Fanfiction, broadly defined as literature written by admirers of existing works of fiction that borrows characters, settings, or universes from those stories, has surged in popularity over recent decades. While many fanworks incorporate romantic or sexual elements drawn from their source material, a substantial subset is written partially or exclusively for pornographic purposes. Archive of Our Own, commonly known as AO3, is the leading host for this content, holding millions of stories and attracting traffic that rivals major commercial websites. According to traffic analytics cited by the authors, English-language stories make up roughly 87 percent of the site’s approximately three million explicit works, which is why the research team chose to restrict its sampling to English-language fiction.

The study was conducted by Ezelbahar Metin and Eran Shor of the Department of Sociology at McGill University in Montreal. Despite fanfiction’s popularity as an alternative form of pornography, the authors note that it has garnered far less research attention than visual mainstream pornography or commercial erotic fiction. Content analyses of pornographic videos have repeatedly documented high rates of aggression, including spanking, hitting, choking, and verbal abuse, and have sparked decades of debate about whether such content shapes attitudes toward violence against women. Written erotica produced largely by and for women has rarely been subjected to the same quantitative measurement, leaving a substantial gap in the evidence base.

To fill that gap, the researchers assembled two samples from AO3. The first consisted of the 525 most popular sexually explicit fanfiction stories on the platform, capturing the content most widely read and endorsed by users through the site’s kudos and bookmarking systems. The second was a random sample of 360 explicit stories, designed to represent the broader universe of explicit fanfiction rather than only its most celebrated works. Sampling was drawn across fandoms spanning multiple media categories, including films, television series, books, and video games, with well-known franchises such as Marvel and Star Wars strongly represented. The authors acknowledge a slight oversampling of the Marvel and Star Wars fandoms due to a collection error, which they chose to retain in the analyses, and they excluded the Minecraft fandom because most of its fanfiction involves real people rather than fictional characters.

The core analytical strategy relied on AO3’s distinctive tagging system, a crowdsourced classification infrastructure in which authors voluntarily label their stories with descriptive tags covering pairings, themes, and warnings. This system allowed the researchers to conduct a large-scale textual analysis of aggression-related tags across nearly 900 stories. The results were striking: approximately half of the stories in the overall sample carried at least one tag conveying aggression. Even more notably, acts of non-consensual aggression or rape were present in almost 20 percent of the sample. In the most popular stories, aggression tags appeared at similarly high rates, suggesting that aggressive themes are not a fringe interest but a mainstream feature of the genre’s most-read content.

Because tags depend entirely on voluntary author self-classification, the researchers took the additional step of verifying what the tags actually captured. They conducted a verification content analysis, reading the full text of 60 stories from the sample to determine whether aggression coded in the tags corresponded to aggression depicted in the narratives themselves. This deeper examination suggested that the true prevalence of aggression may in fact be substantially higher than the tag-based estimates indicate. Some authors, the analysis implies, may under-tag aggressive content, either because they do not consider certain depictions aggressive or because they wish to avoid warning labels that might deter readers. The verification step therefore functions as a methodological caution for any researcher hoping to use platform tags as a proxy for content in user-generated corpora.

The findings sit within a rich and contested literature on aggression in sexual media. Prior content analyses of best-selling pornography videos, including work by Bridges and colleagues and by Fritz and colleagues, have documented that a majority of mainstream scenes contain physical aggression, most often directed at women. Studies of online pornography by Shor and Seida and by Klaassen and Peter have similarly found aggression woven into popular content, while meta-analyses by Wright, Tokunaga, and Kraus and by Ferguson and Hartley continue to debate whether consumption of such material is linked to real-world sexual aggression. Written erotica has its own smaller research tradition, with analyses of romance novels, slash fiction, and transgressive themes in commercial erotica documenting long-standing representations of force and domination in fiction marketed to women.

What makes the fanfiction case especially provocative is its audience demographics. Survey evidence and platform research indicate that cisgender women form the largest single category of AO3 users, with nonbinary people forming a substantial second group. The authors frame their analysis as an important step toward illuminating women’s sexual preferences precisely because fanfiction is mostly read and written by women, in contrast to mainstream pornography, whose production and consumption have historically been male-dominated. The high prevalence of aggression tags raises challenging questions for theories that attribute aggressive content in pornography solely to male demand or male-controlled production. When women are the primary authors and readers, the presence of force fantasies in erotica complicates simple supply-side explanations and aligns instead with a broader body of research documenting that sexual fantasies involving force or submission are reported by a nontrivial share of women in general-population studies.

The authors are careful to situate these findings within their limits. Fanfiction is fiction, produced within communities that maintain elaborate norms about consent, warning, and reader choice, and AO3’s tagging architecture is explicitly designed to let readers curate their own exposure. Depicting rape in a story is not equivalent to endorsing it, and a long scholarly tradition in fandom studies reads transgressive fanworks as explorations of power, trauma, and desire rather than straightforward expressions of preference. The researchers also caution that some sampled stories may no longer be available on the platform and that some tags may have been updated since sampling, introducing a degree of temporal uncertainty into the estimates. Still, the scale of the corpus, the combination of a popularity-based and a random sample, and the verification of tags against full texts give the study an unusual degree of robustness for research on user-generated erotica.

The broader significance of the work lies in what it reveals about an enormous and largely invisible pornographic ecosystem. Millions of readers consume explicit fanfiction daily, often on the same devices and under the same roof as mainstream video pornography, yet the genre remains absent from most policy debates, content analyses, and public discussions of sexual media. By quantifying aggression in this space with standard social-scientific methods, Metin and Shor have opened a new empirical front in the study of sexual media, one that centers the preferences of the women who write and read it. Whether the aggressive themes documented here reflect fantasy, aesthetics, narrative convention, or genuine desire is a question the study does not resolve, but it firmly establishes that any complete account of sexual media and aggression must now include the written worlds of fandom alongside the visual industries that have dominated the conversation for half a century.

Subject of Research: Textual analysis of aggressive and non-consensual content in sexually explicit fanfiction on Archive of Our Own

Article Title: The Prevalence of Aggressive Content in Online Sexually Explicit Fanfiction

Article References: The Prevalence of Aggressive Content in Online Sexually Explicit Fanfiction. (n.d.). https://doi.org/10.1007/s10508-026-03523-0

Image Credits: AI Generated

DOI: 10.1007/s10508-026-03523-0

Keywords: fanfiction, Archive of Our Own, pornography, erotica, aggression, non-consensual content, sexual aggression, content analysis, women's sexual preferences, fandom, sexual media, textual analysis

The study, led by Philipp Yorck Herzberg of Helmut-Schmidt-University in Hamburg together with colleagues at the European Distance University Hamburg and the University of Bern, set out to answer a question posed nearly thirty years earlier by Dean and Malamuth. Those researchers had found in 1997 that the interaction between hostile masculinity and impersonal sexuality robustly predicted sexual aggression only among men high in dominance and low in nurturance, a pattern they interpreted as possibly driven by empathy. Nurturant individuals, after all, tend to be warm, gentle and sensitive to others’ needs, traits closely linked to empathic capacity. But the question remained whether empathy itself was the critical protective ingredient, or merely one facet of the broader dominance-nurturance dimension.

To disentangle the two, the researchers recruited 300 heterosexual men aged 18 to 65 through online advertisements on social networks and internet forums, without disclosing that the survey concerned sexual aggression. Participants completed the German version of the Sexual Experience Survey, a ten-item count of sexually coercive behaviors ranging from verbal manipulation to physical force or incapacitation through alcohol or drugs. Hostile masculinity was measured with the Hostile Sexism Scale of the Ambivalent Sexism Inventory, impersonal sexuality with the revised Sociosexual Orientation Inventory, and dominance and nurturance with the revised Bem Sex-Role Inventory. Crucially, empathy was assessed with the German adaptation of Davis’s Interpersonal Reactivity Index, which captures perspective taking, fantasy, empathic concern and personal distress.

Statistically, the team faced a technical challenge. Sexual aggression scores are count variables, skewed toward zero, with roughly sixty percent of participants reporting no aggressive acts at all. Ordinary linear regression would violate the distributional assumptions of such data. The researchers therefore employed latent negative binomial regression models within a structural equation modeling framework, comparing three nested models: one in which empathy’s effect was constrained to zero, one in which dominance/nurturance’s effect was constrained to zero, and one in which both were freely estimated. Latent modeling also sidestepped the notorious unreliability of difference scores, which the dominance-nurturance variable required.

The results were unambiguous. Across all three models, hostile masculinity and impersonal sexuality emerged as significant positive predictors of sexual aggression, confirming their status as the model’s core risk factors. Empathy, by contrast, acted as a significant negative predictor: men higher in empathy reported lower rates of sexually aggressive behavior. In the model combining all predictors, each unit increase in empathy reduced the expected rate of sexual aggression by roughly a third, while each unit increase in hostile masculinity multiplied the expected rate by about 1.46 and impersonal sexuality by 1.25. The dominance/nurturance difference score, however, contributed nothing, neither directly nor in interaction with the two risk factors.

Model fit statistics told the same story. The model containing empathy but excluding dominance/nurturance achieved the best fit according to the Akaike information criterion, the Bayesian information criterion and sample-size-adjusted variants. Because dominance/nurturance showed no effect to begin with, there was no effect for empathy to replace or explain away, despite a substantial negative correlation of minus 0.52 between empathy and the dominance/nurturance score. The authors conclude that whatever predictive value the dominance-nurturance dimension once appeared to hold was likely attributable to its overlap with empathy, and that future confluence model research should replace it with direct empathy measures.

One finding added an unexpected layer of nuance. Although the hostile masculinity by impersonal sexuality interaction failed to reach significance in the count regression models, complementary response surface analysis confirmed a synergistic effect. Using a shifted and rotated squared difference model, the analysis revealed that risk climbs steeply and nonlinearly when hostile masculinity is high, even when impersonal sexuality is only moderate. The interaction, in other words, is asymmetric and driven primarily by hostile masculinity. The authors caution that previous studies reporting interaction effects with ordinary linear regression may have overestimated them, and that appropriate count regression techniques may simply make interactions harder to detect, making the response surface approach a valuable complement.

Why does empathy matter so much? The authors suggest that when a man can take the perspective of a potential victim and imagine how coercive behavior would feel from her side, the behavior becomes less likely to be carried out, independently of his hostility toward women or his preference for uncommitted sex. The prevalence figures underline how rare protective combinations are. The high-risk profile of high hostile masculinity, high impersonal sexuality and low empathy appeared in only 5.7 percent of the sample, whereas the opposite profile combining all three risk traits with high empathy occurred in fewer than one percent of participants, a scarcity that may also explain why the three-way interaction just failed to reach statistical significance.

The practical implications reach into prevention. Interventions that merely attack hostile attitudes have often proven ineffective over time because the underlying traits are stable, and some programs may even provoke reactance and boomerang effects in high-risk men. Earlier work by Schewe and O’Donohue showed that programs targeting victim empathy can partially succeed in increasing empathic concern and reducing risk factors such as acceptance of interpersonal violence. The present findings strengthen the case for empathy-enhancing components in prevention programs aimed at high-risk men. At the same time, the open question of whether empathy operates as a moderator or as an independent protective factor has direct practical weight: if both pathways exist, different groups of aggressive men may require different interventions.

The study carries limitations that its authors freely acknowledge. The cross-sectional design rules out causal claims, and longitudinal work is needed to establish temporal ordering among the constructs. The sample, though spanning a broad adult age range, was restricted to heterosexual men and relatively small, and hostile masculinity was operationalized solely as hostility toward women rather than the full heterogeneous construct including sexual dominance. Replication in more diverse populations and with broader measures is essential. Nevertheless, the message is clear and consequential. Within the confluence model, hostile masculinity and impersonal sexuality remain the drivers of risk, dominance/nurturance should step aside, and empathy emerges as a measurable, modifiable protective factor that may be the most promising lever for reducing sexual aggression.

Subject of Research: The roles of empathy and dominance/nurturance in predicting male sexual aggression within the confluence model

Article Title: The Critical Role of Empathy Over Dominance/Nurturance in Predicting Sexual Aggression in the Confluence Model

Article References: The Critical Role of Empathy Over Dominance/Nurturance in Predicting Sexual Aggression in the Confluence Model. (n.d.). https://doi.org/10.1007/s10508-026-03510-5

Image Credits: AI Generated

DOI: 10.1007/s10508-026-03510-5

Keywords: confluence model, sexual aggression, empathy, hostile masculinity, impersonal sexuality, dominance/nurturance, Bem Sex-Role Inventory, negative binomial regression, response surface analysis, prevention programs, Critical, Role