How Search Engines Shape Gaze Patterns During Information Seeking: Google vs. Baidu

How Search Engines Shape Gaze Patterns During Information Seeking: Google vs. Baidu

Over decades, the complexity of the search-engine results pages (SERPs) has grown dramatically. In the recent years, interactive and informational features have been added to SERPs.

For example, if you searched for Charles Barkley in 2003, you would get nine organic results and some sponsored links placed at the top of the search results or in the right sidebar. Now, the same query could result in a SERP with many more components, such as a knowledge panel, a People Also Ask element, a video pack, and so forth.

Searching for Charles Barkley in Google Search in 2003 (top) and 2020 (bottom, which is cropped to show the content visible above the folder): The visual components of SERPs are much richer than they were 17 years ago.

This trend is valid for other search engines around the world. Baidu, the most popular search engine in China, also shows a lot more components on its SERPs today compared to 10 years ago.

Searching for Yao Ming in Baidu in 2009 (top) and now (bottom): The current SERP of Baidu (bottom) has more interactive and informational components compared to its 2009 version (top).

Google and Baidu SERPs: Similarities and Differences

Besides sponsored results and organic results, there are several shared features that we observed in both Google and Baidu SERPs:

  • Packs (videos and pictures): Preview of certain media content that is related to the query
  • Featured snippet: A summarized answer from an organic result, showing at the top of the SERP

Despite the similarities, Baidu’s SERP layout showed some differences compared to the Google SERP’s. Let’s use an example to illustrate it: searching for orchid.

The main body (excluding the right rail) of Google’s SERP for orchid contained 5 different components: a local pack, a video pack, an image pack, a People Also Ask element, sponsored shopping cards, as well as organic search results. The right rail contained a knowledge panel.

For the query orchid, Google returned five different SERP features including organic results in the page body and an additional knowledge panel in the right rail. (There was also an image pack so far down the SERP that it’s not visible in this screenshot, but we didn’t include it in our count.)

In contrast, searching for the same query, orchid, in Baidu resulted in only 2 features in the main SERP body — an image pack and a video pack. Baidu’s right rail had more features than Google’s: ads, related content, top searched flowers, and most popular searches. However, the four elements in Baidu’s right rail were less related to the keyword orchid than Google’s knowledge panel. This low relevance decreases the probability that people will look at them.

For the same query orchid, despite richer content displayed in the right rail, Baidu returned fewer SERP features in the main body of the page (organic results, video and image packs). The right rail had ads, related links, most searched flowers, and most popular searches. (Most Popular Searches are the most popular Baidu search queries currently and totally irrelevant to the search query.)

We already know that the evolving layout of the SERP is shaping how people search in Google. A subsequent question comes to mind: Do the changes of Baidu SERP also influence how people react to the search results? Furthermore, will user behavior patterns of these two search engines be different because of the different SERP layouts?

To update our How People Read Online report and answer these questions, we conducted eyetracking studies in both Raleigh, USA, and Beijing, China. We’ve found that many gaze patterns are the same, yet some are different. In this article, we discuss the different eyetracking patterns we observed in Google and Baidu SERPs.

The Pinball Eyetracking Pattern

Unlike the layer-cake and love-at-first-sight patterns, of which we found many Chinese instances, the pinball pattern, newly discovered in the Google SERP, was extremely rare in Baidu during our study.

A pinball pattern refers to a highly nonlinear gaze path on a SERP, where the user bounces around between organic results and SERP features.

For example, when a participant in our Raleigh study tried to learn more about a podcast named My Favorite Murder, her gaze bounced between the first search result, the knowledge panel at the right sidebar, and the carousel of similar podcasts at the top of the page.

A participant’s gaze moved in a pinball pattern while she was searching for the podcast My Favorite Murder. Her attention bounced between the organic search results and other components like the carousel and the knowledge panel.

In contrast, we observed only one instance of the pinball pattern on a Baidu SERP, out of more than 60 searches. Most of our participants ignored the right rail. Their eyes moved in a sequential, love-at-first-sight, or layer-cake pattern on SERPs.

For example, a participant tried to find museums in Shanghai to recommend to his friend. He searched for Shanghai museums and looked at the search results only. There were some museum recommendations in the right rail, but they were below an ad. He didn’t even glance at them.

A participant in the Beijing study who searched for Shanghai Museums didn’t even look at the right rail. His eyes engaged in a sequential scanning of the main SERP body.

We hypothesize that the absence of pinball patterns in Baidu is due to the low relevance of the right-rail elements to the user’s current query. Compared to Google’s knowledge panel, which often is a definition or explanation of the terms in the query, the first element of the right rail in Baidu SERPs was frequently an ad. That ad discouraged users from continuing to scan the right rail in those cases where their eyes looked at it. Moreover, the last part of the right rail was most popular searches, which was irrelevant to the current query at all. Thus, because the right rail had a fair-to-high concentration of irrelevant content, users were likely to quickly abandon it even if they happened to glance at it.

The one instance when we did the pinball pattern in Beijing was when a participant was asked to research Botox for crow’s feet. When she searched for Botox best clinics, several fixations were on the right sidebar, but the majority of her time was spent scanning the organic search results. There were 5 components in the sidebar:

  • An irrelevant ad
  • Related terms (which contained images)
  • People Also Search (also with images)
  • Related procedures (also with images)
  • Most Popular Searches

When she was trying to search for best places for a particular procedure, the elements above were of low relevance to her current task. We think that she looked at the sidebar because of the human faces shown in many of the elements — humans are sensitive to images of faces and eyes. That could explain why her attention was grabbed by the related terms and procedures for a while. She lost interest immediately, after realizing this information was not helpful for her current task.

The only instance of the pinball pattern that we observed in our eyetracking study in China came from a participant who searched for Botox best clinics in Baidu. She directed a few fixations to the right rail, and then quickly bounced back to the SERP main body, where she spent most of her time.

For comparison, here’s a typical pinball pattern for the same task from a participant in Raleigh. When she searched for Botox Raleigh in Google, the right rail contained a knowledge panel with information about a local clinic, Raleigh Botox and Laser Center. Because the knowledge panel was directly related to her task, she spent lots of time on its content, fixating the rating, address, and even the Google reviews, then her eyes bounced back to main body.

A US participant searching for Botox Raleigh in Google exhibited a typical pinball pattern. Her attention bounced between the organic search results and the knowledge panel, which had detailed, task-relevant information about a local clinic offering related services.

The Content-Promotion Model of Baidu

Another interesting comparison is between the content-promotion models of the two search engines. Compared to Google, Baidu promotes its products more obviously. (While there have been accusations that Google surreptitiously promotes its own products in its ranking algorithm, such actions, if they occur, are not easily perceptible to the average user.)

In Baidu, when a participant searched for Shanghai Museums, 4 of the top 6 search results were Baidu properties: 2 Baidupedia (like Wikipedia) results, a Baidu Knowledge result, and a Baidu Maps result. The user did visit one of the Baidu sites and made his decision based on the info on that site.

The Baidu SERP for Shanghai Museums included links to 4 Baidu products in the top 6 search results.

When Baidu’s Approach Works

Out of 10 research sessions performed by our Beijing participants, in half of the sessions people made decisions based on the information they got on Baidu properties.

This is okay when the promoted content is highly related to the search query. People make decisions based on the content they perceive as trustworthy, high quality, and relevant. They won’t count how many Baidu products they have used when making a decision.

For example, a study participant tried to learn more about the Summer Palace, which she planned to visit for the coming weekend. She searched for What to do in the Summer Palace and browsed 3 Baidu-related pages. She first scanned a Baidu Quora page with an article about tourist-spot information. Then she went through a Baidu Knowledge page with pictures and texts introducing each spot. Finally, she opened a Baidu Maps page, where she located the places that she was interested in. She felt well-informed and prepared for her trip after the investigation. In this particular session, Baidu did an excellent job of promoting its content while also answering people’s questions.

When Baidu’s Approach Doesn’t Work

Compare this successful content promotion to the complete disregard that users had for the Baidu links presented in the right rail. The right rail is clearly a lost opportunity for Baidu. The right-rail elements were viewed as either ads or content of low relevance for the current tasks, even though occasionally they were meaningful.

In the long run, people form a mental model of the right rail and develop banner blindness for that area. Even if they direct a few fixations to it sometimes (as in the Botox example), seeing low-relevance content again and again can only reinforce their belief.

However, as Google’s example shows, people can alter their behavior patterns. The Google SERPs also used to have promotional links in the right rail. Still, the informative knowledge panels, visually different from the traditional ads, did change people’s mental models and captured people’s attention.

Thus, it’s okay to promote your content on the SERP, as long as it is relevant to the user’s task.

This guideline is not only useful for designers of global SERPs. When related content is done right, it boosts pageviews — whether on a SERP or on a different page type.

For example, Net-A-Porter, an ecommerce site, promoted related items based on the different characteristics that might interest users. For a Matteau black dress, the similar-item panel offered related products in different categories including Featured, Maxi dresses beachwear, Black linen dresses, Matteau dresses, and Cotton dresses. This design made the recommendations precise and encouraged exploration.

The Net-A-Porter site promoted similar items based on different characteristics of the original item. This approach increased the chance that users explore more related products.

No matter which page types you’re designing for, SERPs, product pages, or article pages, remember that content promotion must be rooted on relevance. A hard sell will discourage exploration and eventually lead to banner blindness — just like it happened with Baidu’s right rail.

Conclusion

Information-seeking behaviors shift alongside design changes: new designs can trigger new behaviors and eyetracking patterns, particularly when they are frequently encountered (like with Google and Baidu SERPs). Thus, big companies should take responsibility because they are shaping user behaviors. It means that a single design decision could be impactful in the long run.

However, not all design changes will affect user behavior — people will change their mental models and actions only if they see a gain in doing so. Google and Baidu clearly illustrate that point. Google’s knowledge panels, which contain relevant, condensed information, have taught users that its right rail is no longer for ads. But Baidu’s popular searches and related links were not interesting enough to alter the perception that the right rail is just for advertisement content.

Site designers should try the following suggestions to get attention to their content on global SERP pages:

  • Optimize your content to answer the most frequently asked questions of your target user group. This can make your content likely be pulled out as a featured snippet or in the People Also Ask element. As we have observed in our study, an informative featured snippet is alluring to participants in both the USA and China. Even if they stop on the SERP because the featured snippet has answered their question, their perception of your brand will improve.
  • Offer content in various formats. As search engines are showing videos, pictures, local packs of specific queries, making content available in different formats increases the visibility of your site.
  • Don’t cheat your users by randomly presenting irrelevant, eye-catching content, aiming at being presented in the SERP features for as many queries as possible. If they search for something and arrive at your page with content that is not relevant to their questions, they will feel cheated.

Designers working on site search can also learn from the big search engines.

  • Adopt suitable SERP features to save user effort. Good SERP features like featured snippets are meant to reduce the time it takes for people to find specific pieces of information. If you’re designing an intranet where employees need to search for profile information or phone numbers every day, use a featured snippet with that information. You may find you’ll save a lot of money for the company.
  • Showcase the variety of content on your site. People may not be aware of the content types available on your site. Using video and picture packs can showcase a range of content and can benefit users who prefer watching to reading. (As long as you keep the showcased content relevant to the user’s current interest, as evidenced by their query.)
  • Be cautious about displaying low-relevance links or promotional content. It can be tempting to put promotional content or low-relevance links somewhere on the SERP —someone may use them! But, like the Baidu’s right rail, it’s actually more likely that people will quickly learn to ignore them. Save the place for critical content.

A final point: if you used to feature irrelevant promotions, don’t despair if you follow our advice and change to relevant information but don’t see an immediate improvement in site metrics. Changing user behaviors take time. If people have been trained (by your old bad design) to ignore certain areas of your page design, they won’t even see the new and better information you place there. At rare occasions, people might steal a random glance at the area, and if they are positively surprised, they will gradually change their behavior. But it can take a long time for intermittent exposure to build into true interest for a previously disliked page area. In one example, it took a year for the difference between two design alternatives to stabilize.