SavvySearch

What SavvySearch was, and where its results came from

SavvySearch never had an index of its own. It was a metasearch engine: you typed one query, and it fired that query at a dozen or more other search engines at the same time, then merged and de-duplicated what came back into a single list. It owned the dispatcher, the ranking logic and the interface, and nothing else. Every result on the page had been found by someone else's crawler.

That is a neutral structural fact rather than a criticism. In 1995 it was arguably the best available answer to a real problem: no single engine indexed more than a fraction of the web, and — crucially — each one indexed a different fraction. Asking twelve engines really did surface material that any one of them would have missed.

The service is gone. CNET absorbed SavvySearch into its Search.com property in October 1999, and the brand was progressively retired from there. No announcement fixing the date on which savvysearch.com stopped answering queries has been found, so the honest statement is that the original service ended some time after the acquisition, and that it has not existed for a very long time.

The idea: an engine that learned which engines to ask

Most metasearch engines of the period simply broadcast every query to every back end they knew about. SavvySearch did not, and this is the reason it matters more than its contemporaries. It kept statistics on which back-end engines had historically produced results people actually selected, for which kinds of query, and used those statistics to decide where to send each new query — while also watching network load and response times so that it neither overwhelmed itself nor hammered the engines it depended on.

Unusually for a commercial web service of the 1990s, this is documented in the peer-reviewed literature rather than in marketing material. Adele E. Howe and Daniel Dreilinger published "SAVVYSEARCH: A Metasearch Engine That Learns Which Search Engines to Query" in AI Magazine 18(2) on 15 June 1997. The abstract states the design plainly:

Metasearch engines reduce the user burden by dispatching queries to multiple search engines in parallel. The SAVVYSEARCH metasearch engine is designed to efficiently query other search engines by carefully selecting those search engines likely to return useful results and responding to fluctuating load demands on the web. SAVVYSEARCH learns to identify which search engines are most appropriate for particular queries, reasons about resource demands, and represents an iterative parallel search strategy as a simple plan.

Three components are worth naming precisely:

• Search-engine selection by learned relevance. SavvySearch built a meta-index associating query terms with the back-end engines that had performed well on similar terms, updated from live usage. Built in 1995, this is recognisably what would now be called learned query routing.
• Resource-aware dispatch. The system modelled network load and engine response time and degraded gracefully under pressure rather than timing out — a practical necessity for a service running on university infrastructure.
• Iterative parallel plans. A query was executed as a plan over several parallel rounds, not as one indiscriminate fan-out.

Which specific engines it queried, and how many, changed continuously and is not reliably documented at any dated point. Contemporaneous accounts put it in the "over a dozen" range, categorised by type — web, Usenet, software archives, news, people-finders — but no count from a dated source could be confirmed, so none is given here.

Where it came from: a graduate project at Colorado State

SavvySearch began as academic work at Colorado State University, created by graduate student Daniel Dreilinger and supervised by Professor Adele E. Howe of the computer science department, who co-authored the published paper. It was later commercialised as a separate business run from Fort Collins, Colorado; the corporate entity name, incorporation details, funding and headcount are not on any record that could be located.

The launch date is contested, and the conflict is unresolved. Search Engine Watch has published both dates, two years apart, in its own pages:

• March 1995 — Search Engine Watch's "Search Engine Birthdays": "Created by Colorado State University student Daniel Dreilinger, SavvySearch was one of the first meta search engines."
• May 1995 — Search Engine Watch's metasearch guide of 23 March 2005: "around since May 1995 and formerly based at Colorado State University".

Same publisher, two answers, no primary source resolving them. Both are printed here rather than averaged or silently chosen. The safe formulation is that SavvySearch launched in 1995 as a Colorado State University project.

The university origin is the origin, not the whole story: by the time of the sale, SavvySearch was a commercial operation, not a research service. But the academic phase is what left the paper trail, and it is why this engine can be described precisely while several of its commercial contemporaries cannot.

How it made money, and why CNET bought it

The model was the standard 1990s one: a free consumer service with banner advertising on the results page. As a metasearch engine it had no crawling costs and no content costs, which made it cheap to run — and left it with nothing proprietary to defend. Every back-end engine it depended on could cut it off, start competing with it, or both. It also had the structural awkwardness that afflicts all metasearch: it sent users onward to engines that would rather have had those users directly.

In October 1999 CNET acquired SavvySearch for approximately US$22 million in cash and stock and folded it into Search.com. CNET's stated rationale, per Ad Age's report of the deal, was integration rather than standalone advertising revenue: it said the acquisition would give users additional speed and access to product information and tech-focused resources, enabling search customisation and expanded comparison shopping including live auction listings. In other words, CNET bought a search front end to attach to its shopping and technology-media business, not to run a general web search engine.

That US$22 million is also the only quantitative anchor available. No market-share or traffic figure for SavvySearch was ever published that could be located, and StatCounter's series does not begin until 2009. A modest sum even by 1999 standards suggests a well-regarded technology asset rather than a mass-market destination — which matches how contemporaneous accounts describe it, alongside MetaCrawler and Dogpile.

Privacy, and a warning about the name today

There is no current privacy question, because there is no current service. One historical point is genuinely worth making, though: SavvySearch's learning system was built on aggregated user behaviour. The relevance model improved precisely because the system recorded which results users selected, per query term, across its whole user base. That was unremarkable at the time and is unremarkable now, but it means SavvySearch was by design a query-and-click-logging service. No privacy policy, regulatory action or independent audit was located, and the absence of complaint in 1996 is not evidence of restraint.

A caution for anyone researching the name today. There is a separate and entirely unrelated browser hijacker documented under names including "savvy search", with malware-removal guides that currently rank well for the term. It has no connection to Dreilinger's project. As of 19 August 2026 the savvysearch.com domain still responds at the network level, but what is served there could not be retrieved and should not be assumed to be benign or historical. Nothing at that address is the 1995 metasearch engine.

What is not known

This is an obscure engine and parts of its record are simply missing. Rather than filling them in:

• The exact launch month — March and May 1995 are both published by the same source.
• The date savvysearch.com stopped serving the original metasearch service.
• The corporate entity name, incorporation date, investors and size of the commercial company in Fort Collins.
• A dated count of how many back-end engines it queried.
• Any traffic, query-volume or market-share measurement, at any point.
• Whether a standalone English Wikipedia article exists at any title — none surfaced, and SavvySearch appears only within Wikipedia's general metasearch coverage.

What it got right, and what it got wrong

It was genuinely good at two things, and one of them still matters. First, in 1995–1997, metasearch was not a gimmick but the correct response to a fragmented index landscape. Second, and more durably, the engine-selection mechanism was serious research: it was published and peer-reviewed in AI Magazine, and it anticipated the query-routing problem that federated and multi-source retrieval systems still face. SavvySearch is the best-documented of the 1990s metasearch engines for a simple reason: it came out of a university and had to justify itself in print.

What it was bad at was speed and dependence. Every query was gated by the slowest back end it chose to wait for, and merging results from engines with incompatible ranking schemes often produced a combined list worse than the best single source. Structurally it owned nothing. The moment one engine's index became dramatically better than all the others — which is what happened when Google arrived in 1998 — the entire premise collapsed. Metasearch is a strategy for a world in which nobody is good, and that world ended.

Nobody can use SavvySearch, so its value now is as history: the best-documented academic origin story in 1990s metasearch, and a clean illustration of why aggregation loses to a superior index.

Common misconceptions

• "SavvySearch was a search engine." It was a metasearch engine. It never crawled the web and never had an index; everything it returned came from somewhere else.
• "It launched in March 1995." Possibly. The same publisher has printed both March and May 1995, and no primary source resolves it. "1995" is the defensible answer.
• "CNET shut it down in 1999." CNET absorbed it into Search.com in October 1999. The technology continued inside a CNET product and the brand faded; there is no documented shutdown date.
• "savvysearch.com today is the old SavvySearch." It is not, and the modern uses of the name in adware contexts are unrelated to the university project.
• "The 'AI' framing is a retrospective label." Unusually, no. The system was published in AI Magazine, the American Association for Artificial Intelligence's journal, specifically for its learning mechanism. There were no AI answers or summaries — the output was a merged list of links — but the learning was real and documented.
• "It was a Colorado State University product to the end." It originated there as supervised student research, but it was commercialised as a separate business before the CNET sale.

Frequently asked questions

Sources

• searchenginewatch.com/sew/news/2064538/search-engine-birthdays
• searchenginewatch.com/2005/03/23/metacrawlers-and-metasearch-engines/
• adage.com/article/news/cnet-acquires-savvysearch/7934/
• ojs.aaai.org/aimagazine/index.php/aimagazine/article/view/1290
• dl.acm.org/doi/abs/10.1609/aimag.v18i2.1290