Sage Journals: Discover world-class research

Abstract

The sharing of internet memes on online social media is increasingly popular form of expressing opinions and complex sentiments in an easily understood image. Memes have been found to be an effective marketing tool if used appropriately. Marketers need to be aware when a meme's use has become saturated and its value is depreciating. As with many consumer uses of online social media, meme sharing is described as market for attention that can display fad cycle dynamics. This analysis models the dynamics of meme attention and estimates these dynamics from a panel of memes scraped from Reddit meme forums and classified using a machine learning algorithm. Empirical analysis reveals patterns consistent with random shocks to meme entertainment value carrying over to the quality and quantity of subsequent expressions of the meme. Estimates indicate that a temporary elevation in attention dissipates within just one to two weeks. However, meme saturation then leads to less spread of the meme. To be effective, meme marketers must remain attentive to where the meme they intend to use is within the fad cycle.

Keywords

meme internet machine learning fads attention

Internet memes are fast becoming part of the marketing toolkit (Tama-Rutigliano 2018). Their growing incorporation into brand campaigns has paralleled the growing popularity of meme sharing in social media (Razzaq, Shao, and Quach 2023). They have been shown to enhance brand recall and brand engagement, especially for millennial-targeted products (Malodia et al. 2022; Razzaq, Shao, and Quach 2024). While appropriate meme marketing can yield impressive returns, practitioners point out that inappropriate use of memes entails risks. The main disadvantages are that they can be open to misinterpretation, they may not align with the brand image, their use entails some relinquishing of control, they can be used excessively, and they can lose their relevance as the current trend represented in the meme fades (GeeksforGeeks 2024). The last two points relate to the fleeting fad nature of the particular meme. I study the fad dynamics of internet memes to estimate how quickly a meme's attractiveness fades. These fad cycles have implications for the structure of the meme segment of the digital marketing industry.

Fads represent short-lived enthusiasm or trends that become popular for a brief period, often due to social influence, before losing their popularity. Some generational fads have included Hula-Hoops, Pet Rocks, Beanie Babies, and fidget spinners. Systematic analyses of fads have been difficult because they emerge at irregular intervals, their origins are hard to discern, and their features appear idiosyncratic. Fads mediated by online social media, such as planking, unboxing videos, TikTok stitching, and the Ice Bucket Challenge, appear to emerge more regularly, but their contexts may be too dissimilar to discern systematic patterns. Internet memes, however, both consistently rise and fall in popularity and share a similar enough context for systematic study. I first model the waxing and waning of internet memes as an outcome of a market for attention with specific implications for their dynamics. I then estimate the parameters of this model from a panel of data on meme appreciation and usage. Highly regarded meme expressions increase the meme's usage, which persists for about two weeks. In addition, consistent with meme usage motivated by attention seeking, internet memes appear to compete with one another.

The popularity of internet meme sharing through online social media has increased dramatically in just a few years. Memes were initially defined as an element of culture that can be passed on through imitation to another individual by nongenetic means (Dawkins 1976). However, the idea of a meme has been appropriated by social media users to describe a specific type of virally spreading idea. Standardization of internet memes has emerged as a combination of an image and a text caption overlaid onto the image. In social media, the image, usually from popular culture, connotes a shared specific sentiment, and the original text added by the creator indicates that the sentiment applies in another context. Also, these expressions usually attempt humor. Different internet meme generator sites (e.g., Imgflip, Kapwing, Make a Meme) have inventories of the more popular meme images, which any user can customize with their own text overlay. Most meme expressions that internet users see are copies of internet memes that friends have come across and inserted into their online social media presence. In this way, the most popular expressions of internet memes can diffuse through a social network quite quickly and broadly.

The popularity of internet memes has not gone unnoticed by marketers. Memes tend to be timely commentaries on current events that brands can embrace to signal that they are at the forefront of popular culture (Razzaq, Shao, and Quach 2023). The democratic nature of the creation of memes generates many perspectives on the same event, and the ease of spreading amplifies perspectives that might not otherwise find voice (Shifman 2014). The use of memes can indicate that the user is on the cutting edge of cultural awareness and increasingly associates the user with a countercultural aura. Marketers can heighten consumer engagement by exploiting commentary on culture and associating with counterculture (Heath and Potter 2004, 2005). However, when the market messaging appears to be less legitimate, memes can generate negative sentiment toward the subject matter. Overuse and inappropriate use of memes in marketing can destroy the brand value. While memes naturally occur in cycles, meme marketing can amplify existing fad cycle dynamics.

The fad behavior of internet memes can manifest itself in multiple ways. An internet meme expression will usually comment on a current aspect of popular culture (e.g., a televised performance or the release of a new movie). Others seeing an original expression can add their own insight on the event by creating another meme expression as part of a public conversation. Alternatively, a particularly humorous insight may lead to further expressions that continue the joke with only vague references to the original event, similar to an improvisational troupe continuing a scene. Finally, some of the luster of an exceptional expression may rub off onto the meme image itself. Subsequent meme creators may be more likely to use the underlying image on expressions unrelated to the original event or insight. The event, the insight, and the image can each exhibit fad behavior in memes. This study focuses on measuring fads in image use because this aspect is more easily measurable across internet meme expressions.

The fad nature of internet memes has strategic implications for their use in marketing messages. One strategy is for brand managers to appeal to current popular culture by embedding internet memes into traditional television or print marketing messages. As with “brand coolness,” timeliness of the meme is key to its success in marketing (Warren et al. 2019). In the 2010s, both Wendy's and McDonald's stumbled with memes in television ads when they referenced outdated memes. Use of outdated memes by the uninformed in marketing has itself become a meme, “When Gen-Z Writes the Marketing Script” (Booth 2024). Another common strategy for internet meme marketing is to create a meme and post it online in the hope that viewers will engage with it enough to disseminate copies of the image through their own social media (Roache 2019). As part of the digital marketing ecosystem, a specific meme marketing segment has evolved and includes services such as creating meme content, setting campaign strategies, and analyzing the resulting engagement metrics. Advertising agencies often contract with meme influencers to place brands in new memes and new versions of existing memes (Kar 2020).¹ This meme marketing segment's comparative advantage at these tasks stems from its investment in understanding how memes respond to the fast-paced changes in popular culture.

Past Research on Memes

An internet meme, the familiar image onto which unique user-generated text is overlaid, succinctly evokes a subtle and complex idea. Analogues from a recent preinternet era might be common setups to jokes (“A guy walks into a bar …”), catchy quotes from movies (“These aren’t the droids you’re looking for”), or famous lines from classic literary works (“No man is an island”). These iconic quotes have become common shorthand phrases that also succinctly evoke a sentiment that would otherwise be difficult to express. An internet meme implementation applies an evocative image template to a new context. It implies that the sentiment in the image applies to the new context. In a more rigid convention that has emerged, the internet meme is a static image, while a meme expression is the image with superimposed text that refers to the current context to which the image applies. For example, the oft-used “Distracted Boyfriend” meme provides a reference to someone getting caught displaying a surreptitious desire. By supplying original text, the creator extends the context of the meme to a new application. For example, Figure 1 depicts the idea of student procrastination with the girlfriend labeled “studying for exams” while the woman receiving the attention is labeled “browsing memes.”

Figure 1.
Static Meme Example.

Creating and sharing internet memes are forms of attention or status seeking that has been found among users of Twitter, now known as X (Iyer and Katona 2016). Content sharing can generate intrinsic utility for the sharer or can satisfy a desire to augment the sharer's social image. Status seekers attempt to enhance their status via signaling with popular internet memes (Nissenbaum and Shifman 2017). The communications literature has applied signaling theory to social network sites to study collective action (Ashuri and Bar-Ilan 2017), social network scale (Donath 2007), social grooming styles (Lin 2019), and the role of influencers (Wasike 2022). In a field experiment of Twitter users, seeking an enhanced social image or status was found to be the largest motive for most users (Toubia and Stephen 2013). Relatedly, competition for Twitter followers accounts for one-third of the content generated by players in a national women's soccer league (Rossi and Rubera 2021). Competition for attention, as well as the social network structure in which a meme implementation originates, explain much of the heterogeneity of Twitter meme popularity and persistence (Weng et al. 2012). Even academic publishing has been modeled as a market for attention (Spitzberg 2018) with the finding that more novel research generates more citations (Chai and Menon 2019). While the particulars of meme sharing differ from either tweeting or academic citations, meme sharing may not be so different from a signaling perspective as it shares the motive of attracting attention or gaining social status.

Memes have been identified in various online formats. Meme implementations alluding to newsworthy events have been identified from shared videos (Xie et al. 2011). Image processing and optical character recognition have been used to develop a method of classifying the sentiment in a meme as pro- or antigovernment (Amalia et al. 2018). Twitter memes have been identified to study propagation of content through social networks (Shabunina and Pasi 2018; Weng et al. 2012). Based on similarity in image, name, and embedded text, static image meme implementations have been classified as competitors or collaborators of other memes (Coscia 2013).

Meme success can depend on the structure of the social network in which a meme implementation is shared. Social interactions on YouTube are influential in determining which videos become successful and how successful they become. These interactions are mediated by both user homophily and direction guidance from the platform (Susarla, Oh, and Tan 2012). Meme uniqueness is related to the social network structure of the meme's users (Segev et al. 2015). Meme attributes and dynamic network structure can be used to evaluate user behavior (He, Zheng, and Zeng (2019). Regularities in sharing can be detected from a graph model connecting people and content (Xie et al. 2011). For example, content that will eventually have a long lifespan can often be predicted within a day of initial posting to social media. However, independent of social network structure, the attributes of the memes themselves, such as their uniqueness from existing memes, play a role in their success (Coscia 2014).

Research on internet memes has largely focused on the determinants of meme success in becoming viral in social networks. Spitzberg (2014) and Spitzberg (2021) synthesize multiple theories into a model of meme diffusion, and Gleeson et al. (2014) show how competition between memes for attention generates popularity measures that can be described by a critical branching process in sharing. This predicts power-law distribution of popularity with heavy tails. A goal of much of this research is to aid platform operators’ efforts to increase network engagement. He, Zheng, and Zeng (2016) overcome impractical assumptions and difficulty characterizing dynamic information to develop a model-free scheme to rank meme popularity in online social media and social networks. Bonchi, Castillo, and Ienco (2013) develop heuristics for a platform selecting among memes to promote maximum virality. Guadagno et al. (2013) show that a stronger emotional response to a meme increases its propensity to go viral.

Social uses of internet technology are well suited for the study of fad behavior. The data are abundant, enabling phenomena to be tracked easily. Marketing professionals have studied internet use to help determine what makes a campaign go viral. However, studies of internet virality tend to consider a “single-humped” life cycle of advertising campaign popularity. Memes persist long enough to have multiple ebbs and flows in popularity more consistent with fads. The sharing of meme images is not the only social phenomenon to emerge from internet usage. Similar faddish behavior may apply to podcasts, emotes, video streams, or whatever will become the next big thing to emerge. This study provides methods and insights that might aid in their study as well.

I develop a simple model of a market for attention to describe the dynamics of meme popularity as fads. While most of the existing literature analyzing memes has been within the communications studies or computer science domains, Schlaile (2021) uses the lens of evolutionary economics. I explore fads in internet memes through the lens of market dynamics. The meme consumers are assumed to respond to the entertainment value of memes, which can persist over time due to improved meme quality. A shock to meme quality increases the entertainment value. The meme producers exploit a temporary increase in entertainment value to produce more meme expressions. Eventually, diminishing marginal utility causes the entertainment value to revert to preshock levels. The dynamics resemble a fad and lead to testable implications.

A sample of meme-related posts was collected from Reddit subforums dedicated to meme culture. A machine learning algorithm was developed to match the image of a meme post to popular meme images. Matches are never perfect, because the text overlay always differs across posts. However, the algorithm correctly classified nearly 70,000 posts into 533 existing meme templates, resulting in a separate time series for various cross-sections of memes. The resulting panel of meme expressions includes information on the underlying meme, its authorship, a popularity score, and the number of comments made about the expression.

Figures 2 and 3 provide motivation for further analyses of internet memes as fads.² Figure 2 compares the fraction of meme expressions posted in each quarter to the number in quarters just before and after a meme image's peak usage. The rapid rise and sudden crash of a meme's usage is consistent with a typical fad life cycle. However, the life cycles of most products also feature a rise and the decline in popularity. Figure 3 provides evidence more consistent with fad behavior. It shows that the appeal of a meme precedes its peak usage. An interpretation of Figure 3 is that meme creators jump on the bandwagon while a meme is gaining broad appeal, which in turn leads to meme overuse that eventually diminishes the meme's appeal. In the next section, a model of a market for attention is developed to represent this general fad dynamic on a shorter time scale. The theoretical model provides the intuition for an empirical analysis of the dynamics of meme appeal and proliferation presented in the following sections. As a robustness check, a market for attention is further implied by popular memes substituting for contemporaneously less popular memes.

Figure 2.
Life Cycle of Meme Usage.

Figure 3.
Appeal of Memes over Their Life Cycle.

The Market for Memes

One characteristic of fads is intertemporal persistence (Bikhchandani, Hirshleifer, and Welch 1992). In the case of internet memes, persistence is generated by the randomness in meme quality in one meme expression affecting demand and supply of subsequent expressions. Variation in meme popularity may emanate from sources like those for other cultural goods such as fashion or slang. Wearing the latest clothing design or adopting the latest vernacular is expected to generate status within one's social circle. Likewise, creating, or even just sharing, a clever meme expression also confers social status. While the popularity of the new element of culture can be highly uncertain, those that succeed will persist in popularity over some period but eventually will die out upon being replaced by another design, phrase, or meme. Here, this persistence is represented as a stochastic autoregressive process linking internet meme popularity across periods. Memes are consumed primarily as entertainment, though they are also increasingly becoming a source of information (Scanlon 2020). A friend on social media may reply to a comment with an image from popular culture that represents a complex social condition. This shared cultural reference may inform the recipient of the replier's sentiment more concisely than “a thousand words.” Alternatively, one might enjoy a post of a yet unfamiliar image that is annotated humorously or perceptively. Images shared enough to gain traction may become part of the canon of popular memes. As with other cultural goods, familiarity with the medium from past exposure can increase the marginal utility derived in each future instance (Brito and Barros 2005; Stigler and Becker 1977). In this case, the time spent consuming meme expressions tends to enhance the consumer's future enjoyment of subsequent expressions of the meme. The primary cost of consumption is not pecuniary but is the opportunity cost of one's time. It may only take a few seconds to read the post, and perhaps to rate it, comment on it, or share it via a personal channel or social media presence such as email, Facebook, or Instagram. Eventually, diminishing marginal utility determines how much time one spends consuming internet memes. Users on Reddit have been found to experience cognitive fatigue, suggesting that time costs are increasing (Glenski, Pennycuff, and Weninger 2017).

These social interactions provide a foundation for a formal model of memes as fads with implications for the available information on the evolution of meme usage and appeal of internet memes over time. These data include both how often variants of meme m are posted to Reddit in time t, the number of meme expressions or e_m,t, and their average appeal to meme users, measured as the average Reddit score, s_m,t. Ideally, the data would include a measure of how much attention a meme generates, such as how often a new meme expression is copied and shared in various forms of social media. Broad attention is assumed to be the meme creator's goal and is the metric most important to meme advertisers. Unfortunately, greater attention must be inferred from its score, s_m,t. When consumers of memes consider a post to be more insightful, clever, and timely (i.e., more appealing), they will be more likely to both “like” it and share it. This leads to posts that are liked by more people tending to receive higher Reddit scores and attract more attention. For modeling purposes, it is useful to think of e_m,t as how many expressions are produced by meme creators and s_m,t as their quality, a measure related to how often these expressions are shared within the social media of meme consumers. The model applies attention-seeking motives of both meme consumers and meme producers to uncover the likely dynamics of meme consumption and production.

On the consumer side, viewers of memes seek greater attention by sharing with their friends the more insightful, clever, and timely meme expressions, that is, those with higher s_m,t. The dynamics of meme quality around a steady state, ${\bar{s}}_{m}$ , can derive from three factors represented as $s_{m, t} = {\bar{s}}_{m} + ρ s_{m, t - 1} + δ e_{m, t - 1} + μ_{m, t} .$ (1)

Random shocks as flashes of creativity can generate different amounts of entertainment value, μ_m,t, modeled as an i.i.d. random variable with mean zero. The mechanisms for these idiosyncratic shocks to cause persistence in meme quality stem from meme consumer assumptions regarding diminishing marginal utility in expression use and in meme spread. For the second to last term, increased use of a meme, higher e_m,t−1, can cause diminishing marginal utility meme proliferation across expressions, implying δ < 0. If the meme was used in more expressions in the last period, it will lose its entertainment value, and hence its appeal, in the current period. Finally, overexposure of a meme through an increase in likes and shares can reduce its appeal in enhancing social status in subsequent shares, leading to ρ < 0. Overexposure causes a decrease in the propensity for the meme to spread into the future.

On the producer side, the model proposes both attention seeking and fad behavior as driving meme creation. Attention is generated by developing expressions of a meme that are timely and clever (Rui and Whinston 2012). A meme post that provides early commentary on a current event or one that does so in a particularly innovative or humorous way will be shared more often and will capture a larger audience. At the same time, a creator can expect a larger audience by using a meme that is currently popular rather than one whose popularity is waning. As a result, the number expressions of a meme, e_m,t, are expected to increase with its recent Reddit score, s_m,t. In addition, a central feature of fads is that they persist over time. Fad behavior is described as simply using a meme because others are using the meme. Evidence of this would be the number of expressions, e_m,t, being positively correlated over time. These considerations lead to the law of motion: $e_{m, t} = {\bar{e}}_{m} + ϕ s_{m, t - 1} + θ e_{m, t - 1} + ν_{m, t} .$ (2)

The steady-state number of expressions, ${\bar{e}}_{m}$ , can be affected by recent past experience. Evidence of attention seeking would be consistent with ϕ > 0, while evidence of fad behavior would be consistent with θ > 0. Eventually, fads die out. This imposes the restriction that 0 < θ < 1. While this is modeled as a simple AR(1) process, the actual process could be more complex, spanning multiple periods with an immediate amplification effect, θ₁ > 1, so long as it returns to a steady state, Σθ_τ < 1.

Equations 1 and 2 describe laws of motion for meme expressions and their quality. Together, they represent a vector autoregression in which both the number and appeal of expressions are functions of past values of their number and appeal. Time dependency ultimately results from how consumers and producers respond to exogenous shocks, μ_m,t and ν_m,t. In the next period, a positive quality shock, μ_m,t, leads attention-seeking meme producers to use the meme more, ϕ > 0, but meme consumers to share less, ρ < 0. Increased use of the meme subsequently causes diminishing marginal value of the meme, δ < 0, and increases the propensity for meme usage to persist into subsequent periods, θ > 0, as a fad. In this way, flashes in current creativity can generate specific future production and consumption dynamics.

One popularity-shifting mechanism could result from a high-status individual with a large follower base, such as a celebrity, sharing an expression of a meme. This would be an example of $ν_{m, t} ≫ 0$ . Other meme producers anticipate that subsequent creations using this meme will also generate a larger audience, perhaps by being noticed and shared by the original high-status individual. In the same way that an influential individual wearing a new piece of clothing or repeating a new slang term can increase these cultural items’ popularity, so too can high-status individuals’ use of a meme influence meme sharing. In this case, the item of clothing, slang term, or meme expression need not be intrinsically appreciated by followers; it need only be expected to confer status onto those who subsequently adopt it. Alternatively, another mechanism for increased meme demand could be the result of an objectively better execution of the meme expression. Some meme expressions are just innately more intriguing or more entertaining. The meme producer may have spontaneously and unexpectedly hit on a formula that works well. This would be $μ_{m, t} ≫ 0$ . Perhaps an existing meme may be found to provide a humorous new context to a fast-evolving current event. When this occurs, the combination of an existing image with a new text reference may suggest a direction for further developing the novelty into future meme expressions that sustains the entertainment value of the meme. To the extent that the later meme expressions can extend this innovation, at least temporarily, the attention this meme will garner will remain above the steady state. With either mechanism, high-status individual recognition or highly innovative posts, the increased attention will persist to subsequent expressions of the meme.

The modeled dynamic patterns are consistent with fads (Bikhchandani, Hirshleifer, and Welch 1992; Mercure 2018). Agents engage in various social activities both to satisfy their innate preferences and to engage with other agents, preferably of high status. With some regularity, participation in one activity attains elevated social status, leading more agents to seek attention by adopting the now elevated activity. The status of early adopters increases more than that of late adopters. Eventually, so many adopt the fad that it no longer elevates status and, instead, signals low status. Agents stop participating in the activity and move on to the next activity that confers high status, at least temporarily. In this way, status is conferred on those with the ability to quickly identify fads. This pattern may also apply to other social activities.

This stylized model of meme production and consumption has specific testable hypotheses regarding the dynamics of the number of meme expressions produced and the attention they generate. A meme expression that is shared or viewed more, that is, one that generates more attention, will tend to be rated higher by meme consumers. In what follows, a meme expression's potential for attention is represented by its Reddit score, a measure of user ratings. The assumption is that internet memes generally considered to demonstrate higher entertainment value will tend to be viewed and shared more often. However, in what follows, entertainment value, popularity, viewing, and sharing are not directly observed. Instead, users who share an internet meme, and hence make it more popular, are assumed to be more likely to also register their pleasure with it by “liking” it. If so, more users liking a meme expression is associated with more users sharing the meme expression and its creator garnering more attention.

A robustness check of attention-seeking behavior is to examine how use of one meme affects other memes. Engaging in one fad crowds out other potential fads. The chief cost of meme consumption is time. If a meme enjoys increased popularity, a larger audience will spend more time consuming it. The time this audience spends on the meme will be diverted from time spent on substitute endeavors, particularly consuming other memes. This will lead to lower ratings and fewer posts for other memes. Consequently, a meme will become temporarily less popular with increased popularity of other memes or an increase in the quality of other meme expressions. A meme will be rated lower with increased popularity of other memes.

Data and Meme Classification

The sample of meme expressions was taken from the meme forums on Reddit (https://www.reddit.com/). Reddit is a social news aggregation, web content rating, and discussion website. Users of the site can post content that is then voted up or down by other users. Posts are organized by subject into user-created boards, called “subreddits,” on thousands of different topics. Volunteer Reddit moderators enforce rules to discourage unwanted activity including abuse, harassment, and deception. Pertinent to this study, meme enthusiasts will post, vote on, and comment on memes in five subreddits: memes, wholesomememes, dankmemes, 2meirl4meirl, and MemeEconomy. Between 2012 and mid-2019, over three million posts were uploaded to these subreddits. While posts to many social networks may only be visible to friends (e.g., Facebook, LinkedIn) or to followers (e.g., Instagram, X), anyone who visits Reddit can see and react to any post. Although Reddit facilitates the scraping of only the most recent 6,000 posts, another site, Pushshift (https://pushshift.io/), records each Reddit post's unique identifier, which enables one to make a call to that post's information on the Reddit site. In that way, nearly all meme posts on Reddit can be scraped. Many posts’ content consists of the sharing of a mildly interesting, but ultimately idiosyncratic, smartphone screenshot. Other posts contain new potential memes drawn from recently transpiring events in popular culture. Some of these will become popular memes, but the vast majority will not. The remaining posts are meme expressions of already established memes.

Reddit posts were classified into these established memes based on a machine learning algorithm that matches posted images to a set of existing images. The convolutional neural network algorithm was trained with the set of popular meme images that existed on Imgflip (https://imgflip.com) in June 2019.³ Imgflip is a popular website that enables users to easily create meme expressions. Users can upload their own images, but most will select from the site's catalog of thousands of meme templates, to which they add their own text.⁴ The catalog is fluid but represents most of the popular meme images used so far. This catalog was the reference set against which Reddit posts were compared. Reddit posted images that had a 98% or higher match with one of the meme templates was classified as being an expression of that template's meme. Perfect matches are precluded by text overlays that differ across meme expressions. In total, 68,325 of the posts in the meme-related subreddit forums were classified into 842 distinct meme templates.

Matching of meme-related posts to templates was not perfect. First, only about 2% of all posts were matched. While this is low, recall that the vast majority of the Reddit posts are idiosyncratic user posts of new images. Others were new uses of popular culture images or images that represent less popular and less well-established memes. The match rate was higher prior to 2016 probably because more of these memes became popular then and were added to the set of templates. Still, the low match rate suggests the possibility of an unrepresentative sample. Figure 4 shows that the monthly volume of both matched and unmatched posts increased sharply over the most recent four years. This indicates a steadily growing popularity of online social interactions through memes. Second, the matching algorithm resulted in some false positives and more false negatives. Spot-checking revealed that about .18% of matched posts were classified incorrectly and about 1.4% of unmatched posts should have been classified. This implies a slight undercount of meme expressions classified to memes.

Figure 4.
Weekly Volume of Posts on Meme-Related Subreddits.

Various data elements were retrieved for each matched post. These include the date of the post, the meme to which it was matched, the title, the creator identifier, the post's Reddit score, and the number of comments. There are large differences in the popularity of memes. For example, 500 meme templates garnered 10 or fewer posts, while 50 memes were depicted in 200 or more posts. Table 1 lists the most popular memes and suggests that meme usage is distributed exponentially. While the creator identifier is unique, there is no demographic information available about each creator. About 57% of the identified creators post just once, but some post regularly, making the average number of posts per creator 2.9. Table 2 reports the number of creators by the number of posts they create. The post's title may provide some context for the post, but creators tend to use subtle references.

Table 1.
Most Popular Memes.

Rank Meme Name Number of Posts Percentage of Posts Ln(Reddit Score)

M SD

1 Drake Hotline Bling 9,121 13.3% 2.58 2.03

2 Expanding Brain 7,569 11.1% 2.64 2.30

3 Tuxedo Winnie The Pooh 3,786 5.5% 2.64 1.99

4 Surprised Pikachu 3,632 5.3% 2.48 2.19

5 Unsettled Tom 2,820 4.1% 2.35 1.87

6 Two Buttons 2,371 3.5% 2.50 1.96

7 Is This A Pigeon 2,300 3.4% 2.74 2.29

8 Distracted Boyfriend 2,140 3.1% 2.45 2.16

9 The Scroll Of Truth 1,873 2.7% 2.81 2.21

10 Left Exit 12 Off Ramp 1,520 2.2% 2.69 2.21

11 Change My Mind 1,404 2.1% 2.53 2.01

12 Me And The Boys 1,124 1.6% 2.53 1.59

13 Well Yes, But Actually No 977 1.4% 2.85 2.16

14 Blank Nut Button 964 1.4% 2.23 2.08

15 Boardroom Meeting Suggestion 901 1.3% 2.77 2.25

16 Futurama Fry 876 1.3% 2.08 1.92

17 American Chopper Argument 844 1.2% 3.21 2.40

18 Success Kid 835 1.2% 2.19 1.96

19 Hard To Swallow Pills 804 1.2% 2.99 2.38

20 First World Problems 800 1.2% 1.94 1.58

21 Bad Luck Brian 786 1.2% 2.05 1.79

22 Philosoraptor 573 .8% 1.98 2.01

23 Who Killed Hannibal 571 .8% 3.14 2.32

24 Car Salesman Slaps Hood 457 .7% 2.81 2.42

25 Confession Bear 453 .7% 2.26 1.96

26 Batman Slapping Robin 451 .7% 2.03 1.99

27 Who Would Win? 433 .6% 3.30 2.44

28 Running Away Balloon 419 .6% 2.84 1.84

29 Roll Safe Think About It 394 .6% 2.10 2.07

30 Sleeping Shaq 383 .6% 3.04 2.25

31 Scumbag Steve 378 .6% 2.02 1.84

32 10 Guy 372 .5% 1.76 1.63

33 One Does Not Simply 353 .5% 1.78 1.80

34 Be Like Bill 316 .5% 3.30 2.75

35 Good Guy Greg 311 .5% 2.47 1.92

36–511 All others 15,014 22.0% 2.11 1.83

Notes: Based on 68,325 posts to Reddit between August 7, 2011, and August 18, 2019.

Table 2.
The Number of Times Someone Posts a Meme to Reddit.

Number of Meme Posts per Creator Number of Creators Total Number of Posts

1 34,629 34,629

2 6,552 13,104

3 1,845 5,535

4 755 3,020

5 362 1,810

6 192 1,152

7 119 833

8 75 600

9 56 504

10 27 270

11 26 286

12 17 204

13–95 65 1,400

Notes: Based on 63,347 posts to Reddit between August 7, 2011, and August 18, 2019, with identifiable authorship.

The Reddit score is the difference between the up votes and down votes that the post receives. These values represent the eventual scores based on up votes and down votes occurring after the expression is posted. Generally, a post will receive more up votes when it is more timely or humorous. Reddit's algorithm for displaying posts to users favors both the most recently posted and those with higher current Reddit scores. This can lead to issues regarding the scale and observability of Reddit scores. Posts that become popular early are viewed by more users, causing their scores to be inflated. Algorithmic virality induced by early popularity could bias tests of meme persistence if early popularity is common for successive expressions of the same meme. In fact, the negative correlation of meme scores over time in Table 3 is inconsistent with successive expressions experiencing virality due to Reddit's display algorithm. The most popular posts can receive votes a week or more after posting, but even these will receive most of their votes on the day of their posting. Nearly all votes for less popular posts occur on the day of their posting.

Table 3.
Determinants of Average Meme Score.

(1) (2) (3) (4)

Number of Past Posts

One-day lag −.004 .001 .002 .002

(.003) (.005) (.005) (.005)

Two-day lag −.007 −.004 −.004

(.005) (.006) (.006)

Three-day lag −.004 −.004

(.005) (.006)

Four-day lag −.000

(.005)

Average Score of Past Posts

One-day lag −.016 −.017 −.017 −.017

(.007) (.007) (.007) (.007)

Two-day lag .002 .001 .001

(.008) (.008) (.008)

Three-day lag −.007 −.007

(.007) (.007)

Four-day lag −.010

(.009)

Observations 26,501 26,468 26,448 26,423

R² .012 .012 .012 .012

No. of memes 470 470 470 470

p < .05.

Notes: Specifications include fixed effects for the meme, the week of the year, and the day of the week. Observations are weighted by the overall number of posts of the meme.

Figure 5 depicts the average score over time for both posts that were matched with templates and those that were not. The score represents a combination of meme expression “quality” and the number of potential users voting on its quality. The rise in average score for nonmatched posts between 2016 and 2018 is probably indicative of the growing popularity of meme sharing in general rather than the emergence of higher-quality memes. Note that the average score of posts matched to memes began to fall in 2018. Imgflip identifies popular memes with some lag. Because of this, the popularity of what it identifies as “popular memes” is already waning. Most of these will eventually be replaced by newer memes with rising popularity that Imgflip has not yet identified as popular. In contrast, memes identified as popular, and so matched, are more likely to have already peaked in popularity, with their decline included within the sample period. This selection process means that toward the end of the sample period, more of the matched memes will be in the last stages of their life cycles, compared with unmatched memes. For our purposes, this means the matched sample of memes in the estimation is more likely to have complete life cycles.

Figure 5.
Average Score of a Meme Post on Reddit.

Figure 6 depicts the histogram of scores for Reddit posts conditional on a score of at least 1. The median score is 17, while the mean is 243 and the highest score is over 100,000. Note that the right tail has more mass than would be expected if the distribution were lognormal. Reddit's ranking algorithm contributes to this fat-tailed distribution. Like most social media platforms, Reddit's ranking algorithm assigns more prominence to posts based on both recency and quality. The posts that are presented earlier in Reddit readers’ feeds are both those uploaded more recently and those that have already attained higher scores. This last aspect implies that posts that initially gain some traction obtaining likes will be viewed by more Reddit readers and are more likely to obtain subsequent likes. This generates a bandwagon effect in which Reddit's algorithm contributes to a post going viral. This will tend to exaggerate the quality of high-scoring meme expressions, which, for our purposes, contributes measurement error to scores. This could tend to bias estimates of the degree of faddishness upward if scores are positively correlated across posts. This would be the case if a viral post is more likely to lead to a subsequent viral post using the same meme than would be the case without Reddit's algorithm favoring higher scores.

Figure 6.
Histogram of Logarithm of Posts’ Scores.

Finally, the overall pattern of meme usage and appeal over the life cycle is characterized by Figures 2 and 3. Figure 2 calculates the usage of a meme relative to the quarter of its peak usage (solid line) with 95% confidence intervals (shaded area) versus a baseline level. The monotonic rise in popularity over the year prior to the peak is consistent with the increased use of memes as they become fads. The monotonic decline over the year after the peak is consistent with the meme becoming passé. Figure 3 displays the logarithm of the average meme score (solid line) and 95% confidence intervals (shaded area) relative to a baseline over the same time span. The appeal of memes also follows an inverted U-shaped life cycle pattern. Importantly, their appeal precedes their popularity, suggesting that the popularity over the life cycle is a result of changes in memes’ appeal.

Empirical Model

The panel nature of the data requires slight modifications to Equations 1 and 2 to account for other factors affecting meme usage, the intermittent posting of memes, and the scale differences across memes. The estimating equations represent a dynamic linear model represented as follows: $\begin{aligned} scor e_{m, t} = \sum_{k = 1}^{K} ρ_{k} scor e_{m, t - k} + \sum_{k = 1}^{K} δ_{k} post s_{m, t - k} + β^{s} X_{m, t} + ε_{m, t}^{s} \end{aligned} .$ (3) $\begin{aligned} post s_{m, t} = \sum_{k = 1}^{K} θ_{k} scor e_{m, t - k} + \sum_{k = 1}^{K} ϕ_{k} post s_{m, t - k} + β^{p} X_{m, t} + ε_{m, t}^{p} \end{aligned} .$ (4)

The vector X_m,t includes other potentially confounding factors. Nine separate year dummy variables were included to capture the substantial growth in the use of memes over the sample period. Time devoted to meme activity could be affected by time participating in other pastimes or activities (e.g., work, sports, vacations). Consequently, X_m,t also includes 52 week-of-year dummy variables and seven day-of-week dummy variables. When the data include no posts for a meme on a specific day, we can infer that zero expressions were posted. However, we do not have information on how well these nonposted memes would have been received. That is, the average score cannot be calculated when there are no items to average over. Rather than imputing a score value, time periods without a post are excluded from the estimation for scores.

An inappropriate interval for t could induce bias in the estimates of interest. Often, some event, such as a fantastic sporting performance, leads two meme creators to both produce meme expressions. Even though they are created simultaneously and independently, they would be posted sequentially. If these are posted in subsequent periods, we might incorrectly infer that one caused the other. This is more likely to happen for small time intervals t. Likewise, if a later expression is derived from an earlier expression but is posted within the same period, we might incorrectly infer no dependence. This is more likely to happen for large time intervals. In general, using a time interval that is too small will tend to bias persistence estimates upward, whereas using one that is too large will tend to bias estimates downward. This is because the model assumes that all posts of a meme in the same period are independent and no posts in subsequent periods are independent. The empirical specification sets t as one day on the basis of the assumption that many social media users are active once per day. Still, coefficients should be interpreted with caution. At the same time, some specifications include longer lags to allow for more complex persistence patterns with AR(K) processes. As alluded to previously, the adjustment processes may be longer than AR(1). In practice, no more than four-day lags (i.e., K = 4) are ever relevant.

The cross-section portion of the panel data represents potentially heterogeneous usage of the different memes. Table 1 indicates that, across memes, the aggregations within a day will include dramatically different numbers of meme expressions per day. The differences in the average propensity to use a meme are captured by including meme fixed effects in X_m,t. Moreover, aggregations of posts within a period will result in differences in variances across memes. Averages over a period for a meme with twice as many posts per period will tend to have a proportionately smaller variance. This is dealt with by weighting observations by the number of posts generated for a meme over the entire sample. Additionally, the units for the average score regressions do not have a clear interpretation. To assist with the interpretation of results, average Reddit scores are transformed into the coefficient of variation in scores for the meme. In this way, the coefficients for scores can be interpreted as the effect of a one standard deviation change.

Results

Table 4 reports the results of the estimation of the number of meme posts per day. Each column represents a separate specification that includes progressively more lags. The control variables mentioned previously are also included in all specifications but are not reported. These regressions include nearly three-quarters of a million observations over 420 different memes. The R² values indicate that the model accounts for a large degree of the variation in the creation of meme expressions.

Table 4.
Determinants of the Number of Meme Posts.

(1) (2) (3) (4)

Number of Past Posts

One-day lag .826* .655* .633* .636*

(.001) (.001) (.001) (.001)

Two-day lag .206* .137* .140*

(.001) (.001) (.001)

Three-day lag .105* .121*

(.001) (.001)

Four-day lag −.024*

(.001)

Average Score of Past Posts

One-day lag .039* .042* .043* .043*

(.002) (.002) (.002) (.002)

Two-day lag .033* .035* .035*

(.002) (.002) (.002)

Three-day lag .021* .021*

(.002) (.002)

Four-day lag −.004*

(.002)

Observations 705,151 704,681 704,211 703,741

R² .750 .761 .763 .763

No. of memes 470 470 470 470

p < .01.

Notes: Specifications include fixed effects for the meme, the year, the week of the year, and the day of the week. Observations are weighted by the overall number of posts of the meme.

The estimates are generally indicative of fad behavior. The positive and significant coefficient estimates for lagged meme posts indicate a large degree of persistence in using any given meme in a post. Column 1 indicates that each additional meme post in a day will generate .826 posts the following day. Moreover, as more lags are included, the estimated effects become progressively smaller. The sum of the lag coefficients in the other columns range from .861 to .875. These values suggest the half-life of a shock is just over five days. A single shock in one day will have dissipated to an expected one-sixth of the initial magnitude after two weeks. Users continue to use a meme in subsequent days at an ever-decreasing rate. Moving to the estimate of the effect on past scores, we find positive and significant estimates. Meme creators are more likely to use a meme that temporarily has broader appeal. Again, the estimates also become progressively smaller with larger lags. Here the sum of the coefficients ranges from .039 to .099, implying that a one standard deviation increase in past meme “quality” on one day generates 4%–10% more memes being created over subsequent days. However, Figure 6 suggests that a viral meme expression could increase the score by ten times the standard deviation. In this case, we could observe up to a doubling in the number of subsequent expressions of the meme. Generating more posts due to increased meme score is consistent with attention-seeking behavior.

Table 3 reports the identical specifications for the average daily meme score. The number of observations is dramatically reduced due to the absence of a recorded score on days in which a meme is not used in a post. While unreported fixed effects for year, seasonality, and meme have large effects on the number of posts in Table 4, they affect the average score much less. Consequently, the reported R² values in Table 3 are much lower than those in Table 4.

In this case, recent experience with a meme is estimated to have little effect on a meme's average score. For example, the number of past posts using a meme has no measurable effect on the current average score. We can reject the hypothesis of diminishing marginal utility from the proliferation of expressions of a specific meme. Additionally, past scores only have a small, but statistically significant, negative effect on current scores. As discussed previously, had this been a positive effect, Reddit's ranking algorithm could have induced bias in this estimate. The effect implies that a one standard deviation in meme score on the previous day may decrease the meme's score on the current day by less than 2% of a standard deviation. This may only be important if the previous expression went viral, increasing the score tenfold. In this case, the subsequent score would fall by 20% of a standard deviation. Recall that higher-scoring meme expressions are believed to be shared more and thus spread more broadly throughout social networks. That meme scores decrease with past scores suggests that meme spread decreases with past spread. This is consistent with some diminishing marginal value in attracting attention when a meme is shared.

Fad behavior is predicated on a desire for attention. The results in Table 4 are consistent with attention-seeking behavior in the time series. Producers are more likely to choose a meme for posts when these memes will garner more attention. Memes continue to be used days after an initial shock to their use or to their scores. A robustness check for attention-seeking behavior is to compare the use of memes in the cross-section. Meme expression creators can choose among hundreds of popular meme templates to express themselves. Does additional attention for other memes divert attention away from the focal meme? If so, meme creators should be less likely to use a meme when other memes are experiencing positive shocks to their popularity. To test this, Equations 3 and 4 are augmented with information about all other posts matched to memes. That is, the number and the average score of past posts of all other memes are included in addition to those for the focal meme. Increased quality and increased usage of these other memes are hypothesized to vie for attention on the focal meme. That is, the effects of experience with other memes on the focal meme should be the opposite of the effects of experience with the focal meme. By including measures for other memes in the model for the focal meme, we can test whether this is true.

Table 5 reports the results of specifications that include information on both the focal meme and all other memes for up to three-day lags. The results for past experience with the focal meme for both the post's duration and score are qualitatively unchanged from those in Tables 3 and 4. Specifically, in Columns 1–3, more posts using the meme and higher-scoring posts increase the use of the focal post. The experience for other memes is estimated to have the opposite effect. More posts of other memes and higher average scores for other memes tend to decrease the focal meme’s use and its Reddit score. As expected, the magnitudes for the other memes are much smaller than those for the focal meme, likely because these additional variables aggregate over so many other memes. This general result is consistent with memes competing for user attention in the cross-section as well as over time.

Table 5.
Competition for Attention Across Memes.

Number of Posts Average Score

(1) (2) (3) (4) (5) (6)

Same Meme—Past Posts

One-day lag .827 .656* .635* −.004 .001 .002

(.001) (.001) (.001) (.003) (.005) (.005)

Two-day lag .205* .140* −.007 −.005

(.001) (.002) (.005) (.006)

Three-day lag .099* −.004

(.001) (.005)

Same Meme—Past Score

One-day lag .038* .041* .042* −.016 −.017 −.017

(.002) (.002) (.002) (.007) (.007) (.007)

Two-day lag .032* .034* .001 .001

(.002) (.002) (.008) (.008)

Three-day lag .022* −.007

(.002) (.008)

Other Memes—Past Posts

One-day lag −.006* −.004* −.004* −.000 −.001 −.001

(.000) (.000) (.000) (.001) (.001) (.001)

Two-day lag −.001* −.004* .001 .001

(.000) (.000) (.001) (.001)

Three-day lag .005* −.000

(.000) (.001)

Other Memes—Past Score

One-day lag −.022* −.019* −.016* −.092* −.088* −.085*

(.004) (.005) (.005) (.028) (.028) (.029)

Two-day lag −.003 −.002 .072 .085

(.004) (.005) (.032) (.033)

Three-day lag .019* −.027

(.005) (.037)

Observations 665,272 634,470 610,725 26,127 25,786 25,516

R² .749 .759 .761 .013 .013 .013

No. of memes 470 470 470 470 470 470

p < .1.

p < .05.

p < .01.

Notes: Specifications include fixed effects for the meme, the year, the week of the year, and the day of the week. Observations are weighted by the overall number of posts of the meme.

Columns 4–6 in Table 5 report results for a post's score. As before, more past posts of a meme have no effect on a post's score, but neither do more past posts of other memes. The estimated effects of past scores for the focal meme are almost entirely unchanged from those in Table 3. Higher past scores for the focal meme tend to reduce current scores by a small amount. However, rather than being in the opposite direction, the estimates indicate that higher past scores of other memes also tend to reduce current scores by a small amount. The sum of these effects varies across specifications from −.097 to −.007, leading to roughly 1% to 10% lower scores. An explanation for this could be that the higher scores of the other memes are crowding out attention across all memes, thereby decreasing the scores for the focal meme.

In general, these estimates are consistent with attention seeking and fad behavior being important factors driving the popularity of internet memes. The impact of the lagged past scores of other memes on the average score of the focal meme may also indicate the production-blocking phenomenon in collaborative ideation (e.g., Diehl and Stroebe 1987, 1991; Stephen, Zubcsek, and Goldenberg 2016). One limitation of these conclusions is that the Reddit sample may be idiosyncratic. The sample was generated from Reddit primarily because of data availability. Memes are posted to, and shared from, many different social networking or bulletin board sites. The operational differences across sites could affect users’ behaviors so that these results do not generalize. With that in mind, this sample and analysis represent the first systematic attempt to estimate fad behavior.

Discussion

The practitioner implications from this research stem from how fast memes rise and fall in popularity over fad cycles. The foremost implication is that marketing managers must act swiftly to leverage the viral nature of trending content. Timeliness is essential, as memes lose relevance and appeal quickly. Overusing memes or using them in a way an inauthentic way can harm a brand's image. Too frequent use or use without genuine relevance can make the brand seem desperate to stay relevant or out of touch with its audience. To effectively engage the audience, managers must understand their humor and references, ensuring the meme resonates and aligns with the brand's voice and values.

These implications are consistent with general guidance to practitioners wishing to engage in meme marketing (Razzaq, Shao, and Quach 2023). Memes often target specific subcultures or online communities. While a meme might perform well with one demographic, it may fall flat or confuse others. Memes can often be edgy, controversial, or humorous in ways that might not align with a brand's image. Using a meme that fails to resonate well with the target audience can backfire, potentially harming the brand's reputation or creating negative perceptions. Marketers must be aware of which audiences are likely to engage with the meme and tailor campaigns accordingly. Relatedly, cultural sensitivity will help avoid unintended offense. The meme should also be relevant to the product or service being marketed, creating a natural connection that enhances brand messaging. Finally, measuring the impact of meme-driven campaigns is essential for assessing their success. Managers should track engagement metrics and monitor for any potential backlash. With a strategic approach, memes can be an effective and fun way to engage with audiences, but they require careful management to ensure they contribute positively to the brand's reputation.

These findings may suggest strategic decisions regarding how to engage in meme marketing. They highlight the risks of using cultural imagery that no longer resonates with the targeted audience. To be successful, a meme marketer would need to know not only how to use the meme appropriately, but also when the meme is associated with growing or waning popularity. Developing the relevant human capital will likely require substantial investment in time and energy. This specific human capital may be more readily found in a marketing agency specializing in this specific form of digital marketing. However, relying on external management weakens control over a marketing campaign.

Conclusion

The use of internet memes continues to evolve, with short video clips currently gaining in popularity relative to static images. No matter the form they take, if memes command user attention, marketers will find a way to exploit this focused attention. This study attempts to understand certain aspects of the dynamics of meme creation and usage. A model based on the market for attention was developed, generating implications for the dynamics of internet meme popularity and appeal that are consistent with fads. Panel estimators were developed to parameterize the evolution of meme usage (i.e., popularity) and entertainment value (i.e., perceived quality). The estimated dynamics appear to conform to implications from a simple model of faddish behavior. Positive shocks to meme expression quality increase the expected attention a meme will attract and are met by an increase in production. This creates dynamics in the number and quality of meme expressions that persist for about two to three weeks. Moreover, the transitory popularity of a meme will depress the popularity of other memes. Memes compete for audience attention. While fad behavior was found for internet meme images, other aspects of meme expressions, the events commented about and the commentary, are also likely to exhibit fad behavior. This is an area for future work.

The estimated dynamic patterns for internet meme fad cycles have implications for the use of internet memes in marketing campaigns. The use of internet memes has been shown to be highly effective in reaching specific audiences. Those targeted appear to be younger, better educated, highly engaged, and internet savvy. However, inappropriate use can backfire. One aspect of inappropriate use is using a meme that has become passé through overuse. The estimates suggest this can happen quickly. The implication is that meme marketing campaigns must be nimble to be effective. Brand managers would have to monitor meme usage regularly, correct strategies, and adjust implementations quickly. More likely, they can outsource this to those with expertise in meme marketing. The patterns uncovered in this research are likely a small portion of the institutional knowledge developed by those steeped in the milieu. Meme marketing campaigns can exploit these and other insights about meme popularity from the meme marketing segment of the industry. Meme influencers are at the forefront of meme creation. Meme marketing agencies work with multiple content creators with heterogeneous specialties and different media outlets that reach specific audiences. Meme consulting agencies specialize in the specific analytics measuring meme effectiveness. Just as digital marketers have specific expertise beyond traditional advertising agencies, meme marketers represent a specialization within digital marketing.

New forms of social interaction in which users vie for attention continue to emerge on the internet. Thousands of users are creating single-season and narrow-topic podcasts. Millions of esports spectators are using and creating new emotes to register their emotional response to game play. Perhaps billions of users are sharing short video selfies via services such as TikTok, Snapchat, or Instagram. Each of these are attempts to garner more attention by delivering entertaining content. Those with larger audiences will seek to monetize the attention, possibly through promoted content. These new phenomena, and others yet developed, could include behavioral dynamics that include faddishness. The analysis in this article may inform the study of these behaviors.

Rank	Meme Name	Number of Posts	Percentage of Posts	Ln(Reddit Score)
1	Drake Hotline Bling	9,121	13.3%	2.58	2.03
2	Expanding Brain	7,569	11.1%	2.64	2.30
3	Tuxedo Winnie The Pooh	3,786	5.5%	2.64	1.99
4	Surprised Pikachu	3,632	5.3%	2.48	2.19
5	Unsettled Tom	2,820	4.1%	2.35	1.87
6	Two Buttons	2,371	3.5%	2.50	1.96
7	Is This A Pigeon	2,300	3.4%	2.74	2.29
8	Distracted Boyfriend	2,140	3.1%	2.45	2.16
9	The Scroll Of Truth	1,873	2.7%	2.81	2.21
10	Left Exit 12 Off Ramp	1,520	2.2%	2.69	2.21
11	Change My Mind	1,404	2.1%	2.53	2.01
12	Me And The Boys	1,124	1.6%	2.53	1.59
13	Well Yes, But Actually No	977	1.4%	2.85	2.16
14	Blank Nut Button	964	1.4%	2.23	2.08
15	Boardroom Meeting Suggestion	901	1.3%	2.77	2.25
16	Futurama Fry	876	1.3%	2.08	1.92
17	American Chopper Argument	844	1.2%	3.21	2.40
18	Success Kid	835	1.2%	2.19	1.96
19	Hard To Swallow Pills	804	1.2%	2.99	2.38
20	First World Problems	800	1.2%	1.94	1.58
21	Bad Luck Brian	786	1.2%	2.05	1.79
22	Philosoraptor	573	.8%	1.98	2.01
23	Who Killed Hannibal	571	.8%	3.14	2.32
24	Car Salesman Slaps Hood	457	.7%	2.81	2.42
25	Confession Bear	453	.7%	2.26	1.96
26	Batman Slapping Robin	451	.7%	2.03	1.99
27	Who Would Win?	433	.6%	3.30	2.44
28	Running Away Balloon	419	.6%	2.84	1.84
29	Roll Safe Think About It	394	.6%	2.10	2.07
30	Sleeping Shaq	383	.6%	3.04	2.25
31	Scumbag Steve	378	.6%	2.02	1.84
32	10 Guy	372	.5%	1.76	1.63
33	One Does Not Simply	353	.5%	1.78	1.80
34	Be Like Bill	316	.5%	3.30	2.75
35	Good Guy Greg	311	.5%	2.47	1.92
36–511	All others	15,014	22.0%	2.11	1.83

Number of Meme Posts per Creator	Number of Creators	Total Number of Posts
1	34,629	34,629
2	6,552	13,104
3	1,845	5,535
4	755	3,020
5	362	1,810
6	192	1,152
7	119	833
8	75	600
9	56	504
10	27	270
11	26	286
12	17	204
13–95	65	1,400

	(1)	(2)	(3)	(4)
Number of Past Posts
One-day lag	−.004	.001	.002	.002
	(.003)	(.005)	(.005)	(.005)
Two-day lag		−.007	−.004	−.004
		(.005)	(.006)	(.006)
Three-day lag			−.004	−.004
			(.005)	(.006)
Four-day lag				−.000
				(.005)
Average Score of Past Posts
One-day lag	−.016**	−.017**	−.017**	−.017**
	(.007)	(.007)	(.007)	(.007)
Two-day lag		.002	.001	.001
		(.008)	(.008)	(.008)
Three-day lag			−.007	−.007
			(.007)	(.007)
Four-day lag				−.010
				(.009)
Observations	26,501	26,468	26,448	26,423
R²	.012	.012	.012	.012
No. of memes	470	470	470	470

	(1)	(2)	(3)	(4)
Number of Past Posts
One-day lag	.826***	.655***	.633***	.636***
	(.001)	(.001)	(.001)	(.001)
Two-day lag		.206***	.137***	.140***
		(.001)	(.001)	(.001)
Three-day lag			.105***	.121***
			(.001)	(.001)
Four-day lag				−.024***
				(.001)
Average Score of Past Posts
One-day lag	.039***	.042***	.043***	.043***
	(.002)	(.002)	(.002)	(.002)
Two-day lag		.033***	.035***	.035***
		(.002)	(.002)	(.002)
Three-day lag			.021***	.021***
			(.002)	(.002)
Four-day lag				−.004***
				(.002)
Observations	705,151	704,681	704,211	703,741
R²	.750	.761	.763	.763
No. of memes	470	470	470	470

	Number of Posts	Average Score
Same Meme—Past Posts
One-day lag	.827***	.656***	.635***	−.004	.001	.002
	(.001)	(.001)	(.001)	(.003)	(.005)	(.005)
Two-day lag		.205***	.140***		−.007	−.005
		(.001)	(.002)		(.005)	(.006)
Three-day lag			.099***			−.004
			(.001)			(.005)
Same Meme—Past Score
One-day lag	.038***	.041***	.042***	−.016**	−.017**	−.017**
	(.002)	(.002)	(.002)	(.007)	(.007)	(.007)
Two-day lag		.032***	.034***		.001	.001
		(.002)	(.002)		(.008)	(.008)
Three-day lag			.022***			−.007
			(.002)			(.008)
Other Memes—Past Posts
One-day lag	−.006***	−.004***	−.004***	−.000	−.001	−.001
	(.000)	(.000)	(.000)	(.001)	(.001)	(.001)
Two-day lag		−.001***	−.004***		.001	.001
		(.000)	(.000)		(.001)	(.001)
Three-day lag			.005***			−.000
			(.000)			(.001)
Other Memes—Past Score
One-day lag	−.022***	−.019***	−.016***	−.092***	−.088***	−.085***
	(.004)	(.005)	(.005)	(.028)	(.028)	(.029)
Two-day lag		−.003	−.002		.072**	.085**
		(.004)	(.005)		(.032)	(.033)
Three-day lag			.019***			−.027
			(.005)			(.037)
Observations	665,272	634,470	610,725	26,127	25,786	25,516
R²	.749	.759	.761	.013	.013	.013
No. of memes	470	470	470	470	470	470

Footnotes

Coeditor

Arvind Rangaswamy

Associate Editor

Koen H. Pauwels

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

Funding

This research was supported by a University of Texas at Arlington,College of Business Research Expense Grant.

ORCID iD

Michael R. Ward

References

Amalia

Sharif

Amer

Haisar

Fikri

Gunawan

Dani

Nasution

Benny B.

(2018), “Meme Opinion Categorization by Using Optical Character Recognition (OCR) and Naïve Bayes Algorithm,” 2018 Third International Conference on Informatics and Computing (ICIC), 1–5, https://doi.org/10.1109/IAC.2018.8780410.

Ashuri

Tamar

Bar-Ilan

Yaniv

(2017), “Collective Action Recruitment in a Digital Age: Applying Signaling Theory to Filtering Behaviors,” Communication Theory, 27 (1), 70–91.

Bikhchandani

Sushil

Hirshleifer

David

Welch

Ivo

(1992), “A Theory of Fads, Fashion, Custom, and Cultural Change as Informational Cascades,” Journal of Political Economy, 100 (5), 992–1026.

Bonchi

Francesco

Castillo

Carlos

Ienco

Dino

(2013), “Meme Ranking to Maximize Posts Virality in Microblogging Platforms,” Journal of Intelligent Information Systems, 40, 211–39.

Booth

Callum

(2024), “What Are the ‘When Gen-Z Writes the Marketing Script’ Memes About?” Forbes (September 27), https://www.forbes.com/sites/callumbooth/2024/09/27/what-are-the-when-gen-z-writes-the-marketing-script-memes-about/.

Brito

Paulo

Barros

Carlos

(2005), “Learning-by-Consuming and the Dynamics of the Demand and Prices of Cultural Goods,” Journal of Cultural Economy, 29, 83–106.

Chai

Sen

Menon

Anoop

(2019), “Breakthrough Recognition: Bias Against Novelty and Competition for Attention,” Research Policy, 28 (3), 733–47.

Coscia

Michele

(2013), “Competition and Success in the Meme Pool: A Case Study on Quickmeme.com,” International Conference of Weblogs and Social Media, arXiv, https://doi.org/10.48550/arXiv.1304.1712.

Coscia

Michele

(2014), “Average Is Boring: How Similarity Kills a Meme’s Success,” Scientific Reports, 4, 6477.

10.

Dawkins

Richard

(1976), The Selfish Gene. Oxford University Press.

11.

Diehl

Michael

Stroebe

Wolfgang

(1987), “Productivity Loss in Brainstorming Groups: Toward the Solution of a Riddle,” Journal of Personality and Social Psychology, 53 (3), 497–509.

12.

Diehl

Michael

Stroebe

Wolfgang

(1991), “Productivity Loss in Idea-Generating Groups: Tracking Down the Blocking Effect,” Journal of Personality and Social Psychology, 61 (3), 392–403.

13.

Donath

Judith

(2007), “Signals in Social Supernets,” Journal of Computer-Mediated Communication, 13 (1), 231–51.

14.

GeeksforGeeks (2024), “Meme Marketing: Meaning, Strategy, Uses, Advantages and Disadvantages,” (February 29), https://www.geeksforgeeks.org/meme-marketing-meaning-strategy-uses-advantages-and-disadvantages/.

15.

Gleeson

James P.

Ward

Jonathan A.

O’Sullivan

Kevin P.

Lee

William T.

(2014), “Competition-Induced Criticality in a Model of Meme Popularity,” Physical Review Letters, 112 (4), 048701.

16.

Glenski

Maria

Pennycuff

Corey

Weninger

Tim

(2017), “Consumers and Curators: Browsing and Voting Patterns on Reddit,” IEEE Transactions on Computational Social Systems, 4 (4), 196–206.

17.

Guadagno

Rosanna E.

Rempala

Daniel M.

Murphy

Shannon

Okdie

Bradley M.

(2013), “What Makes a Video Go Viral? An Analysis of Emotional Contagion and Internet Memes,” Computers in Human Behavior, 29 (6), 2312–19.

18.

Saike

Zheng

Xiaolong

Zeng

Daniel

(2016), “A Model-Free Scheme for Meme Ranking in Social Media,” Decision Support Systems, 81, 1–11.

19.

Saike

Zheng

Xiaolong

Zeng

Daniel Dajun

(2019), “Modeling Online User Behaviors with Competitive Interactions,” Information & Management, 56 (4), 463–75.

20.

Heath

Joseph

Potter

Andrew

(2004), Nation of Rebels: Why Counterculture Became Consumer Culture. HarperCollins.

21.

Heath

Joseph

Potter

Andrew

(2005), The Rebels Sell: Why the Culture Can’t Be Jammed. Capstone Publishing.

22.

Influencer Marketing Hub (2025), “ 8 of the Best Instagram Meme Accounts to Check Out in 2025,” (March 21), https://influencermarketinghub.com/instagram-meme-accounts/.

23.

Iyer

Ganesh

Katona

Zsolt

(2016), “Competing for Attention in Social Communication Markets,” Management Science, 62 (8), 2304–20.

24.

Kar

Sanghamitra

(2020), “Meme Makers Laughing All the Way to the Bank,” Economic Times (January 31), https://economictimes.indiatimes.com/tech/internet/meme-makers-laughing-all-the-way-to-the-bank/articleshow/73755230.cms.

25.

Lin

Jih-Hsuan Tammy

(2019), “Strategic Social Grooming: Emergent Social Grooming Styles on Facebook, Social Capital and Well-Being,” Journal of Computer-Mediated Communication, 24 (3), 90–107.

26.

Malodia

Suresh

Dhir

Amandeep

Bilgihan

Anil

Sinha

Pranao

Tikoo

Tanishka

(2022), “Meme Marketing: How Marketers Can Drive Better Engagement Using Viral Memes?” Psychology & Marketing, 39 (9), 1775–801.

27.

Mercure

Jean-François

(2018), “Fashion, Fads and the Popularity of Choices: Micro-Foundations for Diffusion Consumer Theory,” Structural Change and Economic Dynamics, 46, 194–207.

28.

Nissenbaum

Asaf

Shifman

Limor

(2017), “Internet Memes as Contested Cultural Capital: The Case of 4chan’s /b/ Board,” New Media & Society, 19 (4), 483–501.

29.

Razzaq

Ali

Shao

Wei

Quach

Sara

(2023), “Towards an Understanding of Meme Marketing: Conceptualisation and Empirical Evidence,” Journal of Marketing Management, 39 (7–8), 670–701.

30.

Razzaq

Ali

Shao

Wei

Quach

Sara

(2024), “Meme Marketing Effectiveness: A Moderated-Mediation Model,” Journal of Retailing and Consumer Services, 78, 103702.

31.

Roache

Kiley

(2019), “Brands Are Bypassing Influencers and Targeting Teens with Memes,” Bloomberg (November 27, 2019), https://www.bloomberg.com/news/articles/2019-11-27/brands-are-bypassing-influencers-and-targeting-teens-with-memes.

32.

Rossi

Federico

Rubera

Gaia

(2021), “Measuring Competition for Attention in Social Media: National Women’s Soccer League Players on Twitter,” Marketing Science, 40 (6), 1147–68.

33.

Rui

Huaxia

Whinston

Andrew

(2012), “Information or Attention? An Empirical Study of User Contribution on Twitter,” Information Systems and E-Business Management, 10, 309–24.

34.

Scanlon

Kyla

(2020), “Memes as News: The Importance of Short-Form Content,” The Startup (October 4), https://medium.com/swlh/memes-as-news-the-importance-of-short-form-content-4c12d0718581.

35.

Schlaile

Michael P.

(2021), “A Case for Economemetics? Why Evolutionary Economists Should Re-Evaluate the (F)Utility of Memetics,” in Memetics and Evolutionary Economics: Economic Complexity and Evolution, Schlaile

Michael P.

, ed. Springer.

36.

Segev

Elad

Nissenbaum

Asaf

Stolero

Nathan

Shifman

Limor

(2015), “Families and Networks of Internet Memes: The Relationship Between Cohesiveness, Uniqueness, and Quiddity Concreteness,” Journal of Computer-Mediated Communication, 20 (4), 417–33.

37.

Shabunina

Ekaterina

Pasi

Gabriella

(2018), “A Graph-Based Approach to Ememes Identification and Tracking in Social Media Streams,” Knowledge-Based Systems, 139 (1), 108–18.

38.

Shifman

Limor

(2014), Memes in Digital Culture. MIT Press.

39.

Spitzberg

Brian H.

(2014), “Toward a Model of Meme Diffusion (M3D),” Communication Theory, 24 (3), 311–39.

40.

Spitzberg

Brian H.

(2018), “Framing the Game: An Architectonic Analogue for Meta-Theorizing Academic Activities,” Studies in Media and Communication, 6 (1), 11–25.

41.

Spitzberg

Brian H.

(2021), “Comprehending Covidiocy Communication,” in Communicating Science in Times of Crisis, O’Hair

H. Dan

O’Hair

Mary John

, eds. John Wiley & Sons.

42.

Stephen

Andrew T.

Zubcsek

Peter P.

Goldenberg

Jacob

(2016), “Lower Connectivity Is Better: The Effects of Network Structure on Redundancy of Ideas and Customer Innovativeness in Interdependent Ideation Tasks,” Journal of Marketing Research, 53 (2), 263–79.

43.

Stigler

George J.

Becker

Gary S.

(1977), “De Gustibus Non Est Disputandum,” American Economic Review, 67 (2), 76–90.

44.

Susarla

Anjana

Jeong-Ha

Tan

Yong

(2012), “Social Networks and the Diffusion of User-Generated Content: Evidence from YouTube,” Information Systems Research, 23 (1), 23–41.

45.

Tama-Rutigliano

Karina

(2018), “Memes: A Digital Marketing Tool for Every Industry,” Forbes (August 10), https://www.forbes.com/sites/forbescommunicationscouncil/2018/08/10/memes-a-digital-marketing-tool-for-every-industry/.

46.

Toubia

Olivier

Stephen

Andrew T.

(2013), “Intrinsic vs. Image-Related Utility in Social Media: Why Do People Contribute Content to Twitter?” Marketing Science, 32 (3), 368–92.

47.

Warren

Caleb

Batra

Rajeev

Loureiro

Sandra Maria Correia

Bagozzi

Richard P.

(2019), “Brand Coolness,” Journal of Marketing, 83 (5), 36–56.

48.

Wasike

Ben

(2022), “When the Influencer Says Jump! How Influencer Signaling Affects Engagement with COVID-19 Misinformation,” Social Science & Medicine, 315.

49.

Weng

Flammini

Vespignani

Menczer

(2012), “Competition Among Memes in a World with Limited Attention,” Scientific Reports, 2, 335.

50.

Xie

Lexing

Natsev

Apostol

Kender

John R.

Hill

Matthew

Smith

John R.

(2011), “Visual Memes in Social Media: Tracking Real-World News in YouTube Videos,” in MM ’11: Proceedings of the 19th ACM International Conference on Multimedia. Association for Computing Machinery, 53–62.