Cuil: shut down in 2010

By Bill Hartzer · Updated 19 August 2026

What Cuil was, and what happened to it

Cuil was an independent web search engine that operated from 28 July 2008 to 17 September 2010. It is listed here as no longer available in the fullest sense of that phrase: the company shut down, and as of 19 August 2026 the domain cuil.com does not resolve at all. A DNS lookup fails outright. It is not parked, not redirected, and not serving a holding page. Of all the discontinued engines documented on this site, Cuil is the cleanest disappearance.

The single most common thing said about Cuil is that it was a fake — a search engine that claimed a giant index and never had one. That is wrong, and correcting it is the point of this page. Cuil built and ran its own crawler and its own inverted index (the engine's own stored copy of the web, assembled by software that fetches pages and records what is on them). At no point did it license or re-serve results from Google, Yahoo or Microsoft. Very few organisations have ever built a general web index from scratch outside the Google/Microsoft axis. Cuil did.

What Cuil could not do was rank. The failure was not that the index did not exist; the failure was that a large index with poor retrieval quality is worth less than a small index with good retrieval quality, and Cuil had staked its entire launch on the number rather than the result.

The index-size claim, and why it was contested

At launch Cuil claimed an index of roughly 120 billion web pages. The precise figure circulated at the time was 121,617,892,992. By February 2009 the claim had been revised upward to 127 billion pages.

Three days before Cuil launched, on 25 July 2008, Google published a blog post stating that its systems had seen 1 trillion unique URLs on the web. That is a different quantity from an index count — URLs seen is not the same as pages indexed — but the timing framed the conversation Cuil was about to walk into. Press coverage widely read Cuil's marketing as a claim to be several times larger than Google, a comparison drawn mostly against older Google index-size disclosures rather than any like-for-like measurement. Neither company published a methodology that would have made the comparison meaningful.

Contemporary critics raised three objections, and all three still hold:

No verification was possible. Cuil published no methodology and offered no reproducible index-count test.
The results contradicted the claim. Queries for reasonably common terms returned thin or stale results, and missed pages Google clearly held. That is not the behaviour of a record-breaking index.
An index-size number means nothing without a quality number beside it. Cuil supplied only the former.

Hands-on reviews at Technologizer, TechCrunch and Fortune all reached broadly the same verdict within 24 to 48 hours of launch: bigger claim, worse results.

Twiceler: the crawler, and the complaints

Cuil's indexing robot was named Twiceler, and it was a real and heavily active crawler. Reported user-agent strings took the form Mozilla/5.0 (Twiceler-0.9 http://www.cuill.com/twiceler/robot.html), carrying the original company name Cuill; the name Twiceler is confirmed, while the exact version string cannot now be checked against a primary page, because both cuil.com and cuill.com are dead.

Twiceler drew substantial complaints from site operators. Webmasters reported that it repeatedly hit their sites with randomly generated URLs, apparently probing for pages that were not reachable by any link, on top of ordinary link-following crawling. The effect was heavy and unwelcome server load. This is worth sitting with, because it is the strongest evidence against the "Cuil had no index" story: an engine that was faking it would not have generated years of hosting-forum complaints about bandwidth.

On ranking, Cuil claimed to analyse page content and the relationships between concepts on a page rather than relying primarily on inbound-link popularity, the model Google's PageRank made standard. Results were grouped into topical categories and presented in an unusual magazine-style multi-column layout with extracted page images and topic tabs, rather than a single column of ten blue links. Cuil also stated that it did not personalise results based on a user's history.

Who ran it, and how it was funded

Cuil, Inc. — originally incorporated as Cuill, Inc. — was a privately held startup based in Menlo Park, California, operating in stealth from roughly 2006 or 2007. Its founders were Tom Costello (chief executive; previously founder of Xift, and a researcher at IBM and Stanford), Anna Patterson (Costello's wife, who at Google had built the TeraGoogle large-document index and had earlier built the Recall web-archive search engine that Google acquired), and Russell Power, also formerly of Google.

The business model was advertising-supported in intent only. Cuil launched without advertising and said it would introduce ads later; it was funded entirely by venture capital for its whole operating life, and there is no evidence it ever generated meaningful revenue. The commercial thesis pitched to investors was about cost: the founders claimed Cuil's architecture would make indexing roughly one-tenth as expensive per page as Google's, so a smaller company could economically run a larger index. That was serious engineering rather than vapour, and the people making the argument had built Google-scale systems before.

Approximately $33 million was raised from Greylock Partners, Tugboat Ventures and Madrone Capital Partners, at a reported pre-launch post-money valuation of about $200 million. An earlier self-funded injection of roughly $5 million by Costello and Patterson preceded the institutional rounds. In December 2008 BusinessWeek named Cuil one of the most successful US startups of the year — explicitly on the basis of money raised rather than users or revenue, which is a fair summary of the company in one line.

Privacy, and an awkward asymmetry

Cuil made privacy a secondary selling point. It stated that it did not store users' search histories or personally identifiable search logs, and that it did not build personalised profiles or rank according to prior queries — an explicit contrast with Google's practice at the time. Those arguments would win DuckDuckGo an audience within a couple of years.

The claim should be read with care. No independent audit, regulator finding or technical study tested Cuil's non-retention policy while the service was running, so it remains a company policy statement rather than a verified fact. Cuil was a US company and would have been subject to US legal process whatever its policy said.

There is also an asymmetry that drew criticism at the time and is still instructive: Cuil promised privacy to searchers while being a notably inconsiderate guest on publishers' servers.

Cpedia, and why Cuil was not an AI search engine

Cuil shut down in 2010, well before the generative-AI search era, and it had no chat interface and no language model. Its "understanding content, not just links" language described statistical and semantic content analysis with concept clustering. Any present-day source calling Cuil an AI search engine is applying a 2023 vocabulary to a 2008 product.

One qualified exception deserves recording. In 2010 Cuil shipped Cpedia, described as "an automated encyclopedia ... built by algorithmically summarizing and clustering ideas on the web." Instead of a list of links, it returned generated encyclopaedia-style reports. Structurally that is the same bet AI search makes today — synthesise an answer rather than list sources — attempted more than a decade early with statistical clustering in place of a language model. It is a legitimate historical antecedent, and it did not recover the company's audience.

The collapse, and the shutdown

The traffic figures tell the story without commentary. Per traffic-ranking panels of the period, Cuil reached just over 0.2% of worldwide internet users in its launch week in late July 2008. By 12 September 2008 it was at 0.02%, ranked around 5,340th by traffic. By 13 October 2008 it was at 0.005%, ranked around 21,960th. Those Alexa-style panels are of limited reliability in absolute terms, but the shape — roughly a fortyfold collapse in under three months — is unambiguous and matches contemporary reporting. On 27 December 2008 TechCrunch published a piece headlined "Cuil Fail: Traffic Nearly Hits Rock Bottom." No audited query-volume figure for Cuil was ever published, and StatCounter's public search-engine series does not break it out.

Launch day had already gone badly in two ways beyond relevance. Cuil's servers buckled under the traffic, producing timeouts and empty result pages. And the image-extraction feature, which paired result snippets with images scraped from the pages, in a number of documented cases attached pornographic or wholly unrelated images to innocuous queries — including, notoriously, to searches for people's own names.

On the morning of 17 September 2010, employees were told about Cuil's demise, and the servers were taken offline five hours later. Staff reported that they had not been paid that week. The shutdown followed a failed acquisition attempt. Afterwards, Cuil's patents were sold to Google and Anna Patterson returned to Google. That is not the same thing as Google acquiring Cuil, and Cuil's technology was never folded into Google Search as a product.

What Cuil is worth remembering for

Cuil led with the one metric it could not be held to and could not deliver on. Index size is a vanity number; retrieval quality is the product. By choosing "biggest index" as its headline, Cuil guaranteed that every journalist's first act would be to run test queries — and the test queries failed. It compounded that by launching without the capacity to serve its own launch traffic, and by shipping an image feature capable of attaching pornography to a search for a reader's own name. A search engine gets roughly one first impression. Cuil spent its on all three failures at once.

The convenient explanation — that Cuil failed because Google is unbeatable — lets it off too easily. Independent engines launched before and since, among them Gigablast, Mojeek and Brave, have survived far longer on far less money. Cuil failed a set of specific, self-inflicted product and operations tests inside 48 hours.

Its value as a reference is the opposite of its reputation. Cuil is the clearest available demonstration that building an independent index is the easy half of the problem, and ranking is the hard half. Anyone who reads a modern engine's index-size announcement and takes it as a proxy for quality should read Cuil's first.

Common misconceptions

"Cuil never actually had an index." Not supported. Cuil demonstrably crawled the web — enough that webmasters complained loudly about Twiceler's load. The dispute was about the size of the claim and the quality of retrieval, not the existence of an index.

"Cuil claimed to be three times bigger than Google." Sloppier than that. Cuil claimed about 120 billion indexed pages; Google had said it had seen 1 trillion unique URLs, which is a different quantity. The "3x Google" framing came largely from press comparison against older Google disclosures.

"Google bought Cuil." No. The company shut down. Its patents were later sold to Google and a founder rejoined Google; neither is an acquisition.

"It's pronounced kwil." The company said it was pronounced "cool". The confusion was itself a branding problem: a name nobody can say is a name nobody can pass on.

"Cuil is Irish for knowledge." That was Cuil's own claim — the company said the name came from Irish Celtic folklore and meant "knowledge" and "hazel" — and it was publicly disputed. A terminology official at Foras na Gaeilge, the official Irish-language body, said: "I am unaware myself of the meaning 'knowledge' being with the word 'cuil' in Irish." Treat it as a contested claim, not as etymology.

Frequently asked questions

Is Cuil still working?

No. Cuil shut down on 17 September 2010, when its servers were taken offline. Checked on 19 August 2026, the domain cuil.com does not resolve at all — a DNS lookup fails outright, so there is no parked page, no redirect and no residual content. Anyone encountering a site trading on the Cuil name today should treat it as unrelated to the 2008 to 2010 company.

Did Cuil have its own index?

Yes. This is the most commonly mistaken fact about Cuil. It built and ran its own web crawler, named Twiceler, and its own index, and it never licensed or re-served results from Google, Yahoo or Microsoft. The criticism levelled at Cuil concerned the claimed size of that index and the quality of the results it returned, not whether the index existed at all.

How big did Cuil claim its index was?

At its launch on 28 July 2008 Cuil claimed roughly 120 billion web pages, with the precise figure of 121,617,892,992 circulating at the time. By February 2009 the claim had been raised to 127 billion pages. Cuil never published a methodology or a reproducible test that would have let anyone verify either number, which was a large part of why the claim was disbelieved.

Why did Cuil fail?

It led with an index-size claim it could not substantiate, and its results did not support it. Reviewers found thin and stale answers to ordinary queries within hours of launch. The servers also collapsed under launch traffic, and an image feature attached unrelated and sometimes pornographic images to innocuous searches. Traffic fell from about 0.2% of worldwide internet users in launch week to 0.005% by mid-October 2008.

Did Google buy Cuil?

No. Cuil shut down in September 2010 after a reported acquisition agreement fell through, and staff were told the company was finished on the day the servers went dark. Afterwards, Cuil's patents were sold to Google and co-founder Anna Patterson, who had previously built Google's TeraGoogle index, returned to Google. Neither event amounts to Google acquiring the company, and Cuil's technology never became part of Google Search.

Who founded Cuil?

Cuil was founded in stealth around 2006 to 2007 as Cuill, Inc. in Menlo Park, California, by Tom Costello, previously founder of Xift and a researcher at IBM and Stanford; Anna Patterson, formerly of Google, where she built the TeraGoogle large-document index; and Russell Power, also formerly of Google. It raised approximately $33 million from Greylock Partners, Tugboat Ventures and Madrone Capital Partners.

Was Cuil an AI search engine?

No. Cuil shut down in 2010, years before generative AI search, and it had no chat interface and no language model. Its talk of understanding content rather than links described statistical content analysis and concept clustering. The one interesting antecedent is Cpedia, a 2010 product that generated encyclopaedia-style summaries instead of listing links — the same structural bet AI search makes now, attempted with the tools of 2010.

How is Cuil pronounced?

The company said it was pronounced "cool". Cuil also claimed the name came from Irish Celtic folklore and meant knowledge and hazel, a claim publicly disputed by a terminology official at Foras na Gaeilge, the official Irish-language body. The pronunciation confusion was a genuine branding problem in itself: press coverage had to explain how to say the name before it could explain the product.

Cuil