Can Emotions Be Used as Keywords for Text-Based, Search-Engine Advertising?

Abstract

Click-through rates (CTR) of search-engine advertising remain elusive, thus triggering the need for more research on ad effectiveness. Using simple regression techniques on a data set of 110,000 search query–text ad pairs, we show that it may be meaningful for advertisers to target search queries by (1) using the emotion of the user as indicated by an emotion word in the search query and, then, (2) using positive emotion content in their ads to achieve congruency with the search-query emotion. The results indicate non-spurious correlations that advertisers might consider in their search engine advertising strategies.

Keywords:

• Emotional appeals
• advertising

Notes

1 Available at www.statista.com/ statistics/237974/online-advertising-spending-worldwide/#:∼:text=It%20was% 20calculated%20that%20the, 389%20billion% 20dollars%20in%202021.

2 Available at https://digitalfireflymarketing.com/our-blog/ search-engine-optimization-pay-per-click-importance/#: ∼:text=Paid%20search%20results%20gain%201.5,track %20ROI%20and%20manage%20budgets.

3 Available at https://valveandmeter.com/ pay-per-click-statistics/.

4 Available at https://www.zerolimitweb.com/organic-vs-ppc-2018- ctr-results-best-practices/.

5 The emotion word dictionary is available online at www.dropbox.com/s/7xwgfy6cmqwio6x/emotion_ word_onlineAppendix.xlsx?dl=0. Also, see the Appendix for details about mining emotion words using the word embedding technique.

6 Available at www.wordstream.com/keyword-intent.

7 To give the audience a sense of the scale of the impact of emotion congruency, we did a rough comparison of the impact of emotion congruency and the impact of ad quality score. Specifically, we create a binary variable AdQuality_binary, where 1 indicates high ad quality score and 0 indicates low ad quality score, and the high/low ad quality is split at the median of ad quality scores. Similarly, we create a binary variable Congruence_binary, where 1 indicates high emotion congruence and 0 indicates low emotion congruence, and the high/low congruence is split at the median of the cosine similarity between the word embedding vectors of the emotion words in query and ad in each query-ad pair (when emotion words do exist in both query and ad). Cosine similarity represents the congruency between emotion words in the query and ad, and provides a continuous variable where higher levels indicate higher congruency or sematic similarity and lower levels indicate lower congruency or semantic similarity. Only for this specific analysis in this footnote, we measure congruency between emotion words in query and ad using the cosine similarity technique. We then fit the following model: $Ad CT{R}_i={\beta }_0={\beta }_1\mathit{Congruence}\backslash \_\mathit{binar}{y}_i+{\beta }_2\mathit{AdQuality}\backslash \_\mathit{binar}{y}_i+{\beta }_{3}Q\mathit{ueryHasEmotio}{n}_i+{\beta }_{4}A\mathit{dHasEmotio}{n}_i+O\mathit{ther}\mathit{Control}\mathit{Variable}{s}_i+{\gamma }_1\mathit{CongruencyExis}{t}_i+.$

There can be a confound in the interpretation of the estimate of Congruence_binary (β₁) because the value of 0 in the measure of Congruence_binary includes observations with absence of emotion words, i.e., due to the absence of emotion words in both query and ad or absence of emotion word in only query or absence of emotion word in only ad. In order to control for this issue, we include a dummy variable CongruencyExist, which captures the presence of emotion words in both ad and query in the ith query-ad pair. Hence the coefficients of AdQuality_binary and Congruence_binary represent the scale of the impact on CTR of ad quality score and emotion congruency, respectively. The coefficients for both variables are significant with p < .01. The coefficient for AdQuality_binary is 0.031, while the coefficient for Congruence_binary is 0.0017. The coefficients imply that, by shifting from low emotion congruency to high emotion congruency, the impact on CTR is around 5.5% of the impact when shifting from low ad quality score to high ad quality score. Since the ad quality score is one of the most important factors for effective search advertising (Nabout and Skiera 2012), the analysis suggests that the scale of the impact of emotion congruency is also substantial.

The goal of this analysis is not to compare the significance of emotion congruency and ad quality. It only serves as a “back-of-envelope” estimate and the main purpose is to provide suggestive evidence on the scale of the impact of emotion congruency.

8 Available at www.wordstream.com/keyword-intent, https://support.google.com/google-ads/answer/6167118?hl=en.