A9.com

» see a9.com as it lived, on the wayback machine

~ the obit ~

a9.com was amazon's 2003 attempt to be a search engine. it never reached me. in 2003 i was a 19 year old computer engineering student in ankara on a ttnet adsl line, google had been the default search box for two years and there was no daylight in my own browsing for a second one. a9 was a side product of amazon and most of its traffic came from people who had clicked something on amazon.com itself. it never had a turkish user base.

the startup formally launched in september 2003 in palo alto under the lead of udi manber, who had been a search engineer at yahoo and had written a textbook on string algorithms that turkish algorithms classes still assign in 2026. amazon set up a9 as a wholly owned subsidiary, in a separate building in the former dec systems research center, with the brief to build an engine that could compete with google on the open web. the founding bet was that adding amazon shopping signals to the web results would matter to shoppers.

the public face was the a9.com page, which loaded web results on the left from a google api license, book search results from amazon in the middle, image and movie panels from imdb on the right, and a customizable search history at the bottom. the design was busy but interesting, and search blogs in 2004 and 2005 spent real attention on a9 as the closest thing google had to a credible second engine.

~ the rap sheet ~

~ the rise ~

the high point came in 2005. a9 shipped opensearch, a small xml format that let any website describe its own search interface so a9 could include it alongside the main results. opensearch outlived a9 by a wide margin, firefox and ie7 both adopted it, every search-add-on installer on every browser through 2010 used the same opensearch xml file format under the hood. it is still inside the wikipedia search box in 2026.

the other 2005 shipment was blockview, street-level photos of about ten major us cities taken from amazon trucks driving the streets with mounted cameras. you searched for a restaurant in san francisco and a9 showed you what its storefront looked like. google did the same project under street view a couple of years later. the difference is that google kept driving and a9 stopped after the first cities. the blockview photos were good for one news cycle and then sat as a curio on the a9 dashboard.

~ the postmortem ~

the cracks showed in 2006 when udi manber left for google. that was the search engineer the whole company had been built around. by 2007 the public a9.com page had dropped most of its distinctive widgets. in october 2008 amazon announced that a9.com would stop being a public search engine and would redirect traffic to a generic amazon search instead. the team was kept on, the engine was kept running, both quietly moved to powering amazon.com search itself.

that 2008 redirection was not a normal shutdown. there was no death date, no shutdown blog post, no migration tool. the page went thin, the customization tools disappeared, and the brand stopped being mentioned. amazon never said a9.com was over. a wall street journal piece in 2019 about amazon search bias caused enough discomfort that amazon took the a9.com domain offline entirely and redirected it to amazon.com. that 2019 quiet retirement was the second death, of the brand.

~ the funny part ~

a9 still runs every search on amazon.com in 2026. the team kept building product search inside amazon for two decades, just under a different name. when you type a product into the amazon box on your phone, the result-ranking system is the lineal descendant of the 2003 a9 engine that udi manber set up in palo alto. the public engine that was supposed to compete with google died, the internal engine that nobody outside amazon ever saw became one of the largest commercial search systems in operation.

the broader lesson the search world took from a9 was that even amazon at full size could not start a general web search engine from scratch against google in 2003. cuil tried again in 2008 and died inside two years. powerset tried with a small team and got bought by microsoft. blekko tried with human curation and got bought by ibm. a9 is the version of that story where the company doing it was big enough to absorb the failure and keep the team. the engine survived, the brand did not.

@}-->--

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

« previous  ·  random  ·  next »