Search Engine Bias and the Demise of Search Engine Utopianism

Date: Sep 22, 2006

Due to automated operations, people often mistakenly assume that search engines display search results neutrally and without bias. Like any other media company, however, search engines affirmatively control their users' experiences, which has the consequence of skewing search results (a phenomenon called "search engine bias"). Some commentators believe that search engine bias is a defect requiring legislative correction. But Eric Goldman argues that search engine bias is the beneficial consequence of search engines optimizing content for their users. This article further argues that the most problematic aspect of search engine bias, the "winner take all" effect caused by top placement in search results, will be made moot by emerging personalized search technology.

In the past few years, search engines have emerged as a major force in our information economy, helping searchers to perform hundreds of millions (or even billions) of searches per day. [1] With this broad reach, search engines have significant power to shape searcher behavior and perceptions. In turn, the choices that search engines make about how to collect and present data can have significant social implications.

Typically, search engines automate their core operations, including the processes that the search engines use to aggregate their databases and then sort/rank the data for presentation to searchers. This automation gives search engines a veneer of objectivity and credibility. [2] Machines, not humans, appear to make the crucial judgments, creating the impression that search engines bypass the structural biases and skewed data presentations inherent in any human-edited media. [3] Search engines’ marketing disclosures typically reinforce this perception of objectivity.

Unfortunately, this romanticized view of search engines doesn’t match reality. Search engines are media companies. Like other media companies, search engines make editorial choices designed to satisfy their audiences. [4] These choices systematically favor certain types of content over others, producing a phenomenon called search engine bias.

Search engine bias sounds scary, but this article explains why such bias is both necessary and desirable. I’ll also explain how emerging personalization technology will soon ameliorate many concerns about search engine bias.

Search Engines Make Editorial Choices

Search engines frequently claim that their core operations are completely automated and free from human intervention, [5] but this characterization is false. Instead, humans make numerous editorial judgments about what data to collect and how to present that data. [6]

Indexing

Search engines don’t index every scrap of data available on the Internet. Deliberately or accidentally, search engines omit some web pages entirely [7] or may incorporate only part of a web page. [8]

During indexing, search engines are designed to associate third-party metadata (data about data) with the indexed web page. For example, search engines may use and display third-party descriptions of the web site in the search results. [9] Search engines may also index anchor text (the text that third parties use in hyperlinking to a web site), [10] which can cause a web site to appear in search results for a term the web site never used (and to which the site owner may object). [11]

Finally, once indexed, search engines may choose to exclude web pages from their indexes for a variety of reasons, ranging from violations of quasi-objective search engine technical requirements [12] to simple capriciousness. [13]

Ranking

To determine the order of search results, search engines use complex proprietary ranking algorithms. Ranking algorithms obviate the need for humans to make individualized ranking decisions for the millions of search terms used by searchers, but such algorithms don’t lessen the role of human editorial judgment in the process. Instead, the choice of which factors to include in the ranking algorithm—and how to weight such factors—reflects the search engine operator’s editorial judgments about what makes content "valuable." Indeed, to ensure that these judgments produce the desired results, search engines manually inspect search results [14] and make adjustments accordingly.

Additionally, search engines claim that they don’t modify algorithmically generated search results, but there is some evidence to the contrary. Search engines allegedly make manual adjustments to some web publishers’ overall rankings. [15] Also, search engines occasionally modify search results presented in response to particular keyword searches. Consider the following instances:

Some search engines blocked certain search terms containing the keyword phpBB. [16]
In response to the search term Jew, for a period of time (including, at minimum, November 2005, when this author observed the phenomenon), Google displayed a special result in the sponsored link, saying "Offensive Search Results: We’re disturbed about these results as well. Please read our note here." The link led to a page explaining the results. [17]
Reportedly, Ask.com blocked search results for certain terms such as pedophile, bestiality, sex with children, and child sex. [18]
Google removed some web sites from its index in response to a 512(c)(3) take-down demand from the Church of Scientology. However, Google displayed the following legend at the bottom of affected search results pages (such as search results for scientology site:xenu.net): "In response to a complaint we received under the U.S. Digital Millennium Copyright Act, we have removed two result(s) from this page. If you wish, you may read the DMCA complaint that caused the removal(s) at ChillingEffects.org." [19]

Conclusion

Search engines have some duality in their self-perceptions, and this duality creates a lot of confusion. [20] Search engines perceive themselves as objective and neutral because they let automated technology do most of the hard work. However, in practice, search engines make editorial judgments just like any other media company. Principally, these editorial judgments are instantiated in the parameters set for the automated operations, but search engines also make individualized judgments about what data to collect and how to present it. These manual interventions may be the exception and not the rule, but these exceptions only reinforce the fact that search engines play an active role in shaping their users’ experiences when necessary to accomplish their editorial goals.

Search Engine Editorial Choices Create Biases

Search results ordering has a significant effect on searchers and web publishers. Searchers usually consider only the top few search results; the top-ranked search result gets a high percentage of searcher clicks, and click-through rates quickly decline from there. [21] Therefore, though a search engine may deliver hundreds or even thousands of search results in response to a searcher’s query, searchers effectively ignore the vast majority of those search results. Accordingly, web publishers desperately want to be listed among the top few search results. [22]

For search engines, results placement determines how the searcher perceives the search experience. If the top few search results don’t satisfy the searcher’s objectives, the searcher may deem the search a failure. Therefore, to maximize searcher perceptions of search success, search engines generally tune their ranking algorithms to support majority interests. [23] In turn, minority interests (and the web sites catering to them) often receive marginal exposure in search results.

To gauge majority interests, search engines frequently include a popularity metric in their ranking algorithm. Google’s popularity metric, PageRank, treats inbound links to a web site as popularity votes, but votes are not counted equally; links from more-popular web sites count more than links from lesser-known web sites. [24]

Beyond promoting search results designed to satisfy majority interests, PageRank’s non-egalitarian voting structure causes search results to be biased toward web sites with economic power [25] because these web sites get lots of links due to their marketing expenditures and general prominence.

Indeed, popularity-based ranking algorithms may reinforce and perpetuate existing power structures. [26] Web sites that are part of the current power elite get better search result placement, which leads to greater consideration of their messages and views. Furthermore, the increased exposure attributable to better placement means that these web sites are likely to get more votes in the future, leading to a self-reinforcing process. [27] In contrast, minority-interest and disenfranchised web sites may have a difficult time cracking through the popularity contest, potentially leaving them perpetually relegated to the search results hinterlands. [28]

A number of commentators have lamented these effects and offered some proposals in response:

In addition to plaintiffs, some academics have supported mandatory reordering of search results. For example, Pandey et al. advocate a "randomized rank promotion" scheme in which obscure web sites should get "extra credit" randomly in ranking algorithms, appearing higher in the search results on occasion and accordingly getting additional exposure to searchers. [35] As another example, Pasquale proposes that, when people think the search engines are providing false or misleading information, search engines should be forced to include a link to corrective information. [36]

Search Engine Bias Is Necessary and Desirable

Before trying to solve the "problem" of search engine bias, we should be clear how search engine bias creates a problem that requires correction. From my perspective, search engine bias is the unavoidable consequence of search engines exercising editorial control over their databases. Like any other media company, search engines simply cannot passively and neutrally redistribute third-party content (in this case, web publisher content). If a search engine doesn’t attempt to organize web content, its system quickly and inevitably will be overtaken by spammers, fraudsters, and malcontents. [37] At that point, the search engine become worthless to searchers.

Instead, searchers (like other media consumers) expect search engines to create order from the information glut. To prevent anarchy and preserve credibility, search engines unavoidably must exercise some editorial control over their systems. In turn, this editorial control necessarily will create some bias.

First, searchers have high expectations for search engines: They expect search engines to read their minds [39] and infer their intent based solely on a small number of search keywords. [40] Search engines that disappoint (either by failing to deliver relevant results, or by burying relevant results under too many unhelpful results) are held accountable by fickle searchers. [41] There are multiple search engines available to searchers, [42] and few barriers to switching between them. [43]

As a result, searchers will shop around if they don’t get the results they want, [44] and this competitive pressure constrains search engine bias. If a search engine’s bias degrades the relevancy of search results, searchers will explore alternatives—and that’s true even if the searchers don’t realize that the results are biased. Meanwhile, search engine proliferation means that niche search engines can segment the market and cater to underserved minority interests. [45] Admittedly, these market forces are incomplete—searchers may never consider what results they’re not seeing—but such forces are powerful nonetheless.

In contrast, it’s difficult to imagine how regulatory intervention will improve the situation. First, regulatory solutions become a vehicle for normative views about what searchers should see—or should want to see. [46] How should we select among these normative views? What makes one bias better than another?

Second, regulatory intervention that promotes some search results over others doesn’t ensure that searchers will find the promoted search results useful. Determining relevancy based on very limited data (such as decontextualized keywords) is a challenging process, and search engines struggle with this challenge daily. Due to the complexity of the relevancy-matching process, government regulation rarely can do better than market forces at delivering results that searchers find relevant. As a result, searchers likely will find some of the promoted results irrelevant.

The clutter of unhelpful result may hinder searchers’ ability to satisfy their search objectives, undermining the searchers’ confidence in search engines’ mind-reading abilities. [47] In this case, regulatory intervention could counterproductively degrade the value of search engines to searchers. Whatever the adverse consequences of search engine bias, the consequences of regulatory correction are probably worse. [48]

Technological Evolution Will Make Search Engine Bias Moot

Currently, search engines principally use "one size fits all" ranking algorithms to deliver homogeneous search results to searchers with heterogeneous search objectives. [49] One-size-fits-all algorithms exacerbate the consequences of search engine bias in two ways:

These consequences will abate when search engines migrate away from one-size-fits-all algorithms toward "personalized" ranking algorithms. [51] Personalized algorithms produce search results that are custom-tailored to each searcher’s interests, so different searchers will see different results in response to the same search query. For example, Google offers searchers an option that "orders your search results based on your past searches, as well as the search results and news headlines you’ve clicked on." [52]

Personalized ranking algorithms represent the next major advance in search relevancy. One-size-fits-all ranking algorithms have inherent limits on their maximum relevancy potential, and further improvements in one-size-fits-all algorithms will yield progressively smaller relevancy benefits. Personalized algorithms transcend those limits, optimizing relevancy for each searcher and thus implicitly doing a better job of searcher mind-reading. [53]

Personalized ranking algorithms also reduce the effects of search engine bias. Personalized algorithms mean that there are multiple "top" search results for a particular search term, instead of a single "winner," [54] so web publishers won’t compete against each other in a zero-sum game. In turn, searchers will get results more influenced by their idiosyncratic preferences and less influenced by the embedded preferences of the algorithm-writers. Also, personalized algorithms necessarily will diminish the weight given to popularity-based metrics (to give more weight for searcher-specific factors), reducing the structural biases due to popularity.

Personalized ranking algorithms are not a panacea—any process in which humans select and weight algorithmic factors will produce some bias, [55] but personalized algorithms will eliminate many of the current concerns about search engine bias.

Conclusion

Complaints about search engine bias implicitly reflect some disappointed expectations. In theory, search engines can transcend the deficiencies of predecessor media to produce a type of utopian media. In practice, search engines are just like every other medium—heavily reliant on editorial control and susceptible to human biases. This fact shatters any illusions of search engine utopianism.

Fortunately, search engine bias may be largely temporal. In this respect, I see strong parallels between search engine bias and the late 1990s keyword metatag "problem." [56] Web publishers used keyword metatags to distort search results, but these techniques worked only so long as search engines considered keyword metatags in their ranking algorithms. When search engines recognized the distortive effects of keyword metatags, they changed their algorithms to ignore keyword metatags. [57] Search result relevancy improved, and the problem was solved without regulatory intervention.

Similarly, search engines naturally will continue to evolve their ranking algorithms and improve search result relevancy—a process that, organically, will cause the most problematic aspects of search engine bias to largely disappear. To avoid undercutting search engines’ quest for relevance, this effort should proceed without regulatory distortion.

References

[1] In 2003, search engines performed more than a half-billion searches a day. See Danny Sullivan, "Searches Per Day," Search Engine Watch, Feb. 25, 2003.

[3] There is a broad perception that search engines present search results passively and neutrally. See Leslie Marable, "False Oracles: Consumer Reaction to Learning the Truth About How Search Engines Work," Consumer Reports WebWatch, June 30, 2003; Maureen O’Rourke, "Defining the Limits of Free-Riding in Cyberspace: Trademark Liability for Metatagging," 33 Gonzaga Law Review 277 (1998).

[4] See, e.g., C. Edwin Baker, Advertising and a Democratic Press (Princeton University Press, 1994).

[5] See, e.g., "Does Google Ever Manipulate Its Search Results?" Google.com ("The order and contents of Google search results are completely automated. No one hand picks a particular result for a given search query, nor does Google ever insert jokes or send messages by changing the order of results."); "Does Google Censor Search Results?" Google.com ("Google does not censor results for any search terms. The order and content of our results are completely automated; we do not manipulate our search results by hand."); "Technology Overview," Google.com ("There is no human involvement or manipulation of results...."); "How Can I Improve My Site’s Ranking?" Google.com ("Sites’ positions in our search results are determined automatically based on a number of factors, which are explained in more detail at http://www.google.com/technology/index.html. We don’t manually assign keywords to sites, nor do we manipulate the ranking of any site in our search results.") See also Complaint at ¶¶ 37–38, 52–56, KinderStart.com LLC v. Google, Inc., Case No. C 06-2057 RS (N.D. Cal. Mar. 17, 2006), giving other examples of Google’s claims to be passive. Note that Google has subsequently revised some of these cited pages after its censorship controversy in China.

[6] See generally Abbe Mowshowitz and Akira Kawaguchi, "Bias on the Web," Comm. ACM, Sept. 2002, at 56 (distinguishing "indexical bias" and "content bias").

[8] For example, many search engines ignore metatags. See Eric Goldman, "Deregulating Relevancy in Internet Trademark Law," 54 Emory Law Journal 507, 567–68 (2005). Search engines also incorporate only portions of very large files. See Bar-Ilan, supra note [7]; "Why Doesn’t My Site Have a Cached Copy or a Description?" Google.com (describing how some pages are "partially indexed"); "Has Google Dropped Their 101K Cache Limit?" ResearchBuzz!, Jan. 31, 2005 (discussing how historically Google indexed only the first 101KB of a document).

[9] See "My Site’s Listing Is Incorrect and I Need it Changed," Google.com. Google’s automated descriptions have spawned at least one lawsuit by a web publisher who believed the compilation created a false characterization. See Seth Fineberg, "Calif. CPA Sues Google Over ’Misleading’ Search Results," Acct. Today, Apr. 19, 2004, at 5, available at WebCPA.

[11] For example, the first search result in Google and Yahoo! for the keyword "miserable failure" is President George W. Bush’s home page because so many web sites have linked to the biography using the term "miserable failure." See Tom McNichol, "Your Message Here," N.Y. Times, Jan. 22, 2004, at G1. This algorithmic vulnerability has spawned a phenomenon called "Google bombing," in which web sites coordinate an anchor text attack to distort search results intentionally. See John Hiler, "Google Time Bomb," Microcontent News, Mar. 3, 2002.

[12] See, e.g., Stefanie Olsen, "Search Engines Delete Adware Company," CNET News.com, May 13, 2004 (Google and Yahoo kicked WhenU.com out of their indexes for allegedly displaying different web pages to searchers and search engine robots, a process called "cloaking").

[13] This is the heart of KinderStart’s allegations against Google. See Complaint, KinderStart.com LLC v. Google, Inc., Case No. C 06-2057 (N.D. Cal. Mar. 17, 2006). Although the complaint’s allegations about Google’s core algorithmic search may not be proven, Google does liberally excise sources from Google News. For example, Google claims that "news sources are selected without regard to political viewpoint or ideology," see "Google News," Google.com, but Google dropped a white supremacist news source from Google News because it allegedly promulgated "hate content." See Susan Kuchinskas, "Google Axes Hate News," Internetnews.com, Mar. 23, 2005.

[14] See Eric Goldman, "Google’s Human Algorithm," Technology & Marketing Law Blog, June 5, 2005 (Google hires students to review search results manually for quality purposes).

[15] See Search King, Inc. v. Google Tech., Inc., No. CIV-02-1457-M, at 4 (W.D. Okla. Jan. 13, 2003). ("Google knowingly and intentionally decreased the PageRanks assigned to both SearchKing and PRAN.") This manual adjustment has also been alleged in the recent KinderStart lawsuit. See Complaint, KinderStart.com L.L.C. v. Google, Inc., Case No. C 06-2057 RS (N.D. Cal. Mar. 17, 2006).

[19] Google scientology.site:xenu.net and scroll to the bottom of the page (or click here and scroll to the bottom of the page). See also Chris Sherman, "Google Makes Scientology Infringement Demand Public," Search Engine Watch, Apr. 15, 2002.

This duality, if it ends up leading to the dissemination of false information, could also create some legal liability. See KinderStart v. Google, No. 5:06-cv-02057-JF (N.D. Cal. motion to dismiss granted July 13, 2006), pointing out the potential inconsistency of Google’s position that PageRank is Google’s subjective opinion but an objective reflection of its algorithmic determinations.

[21] See "iProspect Search Engine User Behavior Study," iProspect, Apr. 2006 (62% of searchers click a search result on the first results page); Jakob Nielsen, "The Power of Defaults," Jakob Nielsen’s Alertbox, Sept. 26, 2005 (citing a study by Cornell professor Thorsten Joachims that the first search result gets 42% of clicks and the second search result gets 8%; further, when the first two search results are switched, the first search result gets 34%—meaning that positioning dictated searcher behavior); Nico Brooks, "The Atlas Rank Report: How Search Engine Rank Impacts Traffic," Atlas Institute Digital Marketing Insights, June 2004 (the first-ranked search result may get ten times as many clicks as the tenth-ranked search result).

[23] See Lucas D. Introna and Helen Nissenbaum, "Shaping the Web: Why the Politics of Search Engines Matters," Info. Society, July–Sept. 2000, at 169.

[25] See Niva Elkin-Koren, "Let the Crawlers Crawl: On Virtual Gatekeepers and the Right To Exclude Indexing," 26 University of Dayton Law Review 179, 188 (2001); Frank Pasquale, "Rankings, Reductionism, and Responsibility," Seton Hall Public Law Research Paper No. 888327, at 25, Feb. 25, 2006; Trystan Upstill et al., "Predicting Fame and Fortune: PageRank or Indegree?" Proceedings of the 8th Australasian Document Computing Symposium, Dec. 15, 2003 (showing that BusinessWeek Top Brand, Fortune 500, and Fortune Most Admired companies get disproportionately high PageRank).

[27] See "Egalitarian Engines," Economist, Nov. 17, 2005 ("[T]here is a widespread belief among computer, social, and political scientists that search engines create a vicious circle that amplifies the dominance of established and already popular web sites"); see also Junghoo Cho and Sourashis Roy, "Impact of Search Engines on Page Popularity," WWW 2004, May 2004; Upstill, supra note [25]. But see Santo Fortunato et al., "The Egalitarian Effect of Search Engines," Nov. 2005 (questioning the consequences of the "rich-gets-richer" effect).

[28] See Cho and Roy, supra note [27]; but see Filippo Menczer et al., "Googlearchy or Googlocracy?" IEEE Spectrum, Feb. 2006 (providing empirical evidence suggesting that "search engines direct more traffic than expected to less-popular sites").

[29] See Search King Inc. v. Google Tech., Inc., No. CIV-02-1457-M, at 3 n.2 (W.D. Okla. Jan. 13, 2003): "Google’s mathematical algorithm is a trade secret, and it has been characterized by the company as ’one of Google’s most valuable assets’"; Stefanie Olsen, "Project Searches for Open-Source Niche," CNET News.com, Aug. 27, 2003.

[31] See id.; Eszter Hargittai, "Open Portals or Closed Gates? Channeling Content on the World Wide Web," 27 Poetics 233 (2000); cf. "Cass Sunstein, Republic.com" 170-72 (2001), advocating publicly funded "deliberative domains."

[32] See Kevin J. O’Brien, "Europeans Weigh Plan on Google Challenge," International Herald Tribune, Jan. 18, 2006 (discussing a European initiative called Quaero, which is intended to break the American hegemony implicit in Google’s dominant market position); Graeme Wearden, "Japan May Create Its Own Search Engine," ZDNet News, Dec. 21, 2005.

[33] See Search King, Inc. v. Google Tech., Inc., No. CIV-02-1457-M (W.D. Okla. Jan. 13, 2003); KinderStart.com LLC v. Google, Inc., No. C 06-2057 RS (N.D. Cal. dismissed July 13, 2006); Langdon v. Google, Inc., No. 1:06-cv-00319-JJF (D. Del. complaint filed May 17, 2006); Roberts v. Google, No. 1-06-CV-063047 (Cal. Superior Ct. complaint filed May 5, 2006); Datner v. Yahoo! Inc, Case No. BC355217 (Cal. Superior Ct. complaint filed July 11, 2006). [Note: This list updated as of July 24, 2006.]

[34] As Google said in its response to the KinderStart lawsuit, "Plaintiff KinderStart contends that the judiciary should have the final say over [search engines’] editorial process. It has brought this litigation in the hopes that the Court will second-guess Google’s search rankings and order Google to view KinderStart’s site more favorably." Motion to Dismiss at 1, KinderStart.com LLC v. Google, Inc., No. C 06-2057 RS (N.D. Cal. May 2, 2006).

[36] See Pasquale, supra note [25], at 28–30 (proposing that the link be displayed as an asterisk to the search results).

[37] Every Internet venue accepting user-submitted content inevitably gets attacked by unwanted content. If left untended, the venue inexorably degrades into anarchy. See, e.g., "Step-by-Step: How to Get BILLIONS of Pages Indexed by Google," Monetize Blog, June 17, 2006, (Google indexed more than five billion "spam" pages from a single spammer before manually de-indexing the sites); Alorie Gilbert, "Google Fixes Glitch that Unleashed Flood of Porn," CNET News.com, Nov. 28, 2005 (describing how Google Base, a venue for user-submitted content, was overtaken by pornographers: "[T]he amount of adult content on Google Base was staggering considering Google only launched the tool a week ago"); Josh Quittner, "The War Between alt.tasteless and rec.pets.cats," Wired, May 1994, at 46 (describing how a group of anarchists, for fun, took over a USENET newsgroup about pets).

[38] See Mowshowitz and Kawaguchi, supra note [6], at 60 (market forces are the best way to counter adverse effects of search engine bias).

[39] See "Our Philosophy," Google.com ("The perfect search engine...would understand exactly what you mean and give back exactly what you want"); Chris Sherman, "If Search Engines Could Read Your Mind," Search Engine Watch, May 11, 2005.

[40] Searchers routinely use a very small number of keywords to express their search interests. See iProspect.com, Inc., "iProspect Natural SEO Keyword Length Study," Nov. 2004 (stating that 88% percent of search engine referrals are based on only one or two keywords); see also Declan Butler, "Souped-Up Search Engines," Nature, May 11, 2000, at 112, 115 (citing an NEC Research Institute study showing that up to 70% of searchers use only a single keyword as a search term); Bernard J. Jansen et al., "Real Life Information Retrieval: A Study of User Queries on the Web," 32 SIGIR Forum 5, 15 (1998), stating that the average keyword length was 2.35 words; one-third of searches used one keyword and 80% used three keywords or fewer; Jakob Nielsen, "Search: Visible and Simple," Jakob Nielsen’s Alertbox, May 13, 2001 (stating that the average keyword length was 2.0 words).

[41] See Kim Peterson, "Microsoft Learns to Crawl," Seattle Times, May 2, 2005 (MSN Search "learned that the arcane searches were the make-or-break moments for web searchers. People weren’t just happy when a search engine could find answers to their most bizarre, obscure, and difficult queries. They would switch loyalties."); Bob Tedeschi, "Every Click You Make, They’ll Be Watching You," N.Y. Times, Apr. 3, 2006.

[42] In addition to the recent launch of major new search engines by providers such as MSN, the open-source software community is developing Nutch to allow anyone to build and customize his or her own web search engine; see also Olsen, "Project Searches for Open-Source Niche," supra note [29].

While there are multiple major search engines, the market may still resemble an oligopoly; a few major players (Google, Yahoo!, MSN, Ask Jeeves) have the lion’s share of the search engine market. However, this observation may construe the search engine market too narrowly. Many types of search providers compete with the big mass-market search engines, ranging from specialty search engines (such as Technorati) to alternative types of search technology (adware, for instance) to non-search information retrieval processes (link navigation and such). Ultimately, every search engine competes against other search engines and these other search/retrieval options.

[43] See Rahul Telang et al., "An Empirical Analysis of Internet Search Engine Choice," Aug. 2002. On file with author. For example, search engines use the same basic interface (a white search box), and searchers rarely use advanced search features that might require additional learning time at other search engines.

[44] See Grant Crowell, "Understanding Searcher Behavior," Search Engine Watch, June 14, 2006 (citing a Kelsey Research study showing that 63% of searchers used two or more search engines); Vividence, Inc., "Google Wins Users’ Hearts, But Not Their Ad Clicks," press release, May 25, 2004 (stating that up to 47% of searchers try another search engine when their search expectations are not met).

[45] See Rahul Telang et al., "The Market Structure for Internet Search Engines," 21 J. Mgmt. Info. Sys. 137 (2004), describing how searchers sample heterogeneous ranking algorithms, which support a diversity of search engines; Mário J. Silva, "The Case for a Portuguese Web Search Engine," (describing the value of a Portuguese-oriented search engine); Jeffrey McMurray, "Social Search Promises Better Intelligence," Associated Press, July 9, 2006 (discussing niche search engines that draw on social networking); cf. Jakob Nielsen, "Diversity Is Power for Specialized Sites," Jakob Nielsen’s Alertbox, June 16, 2003 (describing how specialized sites will flourish on the Internet).

[46] See, e.g., Susan L. Gerhart, "Do Web Search Engines Suppress Controversy?" First Monday, Jan. 2004. Gerhart argues that search engines don’t adequately prioritize search results that expose controversies about the search topic. However, her argument assumes that controversy-related information has value to consumers, an assumption that deserves careful evaluation.

[47] See Eric Goldman, "A Coasean Analysis of Marketing," 2006 Wisconsin Law Review (forthcoming).

[48] See Susan P. Crawford, "Shortness of Vision: Regulatory Ambition in the Digital Age," 74 Fordham Law Review 695 (2005), discussing the shortcomings of regulatory intervention in organic information systems.

[49] See James Pitkow et al., "Personalized Search," Comm. ACM, Vol. 45:9 (Sept. 2002) at 50–51.

[50] See Michael Kanellos, "Microsoft Aims for Search on Its Own Terms," CNET News.com, Nov. 24, 2003 (quoting a Microsoft researcher as saying, "If the two of us type a query [into a search engine], we get the same thing back, and that is just brain dead. There is no way an intelligent human being would tell us the same thing about the same topic"); David H. Freedman, "Why Privacy Won’t Matter," Newsweek, Apr. 3, 2006; "Personalization of Placed Content Ordering in Search Results," U.S. Patent App. 0050240580 (filed July 13, 2004).

[53] See Jaime Teevan et al., "Personalizing Search via Automated Analysis of Interests and Activities," SIGIR ’05; Terry McCarthy, "On the Frontier of Search," Time, Aug. 28, 2005: "Search will ultimately be as good as having 1,000 human experts who know your tastes scanning billions of documents within a split second" (quoting Gary Flake, Microsoft Distinguished Engineer).

[55] Personalized algorithms have other potentially adverse consequences, such as creating self-reinforcing information flows. See Sunstein, supra note [31]. For a critique of these consequences, see Goldman, "Coasean Analysis," supra note [47].

[57] See Danny Sullivan, "Death of a Meta Tag," Search Engine Watch, Oct. 1, 2002.