Inference Find (InFind)

By Bill Hartzer · Updated 19 August 2026

What Inference Find was, and where its results came from

Inference Find was an American metasearch engine of the late 1990s. A metasearch engine has no index of its own — no crawler fetching pages, no stored copy of the web to rank. It takes your query, sends it to several real search engines at once, and assembles what comes back. Inference Find sent queries to Yahoo, AltaVista, Lycos, WebCrawler, Infoseek and Excite, merged the responses, stripped duplicates, and displayed a single list.

It is listed here as no longer available, and that is settled. Search Engine Watch reported on 7 March 2001 that the service, by then rebranded InFind, had shut down earlier that week; ABC News confirmed at the end of April 2001 that it had closed in March having run out of capital. A DNS lookup on 19 August 2026 returned no address for infind.com. Any directory still listing Inference Find or InFind as an available search tool has been wrong for a quarter of a century.

One thing it was not: the service did not run on artificial intelligence, whatever its name implies. Its parent, Inference Corporation, was a genuine AI software company — the ART expert-system toolkit, case-based reasoning, diagnostic software for Compaq — but no source connects any of that technology to the search product. The name is corporate heritage, not a description of the engine.

How the metasearch actually worked

The mechanics are documented in a contemporaneous survey of metasearch engines from around 1998, and several of the details are more interesting than the product's obscurity suggests.

Parallel dispatch, with redundancy. Queries went out to the six back-end engines simultaneously. Infoseek was called three times in parallel to pull 30 records — an unusually aggressive fan-out for the period, and a reminder that early metasearch was as much a scraping problem as a ranking one.
A user-controlled timeout. The searcher could set how long the engine would wait for slow back-ends, anywhere from 1 to 30 seconds. This exposed the fundamental trade-off of metasearch — coverage against latency — directly to the person doing the searching, which almost no modern engine does.
Merging and de-duplication. Results were combined and redundancies removed before display, with a title-only display mode for people who wanted density over detail.
Localisation and accessibility. French- and German-language interfaces were added, and a text-only gateway existed for accessibility use.

The service ran first at inference.com/infind/ and later at infind.com under the InFind brand.

Clustering: the thing it did first

The distinguishing feature was the results page. Instead of a single ranked list, Inference Find organised results into groups. Its own description of the process was that it "merges the results, removes redundancies, and clusters the results into neat understandable groupings". Search Engine Watch described the clustering as being by topic and by site type — commercial, educational, non-profit — to help users find authoritative material faster.

This matters because clustering is a better answer to an ambiguous query than ranking is. A flat list forces the engine to guess which sense of a word you meant and commit to it; a clustered page shows you the senses and lets you pick. Inference Find was doing this commercially several years before Vivisimo's Clusty and Grokker made it a category, and the industry has spent the two decades since only half-relearning the lesson.

Who ran it, and why it died

From around 1996 the service belonged to Inference Corporation, a US artificial-intelligence software firm founded in Los Angeles in 1979. Inference Find was a side project and a showcase, not a revenue line: the parent's money came from enterprise AI software. In February 2000 the search service was spun off as InFind, a separate business with Allan Kaplan as president. The following month, eGain Corporation agreed to acquire Inference Corporation for a reported $73 million — a deal that concerned the parent's enterprise software, not the search engine, which had already been separated out.

The spun-off business had no revenue model. It was notable for carrying no advertisements on its search pages, which its president treated as a point of principle and users clearly valued, but nothing replaced the advertising it declined to sell. Search Engine Watch reported insufficient advertisers and no alternative revenue model. Kaplan's own words to Search Engine Watch in March 2001 were: "Times have changed, and the environment has gotten tougher, so at this moment it's not clear what will happen with InFind." He told ABC News a month later that the site had simply "run out of capital".

Its usage figures make the ending sharper. By the operator's own account, reported by ABC News and not independently audited, InFind was serving more than 500,000 hits and 35,000 visitors a day with no marketing at all — "strictly based on word-of-mouth promotion from loyal users". Search Engine Watch put it more modestly: not among the most popular metasearch engines, but with a dedicated following that generated about five enquiries a day after it went dark. No third-party market-share measurement for it has ever been published.

An honest assessment

Two things were ahead of their time: clustered results, and giving the user an explicit dial on the coverage-versus-latency trade-off. Building 35,000 daily visitors purely by word of mouth, in an era when portals were spending enormous sums on advertising, is real evidence that people liked the product.

What it was bad at was everything commercial, and everything structural about metasearch. It refused ads, never built an alternative, was spun out of its parent into a funding market that closed behind it, and was dead within thirteen months. Technically it was hostage to six engines it did not control: it could not rank better than its sources, could not cover more than they covered, and could not survive their decline. Infoseek, Excite, AltaVista, WebCrawler and Lycos were themselves being dismantled by Google over exactly the same period. A metasearch engine whose sources are collapsing has no product.

The trademarks tell the end of the story tidily. The INFERENCE FIND mark, filed on 16 August 1996, was recorded abandoned in 2001 with no statement of use ever filed. The INFIND mark, filed in January 2000 with a broad portal-services specification that shows what the spin-off had hoped to become, was abandoned in October 2001 for failure to respond.

What is not known

This page is short because the record is thin, and the gaps are worth stating rather than filling.

The launch date is unknown. The earliest hard datum is the 16 August 1996 US trademark filing for INFERENCE FIND, which establishes only that the name was in use or intended for use by then. A contemporaneous survey describes a mature service, and a mirrored accessibility page carries a 1998 Inference Corporation copyright. None of that is a launch announcement.
There is no privacy documentation. No privacy policy or data-retention statement for either brand was located, and none would be expected from a service of that vintage. What can be said structurally is that a metasearch engine necessarily forwards its users' queries to every back-end it calls, and each of those logged them under its own terms; whether Inference Find stripped identifiers first is unknown. Its ad-free results pages did mean no third-party ad tracking, which in 1998 terms was meaningfully more private than the alternatives — but that is an inference from the absence of ads, not a claim the company made.
There is no Wikipedia article on Inference Find or InFind. The article on Inference Corporation covers the parent company and does not mention the search engine at all, so it does not support claims about the product.

Common misconceptions

"Inference Find had its own index." It did not. Every result came from Yahoo, AltaVista, Lycos, WebCrawler, Infoseek or Excite, queried live.
"It was an AI search engine, because Inference Corporation was an AI company." The parent was an AI company. No source connects its case-based-reasoning technology to the search product, and grouping results by type and topic was solved in the 1990s with heuristics and string analysis.
"Inference Find and InFind were different services." Same service. InFind was the rebranded name after the February 2000 spin-off from Inference Corporation.
"Google killed it." More precisely, it ran out of capital in a closed funding market in March 2001, during the dot-com collapse, having never built a revenue model. Google was the background condition, not the stated cause.
"eGain shut it down." eGain acquired Inference Corporation, but the search business had already been spun out as a separate company in February 2000 and failed on its own.
"Clustering search started with Clusty." Inference Find was doing it commercially several years earlier.

Frequently asked questions

Is Inference Find still online?

No. The service, by then rebranded InFind, shut down in early March 2001, and ABC News confirmed at the end of that April that it had closed having run out of capital. A DNS lookup on 19 August 2026 returned no address for infind.com. The unrelated domain inference.com resolves today but has nothing to do with the search engine.

Did Inference Find have its own index?

No. It was a metasearch engine, meaning it had no crawler and no stored copy of the web of its own. It sent each query to Yahoo, AltaVista, Lycos, WebCrawler, Infoseek and Excite in parallel, then merged the answers, removed duplicates and grouped them. Its only original contribution was the merging, clustering and presentation layer.

What made Inference Find different from other metasearch engines?

Two things. It clustered results into groups by topic and by site type, such as commercial, educational and non-profit, instead of showing one flat ranked list. And it let the user set the timeout for slow back-end engines, anywhere from 1 to 30 seconds, exposing the coverage-versus-speed trade-off directly rather than hiding it.

What was the difference between Inference Find and InFind?

They were the same service under two names. Inference Find ran as a project of Inference Corporation, an AI software company founded in Los Angeles in 1979, from around 1996. In February 2000 the search service was spun off as a separate business called InFind, with Allan Kaplan as president, running at infind.com until it closed thirteen months later.

Why did InFind shut down?

It had no revenue model and could not raise capital. The service carried no advertisements on its search pages, which users valued, but nothing was built to replace that income, and after the February 2000 spin-off it needed outside money just as the dot-com funding market closed. Its president said simply that it had run out of capital.

Was Inference Find an early AI search engine?

No, despite the name. Its parent company built expert systems and case-based reasoning software, but no source connects that technology to the search product, and no AI technique is documented inside it. Describing it as an early AI engine reads the parent's brand into a product that nothing links it to.

How popular was Inference Find?

No independent measurement exists. The operator told ABC News that shortly before closure the site was serving more than 500,000 hits and 35,000 visitors a day, achieved purely by word of mouth with no marketing spend. Those are unaudited operator figures. Search Engine Watch described it as not among the most popular metasearch engines but with a dedicated following.

Inference Find