The Search Engine Graveyard: A New Resident
May 5, 2026
Another dinobaby post. No AI unless it is an image. This dinobaby is not Grandma Moses, just Grandpa Arnold.
I was working for a search-and-retrieval company when AskJeeves.com became available in 1997. As it turned out, the natural language breakthrough that set AskJeeves apart from the other Web search engines was its question-answering angle. The firm at which I worked hired “content specialists.” From interviewing job seekers, I learned that AskJeeves’ approach was to can certain common questions. The answers to these questions would be updated. Some were automated like “What’s the weather in San Francisco?” but others required a human to craft a response. Other queries were passed to a search-and-retrieval system. Manual processes here are expensive. AskJeeves, therefore, bought “promising” companies for their indexing and content processing capabilities; for example, Jigsaw Technologies in 2000, Direct Hit Technologies in 2000 (specializing in search result ranking), and Teoma Technologies in 2001. AskJeeves tried repurposing its technology for customer service. But Google was maturing into the organization we all know today. In 2005, Barry Diller added AskJeeves to his collection of Internet properties. After the acquisition, Mr. Diller learned that Web search was a difficult and expensive business. The Ask.com service became a metasearch system, recycling search results from other Web indexing outfits in an effort to reduce costs.
Mashable has now reported that Ask.com is dead. “Every Great Search Must Come to an End” said:
Amid an overwhelming shift toward generative AI-powered search engines and a repositioning of AI agents as the future of web browsing, the loss of Ask.com feels like a true end of the early dot-com era. So long Jeeves, hello AI.
I want to add a bit of color to the demise of this Web search system.
My view is that smart software is indeed search-and-retrieval, just with bells and whistles. Systems like AskJeeves knew that handling queries from users was a tricky business. A certain percentage of queries were repetitive. These could be created and later cached. The acquisitions made clear that the original founders could not innovate in substantive ways. Garrett Gruener and David Warthen could recognize interesting technology and its applications. The acquisitions added some scope to the AskJeeves service, but financial realities sparked a sale to Barry Diller’s IAC in 2005. Web search became the province of deep-pocket entities like Google and Microsoft. These firms’ money came from reasonably solid revenue streams. Google sold ads and its pay-to-play model, and Microsoft licensed software. Without meaningful regulation, Google-type organizations trampled over companies like Lycos and All-the-Web, among others. .
This means that today, search-and-retrieval technology exists but has adopted a new vocabulary. The constants are the same: Expensive, complex, and expensive. Did I mention expensive?
The trajectory of AskJeeves is essentially the same for other search-and-retrieval enterprises: Rollout, technical enhancement, utility function, and disappearance or replacement by a spiffed-up version of the old stuff. If this sounds like the trajectory of artificial intelligence, I have made my point. One can apply this general pattern to Autonomy plc, Fast Search & Transfer, and dozens of search-and-retrieval systems that did not evolve into viable businesses. The technology may chug along in a content management system or may be used to perform a background activity, but the spotlight is not on old-school content search. Instead, attention is paid to smart software that requires massive infrastructure to do what humans did for AskJeeves. I would suggest that human-intermediated systems are more common than the marketers want to communicate. Therefore, AI is probably going to follow an AskJeeves type of fate over the next decade or two.?
Why do I suggest this? Here are my reasons based on my research while writing several books about search, including The New Landscape of Search, CyberOSINT: Next Generation Information Access, and The Enterprise Search Report 1st, 2nd, and 3rd editions, among others.
- Indexing can be automated, but one must know what words or phrases to use in the query in order to match certain content. A search in Bing, Google, or Yandex for “financial fraud” will not allow a teen to become a criminal in 10 minutes. Enter the term “carding,” and the game changes. Even today, software cannot replicate this “lingo knowledge.” Many tricks are used to try to know what the user really wants, but these fall short. The tricks like “field codes” themselves become because a person looking for information must know the code to get the chunked results..
- Content is fluid. Language is fluid. Search systems such as those used by Dialog’s or SDC perform best with static terminology. Scholars like static terminology. Indexing conventions try to cope with contextual issues; for example, does “terminal” mean “train station” or does it mean “mainframe peripheral”? The money pumped into smart software is trying to solve this basic problem for many user queries (or in new lingo, “user prompts”).
- The context of information is [a] volatile because today’s problem may not have existed yesterday and [b] situational; that is, every user operates within an “information ecosystem.” Outsiders have a tough time knowing what the characteristics of the ecosystem imply; for example, “loca” may mean one thing to a YouTube cruise personality and another thing to a person working in nuclear safety engineering. That’s why the efforts at personalization are becoming increasingly invasive. Ecosystem information is needed to provide somewhat useful outputs. What if that ecosystem is classified? Well, the big vendors don’t care. They will take what they can get because without it, the outputs are likely to be wrong or potentially quite problematic.
With the reality of change in these three facets of search-and-retrieval, it is appropriate to appreciate the efforts so many people have contributed to making “search” better. Too bad that most of these systems have failed and burned massive sums of money as they trail flames and smoke across the conference rooms in which revenue talks are held.
I have resisted writing about smart software. Everyone I meet is convinced that artificial intelligence is, by golly, the next big thing. Okay. I have other topics to research. I do want to remind readers that smart software is nothing more than search software wearing the latest designer jeans. That does not make it bad. I think the current skepticism about AI is a normal reaction to the discovery that hallucinations, high costs, and AI systems making decisions about health care, education, and judicial actions will present some problems going forward.
Remember. Search is difficult. Knowledge value requires verifiable facts and a foundation of generally accepted information. Without that, system outputs are useless and potentially harmful. Search gets traction because the systems so far developed don’t quite solve a user’s problem. Thus, search is a work in progress, and that progress is expensive. Mr. Diller pulled the plug.
I want to add a bit of color to the demise of this Web search system.
My view is that smart software is indeed search-and-retrieval just with bells and whistles. Systems like AskJeeves knew that handling queries from users was a tricky business. A certain percentage of queries were repetitive. These could be canned and latter cached. The acquisitions made clear that the original ideas and the original founders could not innovate in substantive ways. The founders, Garrett Gruener and David Warthen, could recognize interesting technology and its applications. The acquisitions added some scope to the AskJeeves service, but financial realities sparked a sale to in 2005. Web search became the province of deep pocket outfits like Google and Microsoft. These firms’ money came from reasonably solid revenue streams. Google sold ads or the pay-to-play model and Microsoft licensed software. Without meaningful regulation, Google-type outfits trampled over Lycos- and All-the-Web type outfits.
This means that search-and-retrieval today exists but it has adopted a new vocabulary. The constants are the same: Expensive, complex, and expensive. Did I mention expensive?
The trajectory of AskJeeves is essentially the same for other search-and-retrieval outfits: Roll out, technical enhancement, utility function, and disappearance or replacement by the old stuff spiffed up. If this sounds like the trajectory of artificial intelligence, I have made my point. One can apply this general trajectory to Autonomy plc, Fast Search & Transfer, and dozens of search-and-retrieval systems that have not evolved into viable businesses. The technology may chug along in a content management system or be used to perform a background activity. But the spotlight is not on old-school search-and-retrieval. The bright new manifestations of search and retrieval capture attention. Hint: smart software that requires massive infrastructure to do what humans did for AskJeeves. I would suggest that human-intermediated systems are more common than the marketers want to communicate. Therefore, AI is probably going to follow an AskJeeves type of trajectory over the next decade or two.
Why do I suggest this? Here are my reasons based on my research and writing of a number of books about search, including The New Landscape of Search, CyberOSINT: Next Generation Information Access, and The Enterprise Search Report 1st, 2nd, and 3rd editions, among others.
- Indexing can be automated but one has to know the words or phrases to use in the query in order to match certain content. Today one can navigate to Bing, Google, or Yandex and search “financial fraud.” The results will not allow a teen to become a criminal in 10 minutes. Enter the term “carding” and the game changes. Even today, software cannot replicate this “lingo knowledge.” Many tricks are used to try to know what the user really wants, but these fall short. The tricks themselves become problematic.
- Content is fluid. Language is fluid. Search-and-retrieval, whether old-school like Dialog Information’s or SDC’s approach, likes static terminology. Scholars like static terminology. Indexing conventions try to cope with contextual issues; for example, does “terminal” mean train station or does it mean “mainframe peripheral”? The money pumped into smart software is trying to solve this basic problem for many user queries or in new lingo “user prompts”.
- The context of information is [a] volatile because today’s problem may not have existed yesterday and [b] situational; that is, every user exists within an “information ecosystem.” Outsiders have a tough time knowing what the characteristics of the ecosystem mean; for example, “loca” may mean one thing to a YouTube cruise personality and another thing to a person working in nuclear safety engineering. That’s why the efforts at personalization are becoming increasingly invasive. Ecosystem information is needed to provide useful outputs. What if that ecosystem is classified? Well, the big vendors don’t care. They will take the information because without those data, the outputs are likely to be wrong or potentially quite problematic.
With the reality of change in these three facets of search-and-retrieval, one has to appreciate the efforts so many people have contributed to making “search” better. Too bad that most of these systems have failed and burned massive sums of money as they trail flames and smoke across the conference rooms in which revenue talks are held.
I have resisted writing about smart software. Everyone I meet is convinced that artificial intelligence is — by golly — the next big thing. Okay. I have other topics to research. I do want to remind anyone reading this short blog post that smart software is nothing more than search and retrieval wearing the latest designer jeans. That does not make it bad. I think the current skepticism about AI is a normal reaction to people discovering that hallucinations, high costs, and specter of AI systems making decisions about health care, education, and judicial actions is going to present some problems going forward.
Remember. Search and retrieval are difficult. Knowledge value requires verifiable facts and a foundation of generally accepted information. Without that system outputs are useless and potentially harmful. Search gets traction because the systems don’t quite solve the user’s problem. Thus, search is a work in progress, and that progress is expensive. Mr. Diller pulled the plug.
Stephen E Arnold, May 5, 2026
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
