“Google the Gatekeeper: How Search Components Affect Clicks and Attention”

These components produce tensions between searchers, 3rd-party websites, and Google itself over clicks and attention."

The presentation of results on Google Search has evolved significantly from its humble beginnings as a list of ten blue hyperlinks. In 2007, Google introduced “Universal Search,” blending results from its Images, Maps, News, and Video vertical search properties into a single Search Engine Results Page (SERP).[1] Five years later, Google added the “Knowledge Graph,” which powers the presentation of facts in the main results column and knowledge panels on the right-hand side of the SERP.[2] In 2014, Google added “Featured Snip- pets,” which extract a prominent and readable page description from a website and place it above the website’s listing.[3] Collectively, these additions to SERPs beyond simple, organic links are referred to as components (Robertson, Lazer, and Wilson 2018).

We find that both direct-answers and featured-snippets decrease CTR and increase time on the SERP, which is broadly consistent with prior work (Chilton and Teevan 2011; McMahon, Johnson, and Hecht 2017; Epstein et al. 2022). As expected, direct-answers have a larger effect than featured-snippets on both outcomes. Interestingly, we do not find a significant effect of knowledge-panels on either outcome.

Our findings highlight the tensions that knowledge components produce between different stakeholders. Users may benefit when they can answer their questions marginally faster using information contained directly on the SERP (Williams et al. 2016). However, the effect of good abandonment on clicks is consequential for 3rd-parties, who may lose revenue and visibility. This impact is especially significant for volunteer-driven non-profits (e.g., Wikipedia) and websites whose value is intrinsically built on having an engaged user base (e.g., StackOverflow) (Vincent et al. 2019). Historically, Google has also misappropriated content from small 3rd-parties, e.g., lyrics site Genius and Celebrity Net Worth, for use in knowledge components (Jeffries and Yin 2020). One interesting development here is the Wikimedia Foundation’s release of a commercial product for companies that use high volumes of Wikipedia content.13 Google and the Internet Archive are its first customers.

The effects we observe increase the importance of work on the credibility and quality of content in knowledge components (Lurie and Mustafaraj 2018; Lurie and Mulligan 2021). Future work should heed Mustafaraj, Lurie, and Devine (2020)’s argument for “voter-centered audits” (i.e., user-centered audits) that focus on the types of information needs for which searchers use these components. For example: what important information needs (e.g., political, health, financial, and/or legal) do knowledge components attempt to satisfy? Lab studies have also investigated the effects of featured-snippets and direct-answers on peoples’ beliefs (Ludolph et al. 2016; Epstein et al. 2022). Future work could extend these studies to ecological settings and assess whether behavioral effects are ephemeral.

Of course, the elephant in the room is the introduction of chatbot-based search on Google and Bing, which is nothing if not a knowledge component on steroids. Chatbot-based search exacerbates concerns about abandonment, misappropriation, and information quality (Robertson 2023) that were originally discussed in the context of knowledge components. Future studies could adapt the methods in this paper to study users’ interactions with chatbot-based search.