First Cause Automated SEO - Continued

For our purposes, the key is that this measure is related to the overall image of an online retailer and its attributes, which includes factors like name recognition, product breadth and depth, shopping experience, and reputation show that the utility of this measure is invariant to whether one uses a narrow definition of a navigational search or this broader measure that includes both the retailer’s name and productrelated . 10 Finally, we analyzed each search term and constructed search-term specific variables based on the content of the search term. The first variable is the number of words in the search term. The second variable, denoted branded search term, is an indicator variable for whether the search terms include the brand name of a product (e.g., Nike or Adidas) in the product search. Note that, in our sample, this is different from the brand associated with a particular retailer’s site (e.g., Zappos or Amazon). These two search-term specific variables may tell us something about the intent of search. For instance, an individual searching for “Nike running shoes” is more specific in what she is looking for than someone searching for “shoes online,” and this may affect clicking behavior. 2.2 Econometric Model Our main objective is to study the drivers of organic clicks arising from searches for products on search engines.14 Let  denote the total number of organic clicks retailer  received from individuals searching for search term . Because of the presence of substantial positive skewness in organic clicks data, we use a log-normal regression model to analyze the relationship between organic clicks and the explanatory variables, where  (short for rank not observed) is a dummy variable that equals 1 if retailer  is not observed on the first five pages of search results for search term ,  is the rank (or position) of retailer  on the first five pages of search results for search term ,  is a dummy for whether the retailer had a sponsored link on the first results page for search term ,  a measure of retailer ’s brand equity, and  is a vector of other other controls including demographic variables, search term specific variables, retailer characteristics as well as retail segment fixed effects.15 There are two primary concerns with estimating this equation: (i) it is likely that some of the explanatory variables are endogenous (correlated with  ); and (ii) owing to the nature of the 1 4As a referee notes, clicks for some search terms may be more valuable (in terms of conversions, margins, or other performance measures) than others. Our focus here is on clicks; unfortunately, our data do not permit us to examine the differential effects of rank and other variables on these alternative measures of retailer performance. 1 5Retail segment fixed effects control for systematic differences in clicks across, for example, mass merchants who may receive many clicks owing to product breadth effects and specialty retailers who receive fewer clicks. While this specification assumes the marginal impact of brand equity and rank is identical across retail segments, we show in Section 4 that the results are similar when one excludes mass merchants such as Amazon and Walmart. 11 Search Planner data, we only observe the dependent variable in equation (1) when clicks exceed a certain threshold. Below we discuss how we deal with these concerns. 2.2.1 Endogeneity Google continuously updates its rankings of search results to generate the most relevant search results, which means that our rank variable will depend on past clicks. It is therefore likely that rank is correlated with the error term and thus endogenous. A similar effect may be at work for the ads variable: Ad positions are based on the outcome of a second-price auction that takes the relevance of the bidder with respect to the search term into account, again making it likely that ad positions are based on past clicking behavior on Google. The standard approach in the literature on clicks at platforms (e.g., clicks at price comparison sites or sponsored clicks at search engines) is to assume that such positions are exogenous. Using the Wu-Hausman test for endogeneity, however, we reject the hypothesis that rank and ad positions are exogenous in our data. To account for the potential endogeneity of these variables, we use information about rank and ads on Bing as instruments. These instruments are correlated with the endogenous regressors, but are unlikely to be correlated with the error term, since Bing’s decisions on search result rankings and ad positions are not based on past clicks on Google. Indeed, using the Sargan test for overidentifying restrictions, we cannot reject the hypothesis that these are valid instruments. One might also worry that our measure of brand equity is correlated with the error in equation (1). For example, if users begin searching for products with generic search phrases and end their searches with a navigational search, navigational searches may be driven by organic search results, and thus endogenous. Based on the Wu-Hausman test, however, we cannot reject the hypothesis that our measure of brand equity is exogenous, even at high significance levels. Our main results thus treat only position and ads as endogenous. Section 4 shows that our results are robust to the use of three alternative measures of brand equity that are also unlikely to be correlated with the errors. 2.2.2 Sample Selection As we explained in Section 2.1, a retail site is included as an observation if it appears on the first five pages of the Google search result page for a specific search term, independent of whether the retailer received organic clicks according to Search Planner. Complicating matters, Search Planner 12 only reports the number of organic clicks if those clicks exceed a certain threshold, which means we do not know whether sites receiving zero organic clicks according to Search Planner really received no click-throughs for the search term in question or whether they were censored. What makes our setting different from a standard censoring environment is that the selection rule depends on total clicks (including paid clicks) rather than just organic clicks. This means that a different probability mechanism generates both the zero clicks and the positive clicks, and this cannot be captured by a standard Tobit censoring model. For this reason, we estimate a Heckmantype selection model. As we argued in the previous subsection, endogeneity is likely to be important in our data, so we allow for endogenous explanatory variables. Estimation of the model consists of two stages. In the first stage we regress a dummy for having positive clicks on all exogenous variables (including instruments) . Here, it is important to include at least one more instrument than is necessary for dealing with the endogeneity problem (otherwise identification is purely based on the parametric form of the inverse Mills ratio). This additional exclusion restriction should relate to the probability of observing positive organic clicks. Since this probability relates to the number of total clicks, we use additional variables in the selection equation that are important for getting paid clicks: We add dummies for whether a sponsored link was displayed on each of pages 2 through 5 in the Bing search results. We obtain the inverse Mills ratio, given by …from the selection equation, and add this to the second stage to obtain. We estimate this equation using the selected subsample (for which we observe organic clicks), i.e., by two-stage least squares using instruments ( ˆ) for the endogenous variables ,  and . 3 Results Table 2 provides results for the specification in equation (2), which regresses the logarithm of organic clicks on explanatory variables that account for the impact on clicks of rank, brand equity, retailer characteristics, as well as characteristics including searcher demographics and the nature of search terms. Recall that these results control for potential endogeneity as well as censoring, and include a constant and retail segment fixed effects to account for potential differences in clicks 13 across the 15 retail segments identified in Table 1. All statistical tests are based on the reported robust standard errors, which account for potential…. The estimated coefficient for the inverse Mills ratio is significantly different from zero at the one percent level, which indicates that it is indeed appropriate to control for censoring of the data. We discuss the other estimated parameters of the model below. 3.1 Rank As discussed earlier, one potential goal of automated seo is to increase the ranking (or position) of a retailer’s links in organic search results. But just how important is position in driving a retailer’s organic clicks following a product search? The first two estimated coefficients in Table 2 provide an answer?

 The estimated coefficient for    captures the effect of a retailer’s link not being included on the first five pages of search results for a given search term. The estimated coefficient of −2335 is significant in both an economic and statistical sense, and implies that a firm not appearing on the first five pages receives 90 percent fewer clicks for a given search term. For a retailer whose link is observed on the first 5 pages, the estimated coefficient of −1347 for () implies that a 1 percent decline in rank induces a 13 percent reduction in organic clicks for a given search term. For example, a retailer moving from the fifth to the sixth position in a search for “jeans” experiences a 27 percent reduction in organic clicks for that search term, while moving from the sixth to the seventh position results in a 22 percent decline.16 While these results indicate that rank is a very important driver of organic clicks following product searches, it is important to stress that the unit of observation underlying these results is ; that is, retailer ’s position in the results for search term  Thus, these rank coefficients measure the effect of improving a retailer’s position for a single search term. Consequently, automated seo efforts that are term specific (e.g., designed to elevate a retailer’s rank following a search for “jeans” but that have no effect on positions following other product searches), will result in a much smaller percentage improvement in that retailer’s total organic clicks. By way of example, the average retailer in our sample was relevant for about 60 search terms, so the corresponding effect on total organic clicks is about 160th of the rank coefficients in Table 2. For example, the estimated coefficient of −1347 for () implies that a 1 percent improvement in rank following a given keyword search results in a 002 (= 134760) percent increase in total organic clicks. 1 6We also ran specifications with position bins rather than the log-linear specification and the results were qualitatively similar. 14 These results indicate that the returns to term-specific automated seo critically depend on the breadth and depth of a retailer’s product offerings and hence the number of search terms in which its link is relevant. 3.2 Retailer Brand Equity Table 2 also reports estimates of the direct effect of a retailer’s brand equity on the clicks it receives following a product search. The estimated coefficient for the logarithm of brand equity is positive and very precisely estimated, indicating that the direct effect of brand equity of a retailer’s site is an important determinant of the organic clicks it receives following a product search. It is important to note that, unlike rank, brand equity is not search-term specific. As such, the estimated impact of brand equity in Table 2 captures the impact on a retailer’s total organic clicks: Holding rank and the other factors influencing clicks constant, a one percent increase in a site’s brand equity results in a 0.084 percent increase in a retailer’s total organic clicks. These results, coupled with those discussed above for rank, indicate that a marginal improvement in a retailer’s brand equity has a larger direct effect on its organic clicks than automated seo efforts resulting in a marginal improvement in its position associated with a particular search term. Notice that the estimated coefficient for brand equity in Table 2, which corresponds to 4 in equation (2), measures the direct effect of retailer ’s brand equity on its organic clicks. However, because search engines’ algorithms determine rankings or positions of listings, in part, on past clicking behavior, there is also an indirect effect of brand equity on clicks: Retailers with greater brand equity and stronger brand names enjoy more clicks, which results in better future ranks. Figure 2, which graphs the average number of times retail sites appear on the first page of search results for different sextiles of brand equity, shows that online retailers with stronger brands in our data tend to have better ranks on results pages following non-navigational searches. The total effect of brand equity on clicks includes the direct effect identified in Table 2 as well as the indirect effect through rank. We identify the total effect of brand equity on clicks by using a standard two-step procedure (Alwin and Hauser, 1975). In the first step we regress the rank variables on the logarithm of brand equity to obtain brand-equity adjusted ranks to determine the impact of brand equity on rank. In the second step, we proceed as in equation (2) but using these brand equity adjusted ranks. This regression yields an estimate of the total effect of brand equity on organic clicks, including both the direct effect shown in Table 2 as well as the indirect effect stemming from the impact of brand 15 equity on position. As shown in Figure 3, the indirect effect of brand equity on clicks (through its impact on rank) is slightly larger than the direct effect, resulting in a total effect on organic clicks of 0185–roughly twice the direct effect.17 From the standpoint of automated seo, these results highlight an important interaction between brand equity and rank. A one percent improvement in a retailer’s brand equity directly increases its total organic clicks by 0084 percent, owing to the fact that consumers more frequently click on its link instead of a competing one in the list of search results. Ultimately, this induces search engines to elevate the firm’s position in all relevant searches, which results in an additional 0101 percent increase in clicks. The total effect of a one percent improvement in a retailer’s brand equity is therefore a 0185 percent increase in total organic clicks. Unlike the impact of rank, this percentage increase applies to a retailer’s overall clicks rather than the clicks stemming from a single search phrase. To illustrate, suppose a relatively unknown retailer gets 1,000 organic clicks from navigational searches and 30,000 through non-navigational searches.18 If the retailer makes an investment in brand equity sufficient to increase its navigational searches by 10 clicks (1 percent), it gains an additional 55 (= 30 000 × 0185%) organic clicks from non-navigational searches. 3.3 Consumer Characteristics The specification in Table 2 indicates that consumer characteristics systematically influence organic clicks on search engines. These results are of potential interest to retailers engaging in automated seo to attract customers within particular demographic groups. Notice that all of the income categories are statistically significant at the 5 percent or better levels: Consumers with higher incomes tend to more frequently click an organic link following a product search than do consumers with lower incomes. While not all of the age categories are statistically significant, the general pattern suggests that younger individuals are less likely to click organic links than older individuals. Interestingly, the results also indicate that consumers searching from home are less likely to click following an organic search than individuals conducting a product search from the workplace. These patterns may stem from differences in search behavior 1 7Note that this procedure does not impact any of the other parameter estimates in Table 2, nor does it impact the overall fit of the model. The estimated total effect is significant at the 1 percent level; the robust standard error for the point estimate of 0185 is 0047. 1 8To put this hypothetical in perspective, electricgeneratorsdirect.com received 1,208 organic clicks from navigational searches and 31,955 organic clicks through non-navigational searches in August of 2012. 16 across consumers with different demographic characteristics. For example, consumers with greater incomes may be more likely to conduct product searches on platforms such as Amazon rather than a search engine; individuals with lower incomes may be more likely to search using price comparison sites. 3.4 Keyword-Specific Effects One might worry that the demographic effects documented above stem from differences in the sophistication of searchers with different demographic characteristics. To account for this, we include two controls for the nature of the keyword search: (1) branded search term, which is an indicator for whether the search phrase includes a brand-name product (e.g., “Levi’s Jeans”), and (2) number of words, which is simply a count of the number of words included in the search. The results indicate that searchers who include specific brands of products in their terms, or who use fewer words in their search, are more likely to click an organic link following a product search. These findings are consistent with Yang and Ghose (2010), who find a positive relationship between branded searches and paid clicks as well as a positive relationship between keyword length and paid clicks. Our results are consistent with longer search phrases resulting in organic results that contain less relevant links, which would result in fewer organic clicks but more paid clicks. 3.5 Other Retailer Characteristics In addition to retail segment fixed effects, the results in Table 2 include controls for other retailer characteristics that might impact automated seo. We discuss each of these in turn. First, note that retailers with a sponsored link on the first page of organic results receive 37 percent more organic clicks after controlling for rank, brand equity, and other drivers of clicks. This positive relationship is consistent with findings of Yang and Ghose (2010) and suggests that these sponsored links may provide searchers information about the retailer that increases the perceived value of clicking its organic link. For instance, such a link might lead searchers to conclude that the corresponding organic listing is relevant; alternatively, the sponsored link might have value as an advertisement that increases the brand equity of the retailer, making consumers more likely to click on organic as well as sponsored links. As with rank effects, however,  is a keyword specific variable so this 37 percent increase applies to the base of clicks from that keyword; it does not imply a 37 percent increase in overall organic clicks. Second, the results in Table 2 indicate that web-only retailers receive about 13 percent more 17 total organic clicks than their bricks-and-clicks counterparts. This highlights that drivers of organic clicks through search engines may differ from those through other channels, such as price comparison sites. For example, Baye et. al (2009) find that bricks-and-clicks retailers selling on a leading price comparison site receive over 25 percent more clicks than their web-only counterparts. Finally, notice that the specification in Table 2 includes additional controls designed to capture potential drivers of clicks that are not accounted for in the specification. These include site age (a potential proxy for cumulative brand equity) and whether the site has a presence on social networks (Facebook and Twitter). While the coefficients for these two controls have a positive effect on organic clicks, they are relatively small and not statistically significant at conventional levels. On balance, we view this as evidence that the effects discussed above are not the result of spurious correlation with excluded drivers of organic clicks. 4 Robustness Checks and Additional Results In this section we demonstrate that our results are robust to a variety of alternative specifications, and offer some additional results that are of potential interest for automated seo related to generating traffic from consumers in different income classes. 4.1 Results Based on Alternative Measures of Retailer Brand Equity One may worry that our results stem from endogeneity issues related to our measure of brand equity. While the Wu-Hausman test did not trigger any formal concerns about our measure of brand equity being correlated with the error in equation (2), one may wonder whether our results are sensitive to this particular measure of brand equity. Table 3 shows that our main findings are robust to using three alternative measures of brand equity that are unlikely to suffer from endogenity concerns. The first specification in Table 3 uses navigational searches on Google from June rather than August to construct the measure of retailer brand equity. Since navigational searches in June were predetermined at the time searchers made their August click decisions, this lagged measure of brand equity mitigates concerns that an unobserved factor drives both navigational and nonnavigational clicks in the August clicks data. As shown in column (1), all parameter estimates, including the brand equity coefficient, increase slightly in magnitude but are qualitatively similar to those reported in Table 2. 18 The second specification in Table 3 uses navigational searches from Bing rather than Google to measure brand equity. Since Bing has a different population of users and employs a different algorithm for returning search results, it is unlikely that unobserved factors that affect clicks on Google are correlated with this measure of brand equity based on navigational searches on Bing. The results in column (2) show that our findings are robust to using this alternative measure of brand equity. The final specification in Table 3 is based on an alternative measure of brand equity pioneered by Animesh, Ramachandran, and Viswanathan (2010). This measure is constructed from data produced by the web traffic reporting firm, Alexa, and measures the “Sites Linking In.” It is based on the number of links to a website from sites that are visited by individuals on Alexa’s web traffic panel.19 Animesh, Ramachandran, and Viswanathan use these data to measure seller quality, noting that links pointing to a website can be viewed as a positive recommendation from the referring site. As shown in column (3) of Table 3, our results are also robust to using this alternative measure of brand equity–as well as to interpreting the brand equity effect identified in our earlier results as purely capturing “seller quality.” 4.2 Alternative Censoring Models Table 4 shows that our main results are robust to using a Tobit censoring model rather than the Heckman selection model used in our main specification. The Tobit model can be interpreted as a constrained version of the selection model, with the selection and outcome equations being equivalent while not allowing for any selection bias. Column (1) reports results controlling for both selection and endogeneity, as in our main specification, while column (2) simply controls for selection. Comparing the parameter estimates to those in column (1) of Table 2, most parameters increase in magnitude and are largely consistent with those reported in our main specification in Table 2. 4.3 Brand Equity and Consumer Income Our main specification in Table 2 assumes that the coefficients for the drivers of organic clicks are identical across consumers in different income groups. We conclude by showing that our qualitative 1 9According to Alexa.com, “Links that were not seen by users in the Alexa traffic panel are not counted. Multiple links from the same site are only counted once.” See also Alexa.com. 19 results are not an artifact of pooling across searchers with different incomes.20 These results are potentially of independent interest, since different retailers may use automated seo to target consumers in different income groups. Table 5 reports the results of estimation by income group, and shows that our main qualitative findings hold in the absence of pooling. Interestingly, however, these results suggest that brand equity has differential effects across individuals in different income classes. For the three lowest income groups, the elasticity of organic clicks with respect to brand equity is smaller than the 0.084 reported in Table 2 based on pooled data, while for the top two income groups the elasticity is greater. Although one of the brand equity coefficients is not estimated precisely enough to infer that it is significantly different from the excluded ($51 to $74 thousand income) category, the results on balance indicate that brand equity is a more important driver of organic clicks for richer than poorer searchers. This result is illustrated in Figure 4, where the dots represent the point estimates for the elasticity of organic clicks with respect to brand equity for the five income groups, and the lines represent the corresponding 95% confidence intervals. 4.4 Relevance of Results for Specialty Retailers Finally, one might worry that our main results are driven by the fact that large mass merchants (such as Amazon) have high levels of brand equity because of the breadth and depth of their offerings, and also tend to receive large numbers of clicks because of this; that is, our regressions could simply be picking up “large retailer” effects. While this concern is mitigated to some extent by the fact that all of our specifications include retail segment fixed effects, our specifications assume that the coefficients for brand equity, rank, and other variables are the same for mass merchants and specialty retailers. Table 6 reports our main results (column 1) along with the results obtained when we exclude Amazon (column 2), Amazon-Walmart-Target (column 3), and all mass merchants (column 4) from the data. Notice that the estimated coefficients are remarkably similar across these different samples. 2 0We also ran specifications that did not pool across searchers of different ages or conducting searches from different locations, but those results did not materially differ from those presented in Table 2. 20 5 Managerial Implications for automated seo Our results are intuitive: When confronted with a list of potentially “relevant” search results, consumers are more likely to click the link of the retailer with the greatest brand equity. That is, holding other drivers of clicks constant, consumers tend to click retailers that are more recognized, trusted, have reputations for providing value (in terms of prices, product depth or breadth), service (well-designed websites, return policies, secure payment systems), and so on. Unlike price comparison sites and other online channels where signals of these attributes may be separately observed (through displays that include user feedback ratings, third-party certification, prices, shipping costs, etc.), the only signals consumers observe in organic product search results are sites’ names (which embody their brand equity) and their “relevance” (as proxied by the rank or position that the search engine’s algorithm assigns each organic link). We also showed that our findings are robust to several alternative specifications and, importantly, to controls for censoring as well as the endogeneity of a retailer’s rank or position in the list of organic results. We conclude by discussing implications of our analysis for search engine optimization, and by providing a few caveats regarding their implementation. 5.1 Rank or Position on Results Pages Our results indicate that rank is an important determinant of clicks; it is hard for a retailer to get organic clicks from a specific product search if its link is not observed on the first five pages of results for that search. For retailer’s above this virtual “fold,” the elasticity of clicks with respect to rank is about unity: a one percent improvement in rank leads to a one percent increase in organic clicks (for that search). Our results thus suggest that there are returns to automated seo efforts that make it easier for search engines to determine a site’s relevance for a particular product search. This includes making effective use of anchor texts, descriptive headings and meta tags, robot.txt files, and using accurate and unique page titles. However, while these sorts of strategies for automated seo are necessary and important, our analysis suggests that it would be a mistake to make them the exclusive focus of automated seo. Rankings are a zero-sum game, and other retailers also have strong incentives to ensure that their sites contain the information needed to be properly indexed by search engines. In light of best responses by other sites, these efforts may prevent a retailer from losing ranks due to miscommunication with search engines, but are unlikely to result in improvements in the equilibrium ranks of a particular retailer’s 21 link. Additionally, strategies designed to improve rank need not result in a long-run increase in organic clicks. This is particularly true of efforts to “trick” search engines into viewing a site to be relevant when it is not. It is also important for managers to recognize that estimates of the impact of rank on organic clicks are keyword specific, and that apples-to-apples comparisons of the benefits of rank versus other drivers of clicks requires an adjustment for the importance of that search in generating organic clicks relative to all relevant searches. Among other things, this means that the returns to focusing on improving rank may be larger for a niche retailer (which sells a single product on a site with a single page) than a mass merchant with thousands of products and pages. 5.2 Site Branding The benefits of including brand equity as part of an automated seo strategy are high. Such investments include increasing consumer awareness (through traditional as well as online advertising), making the site more user-friendly (easier to navigate), providing quality content and service (such as one-click purchases, easy return policies, and using a secure payment system), and more generally, enhancing the value of the brand that underlies the retailer’s link. A number of retailers–including both Amazon and Walmart–have successfully used these strategies. Investments in branding have both direct and indirect effects on organic clicks. The direct effect stems from our finding that consumers are more likely to click on links they know and trust–a finding that is consistent with evidence from other channels, including price comparison and auction sites. But brand equity has an equally sizeable indirect effect: Search engines want to provide users with relevant links, and the brand equity of a site is correlated with the relevance of links, which leads to better ranks and positions. Importantly, the brand equity of a site impacts organic clicks for all relevant keywords, not just those related to a particular search, so there is an amplification effect of automated seo strategies targeted to improve the branding of a site. In addition to spillovers on organic clicks related to searches for other keywords, investments that enhance brand equity are likely to lead to benefits in other channels. These benefits are not accounted for in our estimates, nor in the benefits that other papers document regarding the impact of brand and reputation on sponsored clicks. For example, our analysis focuses exclusively on drivers of non-navigational searches at search engines, so the regression coefficients in Table 2 do not include the benefits of increases in brand awareness or site quality on organic traffic from navigational searches at search engines. Likewise, improvements in a site’s brand equity are likely 22 to result in more direct visits to a retailer’s site, as well as more clicks at other platforms including price comparison and auction sites. Finally, for retailers operating both online and physical stores, some investments (such as advertising) may result in positive spillovers into the physical channel. Our analysis indicates not only that investments in brand equity lead to significantly more organic clicks, but also that these investments are more likely to be sustainable than automated seo efforts focused entirely on rank. Additionally, such investments have spillover benefits in other channels as well, as has already been documented in extensive research on other online markets as well as traditional retail channels. For all of these reasons, we conclude that site quality, brand awareness, and other investments that enhance the brand equity of an online retailer are important components of an overall automated seo strategy. 5.3 Search Term Considerations Some search terms and phrases are more likely to generate clicks than other keywords, even when one accounts for differences in the brand equity of different retailers and their ranks in search results. This is potentially relevant for automated seo as well. For example, our finding that searchers including the specific brand of a product (e.g., “Levi’s Jeans”) in a search are more likely to click an organic link suggests that retailers selling branded products benefit by ensuring that their sites present information about the brands in their portfolio in a way that allows search engines to properly index them. Likewise, searchers using longer keywords are less likely to click an organic link, so parsimony in this regard is also important for automated seo. 5.4 Demographic Considerations Our findings that individuals that are older, have higher incomes, or who conduct product searches at work are more likely to click organic links also have ramifications for automated seo. Among other things, these results suggest that automated seo is more likely to be important for retailers targeting consumers with these demographic characteristics. In addition, since the elasticity of organic clicks with respect to brand equity is higher for individuals with higher incomes, the marginal benefits of automated seo efforts targeted at improving the quality and brand awareness of a site are greater for retailers targeting individuals with higher incomes. More generally, the key implication is that the benefits of automated seo vary, depending on the demographic characteristics of the consumers retailers are attempting to attract through this channel. 23 5.5 Retailer Considerations The relationship between sponsored links and organic clicks identified in our data highlights yet another set of spillovers that complicates the calculus of automated seo. Retailers attempting to increase traffic through organic links should recognize that there are possible spillovers from paid links: Consumers are more likely to click organic links associated with sponsored links. On the surface, this might seem like a pure win for retailers, since a sponsored link that results in an organic click rather than a paid click costs nothing. This is unlikely to be part of a sustainable strategy, however. Ultimately, if consumers click a retailer’s organic rather than sponsored link, its prospects for winning that sponsored link in an auction will decline, since search engines have little incentive to allocate scarce ad space to retailers that do not receive sponsored clicks. 5.6 Concluding Caveats Our analysis has focused on the potential benefits of automated seo by focusing exclusively on the drivers of organic clicks. We have not taken into account the costs of improving these drivers, such as the costs of improving the meta tags associated with a particular keyword to improve rankings or advertising through traditional media to improve the brand awareness of a site. Costs are obviously an important component of optimization, and it would be a mistake to base automated seo decisions purely on the drivers documented above. Future research documenting the costs of different automated seo strategies is therefore also important for the automated seo literature. It is also important to recognize that search engines are only one of many online platforms where consumers conduct product searches. Baye et al. (2013) note that in June 2012, consumers using browsers conducted 634 million product searches at retailer sites (such as Walmart.com), 134 million product searches at price comparison sites (such as Dealtime.com), and 877 million searches at marketplace sites (such as eBay.com). They also point out that 70% percent of eBay’s listings were for new products, and over 60% percent of its listings were through posted prices rather than auctions. Unlike automated seo efforts designed to improve rankings at a search engine, automated seo efforts to improve a retailer’s brand equity can improve the clicks it receives from searches in these other channels. Since these spillover benefits are difficult to quantify, it is easy for those engaging in automated seo to underestimate the benefits of investing in the quality and brand awareness of a site