A deconstruction of Alfassi et al.’s “Fanfiction in the Age of AI” for the culturally curious.
The Paper in Brief
Citation: Alfassi, R., Cooper, A., Mitchell, Z., Calabro, M., Shaer, O., & Mokryn, O. (2025). Fanfiction in the Age of AI: Community Perspectives on Creativity, Authenticity and Adoption. Accepted for publication in the International Journal of Human-Computer Interaction, June 2025. arXiv:2506.18706v1.
The Study: A structured online questionnaire of 157 active fanfiction community members (90 writers, 67 readers), conducted May–July 2024, distributed via Dreamwidth, Discord servers, email lists, and social media. After aggressive bot filtering (332 suspected bots removed from an initial 489 respondents), the final dataset of 157 showed strong internal consistency (Cronbach’s α = 0.85).
The Core Question: How does the fanfiction community—one of the largest unpaid creative ecosystems on Earth—perceive the integration of Generative AI into its spaces?
Deconstructed: The Lower-Level Patterns
Pattern 1: The Demographics Are a Political Statement
Before the authors even get to AI, the participant profile tells its own story:
- 65% women, 16.7% non-binary, 12.7% men. The gender skew is not incidental; it is the precondition for everything that follows.
- 81.5% aged 18–34, with 46.5% having been in the community for over a decade. These are not dabblers. These are lifers.
- 80.9% have at least an undergraduate degree. This is a highly educated population performing highly skilled labor for free.
- AO3 dominates at 84.1%. The platform is not neutral infrastructure; it is the architecture of a specific value system (nonprofit, fan-run, anti-commercial).
The pattern: Fanfiction is not a hobby. It is a career of care—unpaid, long-term, emotionally invested, and disproportionately performed by people (women and nonbinary individuals) whose creative labor has historically been devalued in commercial markets. This is the population now being asked to accept AI into their workspace.
Pattern 2: Motivation Is Social, Not Commercial
The study breaks down why people write, and the hierarchy is revealing:
- Character exploration (86.7%)
- Non-canonical relationships (76.7%)
- Expanding the fictional world (61%)
- Developing writing skills (61%)
- Engaging with other fans (56.6%)
- Easier publishing than the industry (43%)
The pattern: “Easier publishing” ranks last. Fanfiction writers are not failed novelists seeking a backdoor into the industry. They are people who want to explore and connect. The gift is the point. The kudos, the comments, the sense of being read by someone who needed exactly what you wrote—these are the currencies. AI threatens not the paycheck (there is none) but the social contract.
Pattern 3: The “Social” in Social Reading
74.5% of participants view reading fanfiction as a social activity; 73% view writing as social. This is not intuitive—fanfiction reading is typically a solitary act, performed on a screen, often in bed at 2 AM. But the study reveals that the community experiences even silent consumption as participatory. Reading is a form of witness. When you read a hurt/comfort fic at 2 AM and leave a kudos, you are performing a social function: you are telling a stranger I saw what you made, and it mattered to me.
The pattern: AI-generated content breaks this circuit. A story written by ChatGPT and posted without disclosure is not just “content.” It is a broken promise. The reader who leaves emotional feedback is performing intimacy with a void. This is why transparency emerges as the community’s central ethical demand.
Pattern 4: The Three Scenarios of Disruption
The paper usefully frames the community’s fears through De Cremer et al.’s three scenarios:
| Scenario | Description | Community Response |
|---|---|---|
| Augmentation | AI assists human creators, reducing effort | Cautiously accepted; some see value for idea generation |
| Flooding | Cheap AI content drowns out human voices | Deeply feared; threatens the gift economy’s scarcity logic |
| Recognition | Human uniqueness is valued and preserved | The aspirational endpoint; what the community wants |
The pattern: The community is not anti-technology. It is anti-replacement. Writers already use AI tools—74.4% report using at least one (Grammarly, ChatGPT, etc.). The line is drawn not at use but at automation. When AI moves from “helping me write” to “writing for me,” the social fabric tears.
Pattern 5: Data Colonialism and the Unconsented Archive
Perhaps the most theoretically sharp section of the paper is its engagement with the ethics of training data. The authors cite Gero et al. (2025) and Couldry & Mejias (2020) to frame a devastating pattern:
- Fanfiction is produced without financial compensation, as “a form of social connection or identity expression.”
- LLMs are trained on this corpus, often via web scraping, without consent.
- The result is data colonialism: the extraction of unpaid, emotionally embedded labor from a marginalized community (women, queer people, nonbinary individuals) for the profit of commercial AI systems.
The pattern: The same demographic that is underpaid in the formal economy is now having its voluntary creative labor harvested to build tools that may eventually replace it. The irony is almost too neat: the emotional labor of women and queer people becomes training data for machines that will generate synthetic emotion back to them.
Pattern 6: The Design Gap
The paper concludes with design interventions, but the underlying pattern is more interesting: the platforms do not know what they are hosting. AO3, FanFiction.net, and their ilk were built for a human-only ecosystem. They have no native architecture for marking AI-assisted content, for consent-based data governance, or for preserving the “gift economy” against automation. The study’s call for transparency, ethical integration, and community engagement is, in essence, a call for platform literacy—recognizing that infrastructure shapes culture.
Synthesized: The Essay
When the Ghost in the Machine Reads Your Diary
On AI, fanfiction, and the crisis of the unpaid heart.
In May of 2024, a team of researchers sent a questionnaire into the fanfiction community. They were looking for something simple: what do writers and readers think about AI? What they found was a community in the early stages of mourning.
The respondents were, by any measure, an extraordinary group. Eighty-one percent were between 18 and 34. Nearly half had been writing or reading fanfiction for over a decade. Eighty percent had college degrees. Two-thirds were women; one in six was nonbinary. They were highly educated, highly engaged, and—crucially—highly unpaid. They wrote not for money but for connection: to explore characters the canon had neglected, to imagine relationships the source material had forbidden, to spend more time in worlds they loved with people who loved them too.
And now, they were being asked to share those worlds with machines.
The study, Fanfiction in the Age of AI by Alfassi et al., is not a polemic. It is a careful, quantitative and qualitative portrait of a community at an inflection point. But read between the lines, and what emerges is a story about labor, consent, and the quiet violence of extraction.
The Gift and the Algorithm
Fanfiction operates on what academics call a “gift economy.” Writers produce stories; readers respond with kudos, comments, bookmarks, and recommendations. No money changes hands. The currency is attention, recognition, and emotional resonance. A writer who posts a 200,000-word slow-burn romance is not performing a transaction. They are making an offering.
This is why the community’s response to AI is not simply “we don’t like robots.” It is far more specific. The study finds that 74.4% of writers already use AI tools in some capacity—Grammarly, ChatGPT for brainstorming, search engines for research. The community is not Luddite. What it resists is automation: the posting of fully AI-generated stories without disclosure, the flooding of archives with synthetic content, the erosion of the social contract that makes the gift meaningful.
When you leave a comment on a fanfiction story, you are not just praising prose. You are witnessing another person’s emotional labor. You are saying: I see the time you spent. I see the care. If the story was written by a large language model, that comment becomes a kind of betrayal—intimacy offered to a void. The reader has been tricked into performing connection with something that cannot connect back.
This is why transparency is the community’s central ethical demand. Not because AI is inherently evil, but because undisclosed AI is inherently deceptive. It breaks the circuit of reciprocity that holds the community together.
The Colonialism of the Training Set
But the deeper wound is not what AI might do to fanfiction in the future. It is what AI has already done to fanfiction in the past.
The study draws on work by Gero et al. and Couldry & Mejias to frame a devastating reality: the large language models now being used to generate fanfiction were trained, in part, on fanfiction itself. Millions of stories—written for free, shared in trust, often exploring queer and marginalized identities—were scraped from AO3 and FanFiction.net without consent and fed into commercial systems.
This is not theft in the traditional sense. It is something more insidious: data colonialism. The creative labor of a community that is already economically and culturally marginalized—women, nonbinary people, queer writers—has been extracted to build tools that may eventually render their voluntary labor obsolete. The machine learned to simulate emotion by reading the diaries of people who were giving their hearts away for free.
The irony is almost architectural. Fanfiction exists because mainstream media failed to tell the stories this community needed. So the community told them itself, in a space it built itself, under a nonprofit charter, protected from commercial exploitation. And then the commercial systems came anyway—not through the front door, but through the API.
Three Futures
The study organizes the community’s fears into three scenarios, borrowed from De Cremer et al. In the first, AI augments human creativity: a useful tool, a second mind. In the second, AI floods the landscape with cheap, personalized content, reducing humans to curators. In the third, human uniqueness is recognized and preserved.
The fanfiction community, the study suggests, is cautiously open to the first, terrified of the second, and desperately hoping for the third. But hope is not a strategy. The second scenario—flooding—feels increasingly inevitable. If AI can generate passable hurt/comfort fic in seconds, and if platforms have no mechanism to distinguish synthetic from human work, the gift economy collapses under the weight of its own abundance. Scarcity is what gives the gift value. When stories are infinite, attention becomes worthless.
What This Means for the Rest of Us
Fanfiction is often treated as a niche interest, a subculture, a footnote to “real” literature. But the study makes clear that it is something else: a canary in the coal mine for every creative community now facing AI integration. The dynamics at play in AO3—unpaid labor, emotional investment, social reciprocity, consentless data extraction—are the same dynamics facing journalists, illustrators, musicians, and poets.
The difference is that fanfiction writers have no union, no guild, no collective bargaining power. They have only norms. And norms are fragile things when platforms are designed without them in mind.
The researchers end with a call for design interventions: transparency tools, ethical AI integration, community engagement mechanisms. But the subtext is bleaker. The infrastructure of the internet was not built to distinguish between human and machine creativity. The platforms that host fanfiction were designed for a world where “author” meant “person.” That world is ending.
What the study ultimately documents is not a technological transition but a grief process. A community that has spent decades building a sanctuary for the stories mainstream culture refused to tell is now watching that sanctuary be colonized by the very systems that refused to tell them. The machine has read their diaries. Now it wants to write them back.
Coda: The Kudos Problem
There is a moment in the study that haunts me. The researchers note that fanfiction is “seen by its members as a socially embedded and emotionally supportive practice.” This is the language of sociology. But what it describes is something simpler and more fragile: the feeling of being known.
When you post a story and someone leaves a comment that says “this is exactly what I needed tonight,” you are not receiving feedback. You are receiving recognition. You are being told that your unpaid labor mattered to another human being. No algorithm can replicate that. But algorithms are very good at making us forget to ask for it.
The fanfiction community’s resistance to AI is, at its core, a resistance to forgetting. It is a demand that we remember what creativity is for when no one is paying: not product, but presence. Not content, but connection. The kudos button is small and orange and costs nothing to click. But it is also, in its quiet way, a vote for the human.
The machines are learning to write. The question is whether we are learning to read.
Sources: Alfassi et al. (2025), “Fanfiction in the Age of AI,” arXiv:2506.18706v1; Gero et al. (2025); Couldry & Mejias (2020); De Cremer et al. (2023); 2024 AO3 Census.
