Newsrooms spend time and resources attracting audiences to their websites. Once audiences get to the site, the goal is to provide content that is compelling enough to keep them there. Recirculation, or “the percentage of users who visit another page of your website after they finish reading their first article,” is driven in part by links that promote other content on the site. This makes links critical components of news sites, as they allow newsrooms to serve their audience with desired content.
The purpose of this report is to explore best practices in using links on news sites. This research was funded by The Lenfest Institute for Journalism. We partnered with seven local broadcast newsrooms owned by the Graham Media Group. We examined the first time each unique visitor went to the site and, in total, looked at 1.8 million observations. We evaluated four factors in our online experiment:
- Whether the links were text-only or a combination of text and images,
- The placement of the links on the page,
- The type of content included in the links, either Related Stories or Popular Stories, and
- The wording used to describe the links.
We collected data for one week, from March 22nd to March 28th, 2018. For each user, we have data on whether the person clicked on a link, as well as the referral site and the device they used. We tested whether the results varied by station, by referral site, or by device.
The results show that few users click on links. Among those who first visited the site during the study period, 1.42% clicked on a link. Although this appears to be a low rate of engagement, content distribution sites like Outbrain suggest that healthy click-through rates are between 0.10 and 0.25%. The results of our study show that links can be designed to generate higher click-through rates. In fact, in nearly all instances, it didn’t matter which device was used, the specific news site visited, or even the referral page. The same design practices generated the highest click-through rates.
In summary, the results show:
- Link layouts containing images generated 63% more clicks than those that consisted of only text
- Links at the end of a page generated 55% more clicks than links in the middle of a page
- Overall, using related content instead of popular content led to a 14% increase in clicks
- Popular content, however, generated more clicks when the referral page was Facebook
- Generic wording (e.g. Related Stories) generated slightly more clicks than more complex wording (e.g. What Else People Can Read on This Topic)
- It didn’t matter whether people used a smartphone, phablet, tablet, or desktop. It didn’t matter which site users visited. It didn’t matter which page referred them to the site. In all of these instances, there were more clicks when links (1) appeared at the end of articles, and (2) contained images.
USE IMAGES IN LINKS
The news sites in this study tested two different types of links:
- Links with text only
- Links with images and text
Across all types of mobile and desktop devices, clicks were higher for pages with links including images than links consisting of only text. On average, including images and text with links generated 63% more clicks than using a text-only layout.
PLACE LINKS AT THE END OF THE PAGE, NOT IN THE MIDDLE
The news sites in this study tested two different locations for links:
- The middle of the article page
- The end of the article page
Across all types of devices, clicks were higher when the links appeared at the end of the article rather than when the links appeared in the middle of the article. On average, placing links at the end of an article generated 55% more clicks than placing the links in the middle of the page.
Placing links at the end of the page was even more successful for mobile (61%), phablet (82%), and tablet (62%) compared to desktop (50%).
RELATED CONTENT IS BETTER THAN POPULAR CONTENT, USUALLY
The news sites in this study tested two different types of link content:
- Popular content, which was generated from articles trending on the site at that time
- Related content, which was generated from articles related to the story on the site
Across all types of mobile and desktop devices, clicks were higher for pages with links using related content instead of popular content. On average, using related content generated a 14% increase in clicks over links containing popular content.
Popular content did outperform related content in one instance: when users visiting the news site were coming from Facebook. Though the difference in clicks is small, users coming from Facebook to the news site were 7% more likely to click on links containing popular content than those containing related content. Related content generated an increase in clicks compared to popular content when users were coming from Google (39%), Outbrain (44%), and Other sites such as the news homepage (16%).
This is especially important given that 416,386 users were referred to the news site from Facebook. By comparison, search sites like Google and Yahoo referred a collective 33,525 users and Outbrain referred only 24,659 users.
HOW YOU LABEL THE LINKS MATTERS, BUT ONLY SLIGHTLY
The news sites in this study used three different types of wording to label the content of the links:
- A generic label (Popular Stories or Related Stories)
- A label to attract users seeking information (Learn More from Trending Stories, Learn More from Similar Stories)
- A label to attract users with social motivations (What People Are Reading Now, What Else People Can Read on This Topic)
Across all types of mobile and desktop devices, clicks were modestly higher for links that used the generic labels of “Popular Stories” or “Related Stories.” The generic wording generated only a 4% increase in clicks over both links that used wording based on social motivations and links that used wording based on seeking information.
Our research aimed to identify what characteristics of links on news sites are important in encouraging individuals to click on links and increasing recirculation. We found significant differences for each of the factors, but recognize that, consistent with other industry estimates, individuals rarely click on links.
When individuals do click on links, they are more likely to click when the links appear at the end of the page, reflect related content and use images in their layouts. These findings hold true for users on smartphone, phablet, tablet and desktop devices, which provides newsrooms with a set of best practices for constructing their links without having to make changes for different types of devices their users might use to access their site.
These factors are not exhaustive of the characteristics that might influence the propensity to click on links, but rather provide a starting point to help newsrooms understand characteristics that might increase clicks. It is important to keep testing these sorts of factors to see if there are ways of increasing the overall click-through rate. This is important not only for business purposes, but also for the purpose of providing audiences with additional information that meets their needs.
It is important to note that this research is based on the findings of seven news organizations, but it is not clear whether the same results will persist for others. The organizations were geographically diverse, although all were in mid-sized markets. We also note that these patterns persisted across all seven outlets.
Newsrooms can utilize the findings presented here to refine their existing links, but should also consider ways in which to innovate upon links that might generate greater user engagement overall.
From March 22 to 28, 2018, seven mid-sized local broadcast news sites randomly varied four characteristics of the links on their pages: (1) link placement, (2) content type, (3) link wording, and (4) link layout. The news sites collected data for all 4.5 million users visiting the sites over the course of the week. Multiple visits for the same users were omitted from the analyses so that only users’ first interaction with the site was tested, resulting in 1.8 million observations.
The first characteristic, link placement, varied whether links appeared in the middle or at the end of an article page. We also varied whether users saw links to related content or links to content that was trending at the time. The wording used to describe the links was either generic, targeted at users’ cognitive motivations, or targeted at users’ social motivations. The variations of that wording differed according to the type of content displayed and are as follows:
|Wording Type||Popular Content||Related Content|
|Generic||Popular Stories||Related Stories|
|Cognitive||Learn More from Trending Stories||Learn More from Similar Stories|
|Social||What People Are Reading Now||What Else People Can Read on This Topic|
Lastly, the links varied in their layout style where half displayed images and text, and the other half displayed text-only.
To analyze data, we used logistic regression models to compare differences in clicks. Only significant results are discussed in the text. The results of these tests are included in the endnotes.
*Jessica Collier is a Research Associate with the Center for Media Engagement and a doctoral student in the Department of Communication Studies at the University of Texas at Austin. Natalie (Talia) Jomini Stroud is an Associate Professor in the Department of Communication Studies and School of Journalism and Director of the Center for Media Engagement at the University of Texas at Austin. This research was made possible thanks to a generous grant from The Lenfest Institute for Journalism. The Lenfest Institute staff and the other grantees also helped to push this project forward, and we appreciate their thoughts. The authors wish to thank Jerry Jones for his help in creating the code used in this study. We are incredibly grateful to the Graham Media Group, especially Catherine Badalamente, Jonathan Beard, Ian Bonner and Mike Katona for their assistance in making this project possible. We also thank Melody Avant, Johanna Dunaway, and the Center for Media Engagement team for their assistance and advice throughout this project.
 We ran a logistic regression model to calculate differences. Compared to the layout with images (included as the reference group), the text-only layout (B = -0.49; SE = 0.01; p < .001) generated statistically significantly fewer clicks. There were some differences depending on the referral site, although in all instances, clicks were higher for images than text. When Google was the referral site, clicks increased by 88% when images accompanied the links compared to text-only. When the referral site was Outbrain, the percentage was 85%, 43% for Facebook, and 66% for other referral sites, such as the homepage.
 We ran a logistic regression model to calculate differences. Compared to placing links in the middle (included as the reference group), placing links at the end of the article (B = 0.44; SE = 0 .01; p < .001) generated statistically significantly more clicks.
 We ran a logistic regression model with interactions between link location with middle placement included as the reference group. The two-way interactions between link location and device type (desktop was included as the reference group) show that for mobile (B = 0.07; SE = 0.03; p < .05) and phablet (B = 0.19; SE = 0.08; p < .05) placing links at the end of an article on each of these devices generated statistically significantly more clicks.
 We ran a logistic regression model to calculate differences. Compared to the popular content (included as the reference group), related content (B = 0.13; SE = 0.01; p < .001) generated statistically significantly more clicks.
 We ran a logistic regression model with interactions between the type of content in links (popular content included as the reference group) and the referral page users were coming from prior to viewing the links (“other” pages were included as the reference group). The two-way interactions between content type and referral page show that related content generated statistically significantly fewer clicks when the referral page was Facebook (B = -0.22; SE = 0.04; p < .001). On the contrary, when the referral page was Google (B = 0.17; SE = 0.10; p < .10), Yahoo (B = 1.12; SE = 0.51, p < .05), and Outbrain (B = 0.21; SE = 0.07; p < .01), related content generated statistically significantly more clicks.
 We ran a logistic regression model to calculate differences. Compared to using generic text (included as the reference group), using labels to attract users seeking information (B = -0.04; SE = 0.02; p < .05) or using labels to attract users with social motivations (B = -0.04; SE = 0.02; p < .05) generated statistically significantly fewer clicks.
 We ran logistic regression models with interactions between each of the characteristics tested and each of the news sites to observe whether different characteristics were better (or worse) for different sites (site 7 included as the reference group). For each of the seven outlets, links with text-only (links with images included as the reference group) were always better and for two outlets, the difference was statistically significant (site 5: B = -0.26; SE = 0.12, p < .05; site 6: B = -0.26; SE = 0.12, p < .05). For each of the seven outlets, related content (popular content included as the reference group) generated greater clicks and for three outlets, the difference was statistically significant (site 1: B = 0.22; SE = 0.12, p < .01; site 3: B = 0.30, SE = 0.14, p < .05; site 5: B = 0.37; SE = 0.12, p < .01). For each of the seven outlets, there were no significant interactions between placement (middle included as the reference group) and site. For each of the seven outlets, there were no significant interactions between wording (generic included as the reference group) and site.