Based on a brief from the "Generic Visuals in the News" research team, we analyzed the spread of “generic visuals” in the news by focusing on the engagement, circulation and recontextualization of ‘generic’ visuals from a selected group of news media accounts. We developed a comprehensive classification and general description of the visual genres of these generic visuals.
Our key findings are
Concerning engagement, we observed that the most engaged visual genres are platform-related: whereas the “web card” genre is the one that sparks the most engagement on Facebook, according to our Twitter dataset, the most engaging visual genres are the data visualizations.
The network analysis of the circulation of images from the platforms to different domain names reveals that the images circulate the most on Facebook and Twitter. Through the clusters in the network visualization, we also identified relations between Facebook and clickbait websites on one side, and between Twitter and newspaper websites on the other.
While looking at the recontextualization, we were able to analyze in-platform mechanisms of recontextualization for a data visualization on Twitter and to trace the history of de-contextualization and recontextualization for a web card on Facebook. We found that the data visualization on Twitter often doesn’t detach from the original context - perhaps because of the “retweet” function that allows to repurpose the original contextual markers. Contrastingly, the image on the web card analyzed for Facebook inserts itself in a network of very heterogeneous contexts, probably because it went through an initial process of de-contextualization.
Web cards as one of the most-engaged visual genres which may respond to a current trend of algorithmic-driven content production. This type of visual acts as engagement bait to prompt user engagement, which informs recommendation algorithms used on social media platforms.
Regarding the data collection and analysis for image circulation on Twitter, we identified some limitations. The API only offers the option to collect data through "quote tweet" and "retweet". However, recontextualization also occurs on Twitter as shared screenshots of a tweet. This finding begs for further methodological exploration.
This project was conducted based on a research brief from the "Generic Visuals in the News" research team. Despite our increasing exposure to generic visuals, very little is known about the role that they play in relation to the assembling of publics. Very little research has acknowledged the significance of visuals in the news, while existing analyses prioritize arresting and iconic images (e.g. Zelizer 2010), and recent studies of data visualization in the media have likewise focused on those that are award-winning or considered beautiful (e.g. McCosker and Wilken 2014).
Generic visuals are here defined as “visuals which have standardized formats and appearances, which perform particular design functions and which circulate with increasing frequency in the news media, such as stock photos and simple data visualizations,” according to the "Generic Visuals in the News" project.
Originally, the Generic Visuals in the News team had identified 4 research questions to work as a guide for the empirical research of the Data Sprint:
What do these generic visuals represent? What do they communicate and what are the visual characteristics of generic visuals that have created a high engagement rate?
How do these generic visuals circulate? How do they exist and move across different online platforms and media outlets? Who shares them and for which purposes?
How are these generic visuals recontextualized? How are they used across different media texts? And by whom and for which purposes?
How are these generic visuals connected, with each other and with particular types of news stories? For example, are certain types of stock photos consistently used together with specific types of simple data visualizations, and if so, how? Are there groups of visuals that are consistently used for the same types of news stories?
For both methodological and empirical purposes, we reformulated the research questions focusing on generic visuals’ engagement, circulation and recontextualization.
We’ve collected the 500 most retweeted tweets containing images published by the Financial Times Twitter account with the help of 4CAT. We downloaded the images and saved them in a shared folder.
RQ1: What are the most engaged with “generic visuals” on social media news accounts between 2020 and 2022?
Sub question 1: What are the most engaged with generic visuals on the Facebook accounts of six UK Tabloids in the Reach Group?
Sub question 2: What are the most engaged with generic visuals on the Twitter account of the Financial Times?
Sub question 3: Is it possible to identify different genres of the most engaged generic visuals in these media outlets?
RQ2: How do generic visuals circulate in the news? Do they go through processes of recontextualisation?
We decided to investigate three different dimensions of the spread of generic visuals:
Engagement is understood as how much an image was commented on, reacted to or shared.
Circulation is understood as a collection of the sites in which the image can be found outside of the observed initial data.
Recontextualization and “assembling publics”, are understood as the different uses in different domains of a single picture and the different publics assemble around it.
After downloading the 500 most engaged tweets containing images by the FT Twitter account, we performed a cluster analysis along visual features of the images with PixPlot.
Figure 1: Screenshot from the visual analysis with PixPlot showing the 500 most retweeted images published by the Financial Times Twitter account.
We used Crowdtangle to identify and download the 700 most engaging images of four Tabloid accounts on Facebook. We chose six news media accounts from the Reach Group, selecting both national and local tabloids:
|National Brands||The Mirror:||https://www.facebook.com/dailymirror|
|Local Brands||Birmingham Live:||https://www.facebook.com/birmingham.live|
Table 1: Selected tabloid Facebook accounts for engagement analysis.
Subsequently, we used PixPlot to analyze the 700 most engaging images from Facebook. Noticing that the visual features of the images did not emerge in PixPlot, we opted to use PicArrange (see figure 2).
Figure 2: Screenshot of PicArrange analysis of the Facebook data set.
We also stacked the images within each category of generic images to further examine their visual features.
For both platforms, we developed a vocabulary for different genres of generic visuals in the news.
We retrieved the URLs of all the images via Google Vision API:
We then cleaned the data by converting all the URLs to domain names and created a network on Gephi to visualize how the images circulate on different websites. In the networks, nodes represent domain names of websites on which the social media images were circulated, and edges represent the social media images. We visualized the network in ForceAtlas2 layout, and assigned different colors to clusters of nodes based on the calculation of “Modularity.”
From the 500 most engaging images from the Financial Times Twitter account, we selected a data visualization and traced where and how it has been re-contextualized on the web and Twitter. We chose this visualization because it corresponds with the working definition of generic visuals and received many retweets (2343).
Figure 3: Image embedded in a tweet by @FinancialTimes from 27. October 2020.
We traced the spread of the visualization across different web pages with the Google Vision API (with the help of the Memespector GUI): all ten results were found on Twitter.
The recontextualization on Twitter was traced by following the practice of quote tweets: We retrieved all quote tweets of the original tweet containing the image with the help of the Python package Tweepy and the academic access to the full Twitter archive.
Quote tweets were visualized as a bees’ swarm, showing the quote tweet’s publication time and the number of retweets per quote tweet.
The selected case study was the web card with Meghan Markle’s photo. We based this decision on the image engagement metrics: it is the post with the highest engagement in the Facebook dataset. As we will go on to discuss, there is a high probability that this generic visual was the most engaging image because it includes a prompt inviting social media users to interact with the content.
We used the “GV_Web_PagesWithFullMatchingImages” column from the Google Vision API results, where we collected 10 URLs.
Additionally, we reverse image searched them with Google Lens and copied the top ten URLs using the identical image.
We reviewed all the 20 URLs from the previous two stops. We verified that the URLs linked to a post in which the image was used, and removed duplicates. After this process, we reached a total of 18 URLs for our analysis.
When analyzing the 500 most retweeted images published by the Financial Times Twitter through PixPlot, we identify seven different clusters. Thereafter we summarized them as five clusters: data visualizations; photographs; quote cards; breaking news cards; front pages. It becomes clear that Financial Times makes use of corporate design and image templates to draw attention to recently published articles. Oftentimes, images in tweets are identical to the images on the FT website, which suggests that they are transferred from the website to Twitter by default when publishing the URL of an article.
Data visualizations include more complex visualizations, but also simplistic graphs that meet the definition of generic visuals. Photographs are mostly stock photos used to illustrate an article. Quote cards are monochrome images containing short and exchangeable quotes or a single fact. Breaking News cards consist of an identical breaking news alert and Front pages are title pages of the current Financial Times print edition. The most retweeted image category is data visualizations, while photographs are liked the most in our data sample.
After analyzing the 700 photos from the six Reach Group’s local and national newspapers’ Facebook accounts in PicArrange, a few clusters with distinctive visual features emerged immediately.
First of all, two clusters of photographs of people stood out in the data set (marked by pink and orange in figure 2). While the pink cluster is mostly full- or half-body photographs of people, the orange cluster is mostly close-up images of people’s faces. Some of the photos in the orange cluster are similar to mugshots.
Another prominent feature that emerged in our analysis is that many of the images have text overlays, which we decided to call “web cards” (see the red and blue clusters in figure 2). Some of the web cards consist of news images and text. The message usually offers context to the story, presents a quote, or poses a question to social media users. Other web cards are text on top of a solid-color background. The text in those text-only web cards is usually a question that concerns everyday topics and ‘light’ contents (for example: “What’s a job you think deserves higher pay?”). We believe the questions invite social media users to respond. Therefore the web cards act as engagement bait, to boost the media’s social engagement numbers.
Lastly, the front pages of the newspapers form a cluster in our data set (see the green cluster in figure 2).
Through stacking the images in a category, some visual characteristics within the category can be observed (see figure 4). In the web cards with image category, images tend to appear on the bottom or in the top right corner of the web cards, while text can appear at any position. In the web cards with text category, most of the web cards have a thick border around the webcard, and the font size appears to be bigger than the previous category. It can also be observed that in this category text is usually centrally aligned.
Figure 4: Image stacks of web cards with image (left) and web cards with text (right).
Based on our preliminary analysis of the visual features emerging from our datasets from Twitter and Facebook, we further developed a more comprehensive categorization and description of the visual genres in generic journalistic imagery (see table 2).A significant observation is that a portion of visuals are created to tell a story in its own right on the social platform, while some (that might resemble ‘ads’ for stories) require the audience to go to the news website to get the full story.
This category denotes visuals that are previews of the frontpage of the paper version of the new outlet. These could be categorized as generic (it is a recurring visual, likely based on a graphical template).On the other side they require a substantial amount of work in that it is a summary of the collective journalistic effort of the given day. The example here is from the Financial Times Twitter. Daily Mirror, Sunday Express, Daily Express, The Daily Star and Sunday People post the front page on their Facebook page as well. Like the Exclusive card and the Breaking news alert (see further below), they point to the media website - away from the social media.
Photographs (no text)
Photographs denote photos used in social media posts. In this project we did not specifically identify photos that are part of a stock photography service or database. This would be relevant as well as going deeper into the context of the photo use. And we are thinking if this cluster needs a subcluster. In terms of the subject on the photos, we observe that the news outlets very often depict portraits, both of “regular” people (which could perhaps be viewed as a parallel to the journalistic genre of a case story). Other portraits are usually of politicians and celebrities. In this category we also see mugshots.
Breaking news alert
We consider Breaking news alerts all graphic posts containing the word breaking news. These images are used to draw the attention of users when an event considered significant occurs. These can be classified as quite generic, meaning they are easily made and are likely part of an image database that journalists can access as needed - for Financial Times, this image is reoccuring. This flagging of breaking news along with the ‘Exclusive alert’ category are meant to attract readers to the news site. Interestingly, the two are similar to the news values of exclusivity and timeliness, but here repurposed as a sort of advertisement for the given content and an overall positioning of the media outlet.
We consider Exclusive alert all graphic images containing the word exclusive. These images are used to attract users' attention when the news media has come into possession of exclusive news. It can also be interpreted (as is the case with the breaking news alert) an advertisement for the newspaper and positioning in the media market.
Cartoons or drawings
This category denotes drawings or cartoons from the newspaper, shared on social media. Like the front page category, these may be considered a recontextualization of the paper version. In terms of effort, these would require a lot of resources to make.
Infographics denote all images of data depictions/visualizations. High level of generity, it seems: In the dataset, we see several bar graphs and line graphs.
The news swipe could be viewed as a news genre format in itself, facilitated, hosted and architectured by social media companies, especially Instagram/Facebook in this research. The news swipe allows users to get an overview of a news story without leaving the social media platform.
We consider web cards or social cards to be all images composed by a basic image modified by inserting words or sentences and logos. In terms of content, they often feature quotes from a current news article and, for certain outlets, memes. Quote cards, news-you-can-use cards and question cards (see below) may be considered subgenres to this.
Quote cards contain a quotation or single fact from a well-known person or source: Current and historical source material appeared in our dataset.
Question cards contain calls for debate/reflection. The aim is likely to create more engagement with followers or trying to understand their sentiment about a specific topic.
The news-you-can-use card can be viewed as an appropriation of the news story genre made for social media. In terms of content, it contains information useful for everyday (public) life. Visually it can be considered quite generic as it is most often text on colored background. Like it is the case with News Swipe, the user gets ‘the story’ without having to leave the platform.
Table 2: Observations of visual genres in generic journalistic imagery
Previews of the paper version frontpage
Photos used in social media posts (no logo or text)
*regular people, *celebrities, *politicians, *mugshots
Composition of a photo and
*words *sentences *logos
*message regarding the news outlet
Breaking news alert
Graphic posts containing the words breaking news
Graphic posts containing the word exclusive
*including (generic) photo
Artwork originating in the paper (non-generic)
Graphic post with quotes or single fact, no images
*quote from news story
*quote from other source
*single statistical fact
information useful for everyday (public) life, text on colored background
Generic data visualizations
The news article format appropriated to social media
Needs further exploration
Cards with a question on colored background
*call for debate
*call for shout-outs
*call for experience sharing
Table 2: Classification: Observations of visual genres in generic journalistic imagery.
Figure 5: Network analysis showing domains on which the 500 most retweeted images from the Financial Times Twitter account re-appear.
The 500 most popular images from the FT Twitter account were mostly found on social media platforms, e.g. fbsbx.com and twimg.com, which are image hosting sites from Facebook and Twitter. Moreover, images were found on stock photo databases, like Getty Images or alamy.com (green), on newspaper title page archives (blue and pink), quote collections (pink), tabloid websites (green) and Amazon (pink).
Figure 6: Network analysis showing domain names on which the 700 most engaged with images from the selected Tabloid of Reach Group accounts re-appear.
Similar to the findings on Twitter, the 700 most engaged with images from the selected Tabloid from Reach Group accounts were mostly found on the social media image hosting sites, as reflected by the font size of the label in the Gephi network. Moreover, we can identify some clusters: the pictures coming from the Facebook hosting domain (fbsbx.com) appear also on Clickbait sites (green cluster), whereas pictures found on the twitter domain (twimg.com) appear on newspaper websites (violet cluster). The orange cluster consists of mostly fashion and style magazines and tabloids.
For a further inspection of the recontextualisation of the image on the platform, we chose to retrieve all quote tweets of the original tweet containing the image with the help of the Python package Tweepy for retrieving Twitter data. Generally, there are three different ways to circulate and recontextualize tweets: retweets, quote tweets and practices that circumvent the platform’s “grammar of action” (Agre, 1994; Paßmann and Gerlitz, 2014): sharing an image without using the predefined features of Twitter, e.g. by embedding a screenshot into a tweet. The latter cannot be traced with currently existing digital or visual methods, although these might be the most interesting objects to study recontextualization across the platform, as users can fully detach the image from its original context.
Because of this limitation, we chose to trace the image by following the quote tweets of the original tweet. Unlike retweets, quote tweets allow commenting on the reposted tweet and thus, a more profound recontextualisation. For the selected data visualization we retrieved 1136 unique quote tweets (bees swarm quote tweets).
We created a bees’ swarm diagram showing the occurrence of quote tweets per time and their number of retweets, and then zoomed into the most retweeted quote tweets.
Figure 7: Bees’ swarm diagram showing the publication of quote tweets of our example image over time and their number of retweets.
Figure 8: Screenshot of a quote tweet of our example image.
Figure 9: Screenshot of another quote tweet of our example image.
The examples show how supporters of both candidates used the image to mobilize the public for the upcoming election. The visual of FT is turned into a call to action. The account and the assumed audience add to the context: when a supporter of Trump publishes a quote tweet, only adding “Texas” to it, he addresses fellow supporters in Texas.
In another example, a Japanese journalist comments on the image in Japanese “In a Texas poll, Trump is just ahead and has a higher approval rating than Biden. It's a close battle.”. This example transfers the data visualization to a non-English-speaking audience, but at the same time adds a new interpretation to the visualization: While the Financial Times describes Trump and Biden as being “neck to neck”, the journalist emphasizes Trump’s advantage. For non-English-speaking followers, the interpretation of the Financial Times moves in the background.
Figure 10: The most engaged picture from the Facebook Dataset.
As a case study, we chose a web card with a photo of Meghan Markle. An examination of the publication date of the 18 posts reveals that the photo was used in various stories between 2018 and 2022, with 13 posts published between 2020 and 2021. This concentration in the two years could be a result of the press coverage of the decisions of Prince Harry and Markle to leave the Royal Family.
We then coded each post based on the themes of the news articles to understand in which contexts the generic visual was used. The results show that although the photo was used in the context of Markle’s new book in the most engaged Facebook post, the photo was recontextualized in themes such as Markle’s mother, her hair, her miscarriage and bullying accusations against her.
Through means of the Google Lens tool, we traced the source of the image: it was originally a news photo taken by Chris Jackson during Prince Harry and Meghan Markel’s visit to a science park in Northern Ireland. The image was first uploaded to the stock photo agency ‘Getty Images’ on 23 March 2018. Within our dataset, the first news article that recontextualizes this photo was about Markle’s face exercises.Figure 11: A screenshot of Getty Images on which the photo of Meghan Markle was first published.
Figure 12: The timeline of news articles that recontextualized the generic visual of Meghan Markle.
Since the photo was taken while Markle visited entrepreneurs in Northern Ireland in March 2018, it has been recontextualized in many different news stories on various topics throughout the next four years. Moreover, our dataset shows that the image was repeatedly recontextualized around 2021. The 18 news articles are neither about Markle’s visit to Northern Ireland (the image’s original context) nor her new book (the most engaged Facebook post’s context).
Figure 13: The data visualization that we chose as an example for our case study on recontextualisation, showing results of a Texas election poll before the US presidential election in 2020.
Our case study shows that the data visualization published on Twitter remained on the platform and circulated with and without the help of the platform’s grammar of action. When tracing the circulation of all top 500 images from the FT Twitter account, we found that this cannot be generalized: Many images were found on other websites and platforms. Circulation might affect stock photos more than data visualizations, as locations like tabloid websites and stock photo databases suggest. Data visualizations are produced by FT and cannot be found on these sites. However, Twitter data shows that FT uses the social media management tool SocialFlow, which allows simultaneous cross-posting on different platforms. Thus visuals, including data visualizations, might also be published on other social media channels.
By following quote tweets of our data visualization example, we mostly found conclusions and interpretations added to the original context, but no detachment from the context. However, quote tweets only allow detaching images from their origin to a small extent, as the original tweet is always cited. In our example of the tweet by the Japanese journalist, detachment is possible for a non-English-speaking audience that cannot follow the original description and relies on the author's comments.
For analyzing how publics are assembled by generic visuals on Twitter, it’s vital to include the different practices of spreading them - and the ways they enable recontextualisation. Context on Twitter doesn’t only include new or different information or interpretations added to them in the tweet’s text, but also platform features, like the display of the number of likes and retweets, or the account name. These similarly add further information to a tweet. A quote tweet of our example data visualization by a designated Trump supporter changes the visual in a different way than a quote tweet by the account of “Progress Texas”.
Figure 14: A screenshot of Getty Images that displays other photographic resources from the visit of Prince Harry and Megan Markle.
RQ1: What are the most engaged with generic visuals on social media news accounts between 2020 and 2022?
The types of generic visuals on the social media accounts of Reach Media and Financial Times are common to both platforms – Facebook and Twitter. These types were identified through the use of visual analysis tools and later synthesized with an in-depth categorization. The most retweeted type of generic visuals was data visualizations, while stock photos gained the highest average number of likes per photo on Facebook.
However, Facebook and Twitter presented differences regarding the engagement of each category of generic visuals. In the case of Facebook, the web card category was the generic visual type that presented the most engagement. In this type of visual, a question or prompt for interaction within the platform's grammars of action (Agre, 1994) –such as like, reaction, or comment– is always present. Within these categories, the most engaging format is questions to the audience. Whereas in Marketing Studies a “Call to action” is generally a short exhortation to click on a button in order to purchase goods on an e-market, it can be debated whether this kind of generic visual could also fit into this category, for it exhorts users to engage with the post. This type of generic visual signal to tactics that prevent algorithmic invisibility (Bucher, 2012). By generating engagement that boosts the reach of the post on the EdgeRank algorithm these news accounts on Facebook strive to remain relevant on social media.
Similar ambitions can also be observed on Twitter: FT mostly uses images to announce recently published articles. Embedding the article image into the tweet appears to be a default setting of FT’s website, which can be seen as a means to prevent algorithmic invisibility: It makes the tweet appear more prominent in the Twitter feed of users, which affects click rates of the article. This partly automated practice contributes to the fact that both stock photos and simple data visualizations become “increasingly ubiquitous” (Aiello et al., 2022, p. 327).
RQ2: How do generic visuals in news circulate? Are they recontextualised?
Our two case studies, the line graph published by the FT Twitter account and Megan Markle’s photo published on Facebook, reveal that generic visuals circulate differently. While the stock photo was found on different websites, the Twitter data visualization remained on the platform. Twitter’s grammar of action (Agre, 1994) suggests circulating the image with all its context markers, such as the original tweet text, the source account, and the publication time. However, users usually work around this grammar and spread images via screenshots that enable excluding these markers or editing the photo. Twitter’s circulation features, namely retweet and quote tweets, recontextualize in their own specific ways, but in most cases, do not detach the tweet from the original context. Apart from practices that explicitly attempt to spread an image, the platform further circulates images algorithmically, by showing them as liked tweets in news feeds and Twitter trends. This regime of visibility further contributes to the public’s assemblage through images on social media platforms.
Regarding the circulation of generic visuals in the news, we observe that in the case of Megan Markle’s photo, the source was indeed Getty Images, a stock photo agency. The image started circulating only five days after the image was posted. The photograph contained few contextualization markers in its composition given its blurred background. From the outset, the official visit was recontextualized with news covering distinct aspects of Megan’s life. Three years later the image started recirculating in a myriad of news reports on different themes, one of these being the image that we identified as one of the most-engaged images on Facebook. We conclude that the photograph category that includes media personalities needs from the start to have a composition that de-contextualizes the image. In this way, these images are prone to be re-contextualized by different media outlets.
Further research could be done in investigating possible links between the different modes of circulation of generic visuals - in-platform data visualizations for Twitter and across platform stock photos for Facebook - and their engagement rates. In this direction, the relationship among engaging visuals on Facebook and clickbait sites could give significant insights for an analysis and a critique of the process of public assemblage.
Aiello G, Kennedy H, Anderson CW, Mørk Røstvik C. (2022) “‘Generic visuals’ of Covid-19 in the news: Invoking banal belonging through symbolic reiteration”, International Journal of Cultural Studies; 25 (3-4), 309-330. doi:10.1177/13678779211061415
Agre, Philip E. (1994) “Surveillance and Capture – Two Models of Privacy”, The Information Society; 10 (2), pp. 101-127.
Bucher T. (2012) “Want to be on the top? Algorithmic power and the threat of invisibility on Facebook”, New Media & Society; 14 (7), 1164-1180. doi:10.1177/1461444812440159
Anthony McCosker & Rowan Wilken (2014) “Rethinking ‘big data’ as visual knowledge: the sublime and the diagrammatic in data visualisation”, Visual Studies; 29 (2), 155-164, DOI: 10.1080/1472586X.2014.887268
Paßmann, Johannes and Gerlitz, Carolin (2014) “ ‘Good‘ platform-political reasons for ‚bad‘ platform-data. Zur sozio-technischen Geschichte der Plattformaktivitäten Fav, Retweet und Like”, www.medialekontrolle.de; 3.1, URL: https://www.medialekontrolle.de/beitrage/good-platform-political-reasons-for-bad-platform-data-zur-sozio-technischen-geschichte-der-plattformaktivitaeten/ (last accessed: 20.07.2022)
Zelizer, B (2010) “Journalism, Memory, and the Voice of the Visual”, About to Die: How News Images Move the Public; 1-27, Oxford University Press.
BuzzSumo - https://buzzsumo.com
Crowdtangle - https://www.crowdtangle.com
Gephi - https://gephi.org
Google Vision API - https://cloud.google.com/vision
Memspector - https://github.com/jason-chao/memespector-gui
PicArrange - https://arxiv.org/abs/2111.13363
PixPlot - https://dhlab.yale.edu/projects/pixplot/
Tweepy - https://www.tweepy.org
|png||pasted image 0.png||manage||1 MB||21 Oct 2019 - 13:23||EmilieDeKeulenaar|
|png||pasted image 2.png||manage||1 MB||21 Oct 2019 - 13:32||EmilieDeKeulenaar|
|png||unnamed-2.png||manage||12 K||21 Oct 2019 - 13:34||EmilieDeKeulenaar|
|png||unnamed-3.png||manage||11 K||21 Oct 2019 - 13:34||EmilieDeKeulenaar|