Cuil

» see cuil.com as it lived, on the wayback machine

~ the obituary ~

cuil launched on 28 july 2008. the founders were tom costello, an irish search-engine researcher who had worked at ibm and stanford, and anna patterson, a former google engineer who had run a large-scale indexing project there. russell power, also ex-google, was the third co-founder. the pitch was that cuil indexed 120 billion pages, three times as many as google's claimed index at the time, and that its ranking was based on content analysis rather than link popularity. the press picked it up as a "google killer" almost reflexively. the name was pronounced "cool," from the irish word for hazel or knowledge.

launch day broke immediately. the servers buckled under traffic. the results were the bigger problem. queries returned obviously wrong matches, including unrelated images attached to result snippets in ways that produced widely-shared screenshots of porn appearing next to children's media queries. the privacy story (no ip logging, no user accounts) drew some interest, but it was overwhelmed by reviewers writing that cuil had no business calling itself a google challenger when basic queries did not work.

cuil tried for two more years. the company shipped an automated encyclopedia called cpedia in 2010, generating wikipedia-style entries by stitching crawled web text together algorithmically. cpedia produced output that read like word salad on most topics. an acquisition deal was reportedly close in mid-2010 but fell through. the servers shut down on 17 september 2010. employees were laid off without severance pay. the cuil.com domain went dark and stayed dark.

i never used cuil seriously. i was a working engineer in 2008 and would have heard about it the week of launch from tech press, opened cuil.com once, run a query, gotten useless results back, and gone back to google before the tab finished loading. that was approximately the experience the rest of the internet had as well. cuil was the kind of product i remember as a brand more than a tool. a logo that flashed past in news feeds for two years and then was gone.

~ the case file ~

~ what it was ~

cuil's technical pitch was a content-based ranking system. instead of pagerank, which weighed links between pages, cuil claimed to analyze the text of every indexed page and group results by the concepts they shared. the result page showed clusters of related topics down the right side, intended to help users narrow a query the way faceted search did in product catalogs. the visual layout was three or four columns of result thumbnails rather than the linear ten-blue-link list google had standardized.

the index size was the marketing centerpiece. cuil announced 120 billion pages indexed at launch, a number disputed by competitors and never independently verified. the figure was meant to signal scale at a moment when google's ability to crawl the entire web was a major part of its mystique. cuil could not deliver the search-quality experience that index implied, and the size number became a punchline once the result quality became known.

~ why it failed ~

the central failure was relevance. search ranking is the actual product, and cuil's ranking did not work. wrong results made the index size irrelevant. the irrelevant-images bug at launch was the most-cited example, but the deeper problem was that the content-cluster approach surfaced topical neighbors of a query rather than the specific page the user wanted. for a navigational search like "wells fargo login," that approach is an actively wrong shape for the problem.

the secondary failure was operational. servers buckled under launch traffic that was not, by google's standards, particularly heavy. the company had raised $33 million in private funding, which sounds large but is small for the kind of infrastructure required to crawl and serve at the scale cuil claimed. the press cycle moved on quickly when the obvious follow-up demos kept producing odd results. by 2009 most coverage of cuil was retrospective, treating it as a story already over.

~ what carried over ~

the lesson that scale claims do not save a search engine. cuil had ex-google engineers, real funding, and a credible technical premise. it failed anyway because the results page did not work for the queries users typed. that lesson informed every later search-engine attempt, from blekko to neeva. each one led with relevance demos before any kind of size claim.

the dictionary entry. internet writing now uses "cuil" as light slang for a confidently launched product that immediately falls apart. the irish-language pun built into the brand survived in jokes long after the product itself stopped resolving. it sits next to flooz and pets-dot-com in the small list of dot-com era names that stuck around mostly as warnings.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »