary sources are worth reading directly: DuckDuckGo's own account of how it makes money and the history of GoTo.com, the company that built the auction everything else is descended from.
Frequently asked questions
How do search engines make money?
Through five structural models, usually in combination: advertising sold at auction against the query, direct subscriptions paid by searchers, metered API access selling the index wholesale to developers, affiliate commissions on transactions beginning with a search, and mission funding where advertising revenue is committed to a cause. Advertising dominates by a very large margin.
Who invented paid search advertising?
GoTo.com, founded by Bill Gross at the Idealab incubator, demonstrated the model publicly at TED in February 1998. It combined keyword bidding, ranking by bid order, and payment only on a click. Google launched AdWords in 2000 and moved to a pay-per-click auction in 2002; Overture sued in April 2002, settling in 2004 for 2.7 million Google shares issued to Yahoo.
Can a search engine sell ads without tracking users?
Yes, and DuckDuckGo is the working demonstration. Its ads are matched to the query and a coarse region rather than to a behavioural profile, on the basis that commercial-intent queries are valuable from the query alone. Most ad clicks are handled through Microsoft Advertising. The trade-off is lower revenue per user, which constrains the scale the model can support.
How does a paid search engine like Kagi work?
Subscribers pay directly and there is no advertising anywhere in the product, so no advertiser influences ranking. That funding model enables features advertising cannot support — notably per-domain controls letting a subscriber raise, lower, pin or permanently block any site from their own results. Kagi crossed $1 million in annual revenue in 2023 and converted to a public benefit corporation in 2024.
Why do search engines sell API access to their indexes?
Because a web index is a scarce asset with demand beyond the consumer product. Microsoft raised Bing API pricing sharply in 2023 and retired all public Bing Search APIs on 11 August 2025, which made the remaining independent indexes commercially valuable. Mojeek sells Web Search and Site Search APIs; Ecosia and Qwant built Staan partly to sell index access.
How does Ecosia's tree planting model work financially?
Ecosia is an advertising business with a distribution rule attached. It shows search ads through Microsoft Advertising and Google and commits the large majority of profits to tree planting and landscape restoration. Its published 2021 split was 80% of profits, which Ecosia described as 47.1% of total income. The tree count is self-reported through planting partners, not independently audited.
Has any search engine failed because of its business model?
Several. Cuil launched in July 2008 with no advertising, funded by roughly $33 million of venture capital, and shut down in September 2010 having never generated meaningful revenue. Neeva, founded by former Google advertising executives on an ad-free subscription premise, did not reach sustainable consumer scale and closed. Subscription search is viable but has not proven broadly scalable.
Do search engines pay to be the default on your device?
Yes, and it is one of the largest transactions in the industry. Google's payments to Apple for default placement were central to the US antitrust case; a November 2023 disclosure put the arrangement at 36% of search advertising revenue generated through Safari. A user who never changes a default is, in commercial terms, the thing being traded.
Sources
- thehistoryoftheweb.com/goto-forgotten-search-engine/
- bgtheory.com/blog/goto-to-overture-to-ysm-timeline/
- duckduckgo.com/duckduckgo-help-pages/company/how-duckduckgo-makes-money/
- help.kagi.com/kagi/company/history.html
- en.wikipedia.org/wiki/Ecosia
- learn.microsoft.com/en-us/lifecycle/announcements/bing-search-api-retirement
- en.wikipedia.org/wiki/Alphabet_Inc.
- en.wikipedia.org/wiki/Google_Ads