How Do Fans Use AI Social Chatbots? A Multi-Method Exploration of Fan Accounts

It may state two separate claims as one.

Evidence from the thematic and n-gram analyses suggested that fandom engagement and social motivations were identified by users as important drivers of use, while utilitarian functions also played a role.

“Evidence from the thematic and n-gram analyses suggested that fandom engagement and social motivations were identified by users as important drivers of use, while utilitarian functions also played a role.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

After three iterations of independent review and collaborative refinement, the authors converged on a final set of themes representing 25 recurring motives.

“After three iterations of this process, we had converged on our understanding of the themes present in each comment we discussed; this final set of themes represented 25 recurring motives.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The resulting 813 comments had a mean length of 30.82 words.

“The resulting 813 comments identified by the research assistants had a mean length of 30.82 words.”

Probably not a substantial claim.

Virtual Social Compensation motives cover users who utilize Character.ai to alleviate a lack of social experiences.

“The next broad category of motives was Virtual Social Compensation, which covers users who utilize Character.ai as a method to alleviate a lack of social experiences.”

• AssumptionRests on an assumption: Users' comments accurately reflect their motivations for using Character.ai.

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The IRL Social Contact category includes using the website alongside real-life friends to facilitate real-life social development.

“The IRL Social Contact category also captures motives related to using the website alongside real-life friends, allowing for real-life social development and facilitation, as well.”

• AssumptionRests on an assumption: Users' comments accurately reflect their motivations for using Character.ai.

It may state two separate claims as one.

The data suggest that many fans engage in extension of text without community engagement, reporting they use C.ai specifically because it does not involve other people.

“Our data suggest, however, that many fans, when given the opportunity, engage in this extension of text without the community engagement, in fact reporting that they use C.ai to do so specifically because it does not involve other people.”

• AssumptionRests on an assumption: Users' comments accurately reflect their motivations for using Character.ai.

It may state two separate claims as one.

Fans reported using chatbot characters for involvement with the main story, such as self-insertion, or for parasocial attachment to a specific media figure.

“This category captures individuals who reported utilizing the chatbot characters for involvement with the main story, such as engaging with the world as a self-inserted character, or for those who have a parasocial attachment to a specific media figure and would like to engage with them specifically.”

• AssumptionRests on an assumption: Users' comments accurately reflect their motivations for using Character.ai.

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The comments ranged in date of creation from November 2023 to January 2026; most (782, or 96%) were posted in 2025, 5 (0.6%) near the end of 2023, 11 (1.1%) in 2024, and 13 (1.6%) in 2026.

“They ranged in the date of their creation from November 2023 to January 2026; most (782, or 96%) of the comments were posted in 2025, 5 (0.6%) were made near the end of 2023, 11 (1.1%) in 2024, and the balance (13, or 1.6%) were made in 2026.”

Probably not a substantial claim.

Fan Facilitation motives indicated using Character.ai to engage with preferred fandom, character, or narrative adjacent or continuous to the main story.

“Fan Facilitation motives were those that indicated motives to use Character.ai as a method of engaging with their preferred fandom, character, or narrative adjacent or continuous to the main story.”

• AssumptionRests on an assumption: Users' comments accurately reflect their motivations for using Character.ai.

It may state two separate claims as one.

Many fans who admitted to darker motives acknowledged they are problematic but believed acting them out with AI chatbots harms no one.

“Many of the fans who admit to these darker motives acknowledge that they are problematic but recognize that acting them out with AI chatbots harms no one, as the fan who reported that “I know it’s not right…, but it’s just what I need to do to be safe from myself.””

• AssumptionRests on an assumption: Users' comments accurately reflect their motivations for using Character.ai.

It may state two separate claims as one.

Enjoyment was reflected in character interaction, worldbuilding, roleplaying, and customizing bots for others.

“That enjoyment is reflected in many different types of interaction, including things like character interaction (e.g., “I just like chatting with characters in a show or anime”), worldbuilding, roleplaying in general (e.g., “I create my own stories with characters I love—for fun”), or even customizing bots for others to use (e.g., “I like creating bots for others to enjoy”).”

• AssumptionRests on an assumption: Users' comments accurately reflect their motivations for using Character.ai.

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Through further discussion and review, the authors came to a consensus on a group of five overarching motives under which each of the 25 could be grouped.

“Through further discussion and review of specific exemplars, we came to a consensus on a group of five overarching motives under which each of the 25 could be grouped.”

It may state two separate claims as one.

Several fans reported talking with bots enacting favorite performers who had passed away, and found the experience comforting or healing.

“Others used it for a particular parasocial purpose, including several fans who reported talking with bots enacting favorite performers who had passed away, finding the experience to be comforting or healing.”

• AssumptionRests on an assumption: Users' comments accurately reflect their motivations for using Character.ai.

It may state two separate claims as one.

The thematic analysis identified 25 definite themes and generated five categories of motivations: Fan Facilitation, Safe Harbor, Virtual Social Compensation, IRL Social Contact, and Entertainment.

“After identifying 25 definite themes, we generated five categories based on the identified motivations: Fan Facilitation, Safe Harbor, Virtual Social Compensation, In Real Life (IRL) Social Contact, and Entertainment.”

• AssumptionRests on an assumption: Users' comments accurately reflect their motivations for using Character.ai.

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Analogies were present in 2.55% of the data.

“For parasocial interaction, we looked at our parasocial attachment and analogy categories, with parasocial attachment language being present in 0.159% of the text corpus and analogies being present in 2.55% of the data.”

Probably not a substantial claim. It may state two separate claims as one.

These five overarching themes represented distinct motives or drivers of C.ai use rather than distinct and exhaustive classifications of comments; a single comment might convey a single theme, multiple themes, or, rarely, no theme at all.

“These five overarching themes represented distinct motives or drivers of C.ai use rather than distinct and exhaustive classifications of comments—that is, a single comment might convey a single theme, multiple themes, or, rarely, no theme at all.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

A review by Hung et al. (2026) identified benefits of parasocial relationships with social chatbots, including emotional support and social needs, enjoyment, and community belonging and participation.

“A review by Hung et al. (2026) identified benefits of what they characterized as parasocial relationships with social chatbots, including emotional support and social needs, enjoyment, and community belonging and participation.”

• cites Hung et al., 2026 sourceCheck pending

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Many studies on social AI chatbots strongly underscore their use to meet social needs or perform social functions, including the experience of romantic or para-romantic relationships.

“Many of the studies on social AI chatbots strongly underscore their use to meet social needs or perform social functions, including the experience of romantic or para-romantic relationships (e.g., Ma & Koike, 2025).”

• cites Ma & Koike, 2025 sourceNot checked — no full text of the source

It may state two separate claims as one.

Fan C.ai use has a dual character, namely as a tool to both meet social needs and avoid social contact, with potential benefits and hazards for fans engaging with social chatbots.

“Findings are discussed in terms of the dual character of fan C.ai use, namely as a tool to both meet social needs and avoid social contact, with its potential benefits and hazards for fans engaging with social chatbots.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

A review by Hung et al. (2026) identified a pattern in which use can lead to emotional and behavioral dependency and the displacement of human relationships.

“On the other hand, they also identified a pattern in which use can lead to emotional and behavioral dependency and the displacement of human relationships.”

• cites Hung et al., 2026 sourceCheck pending

Probably not a substantial claim. It may state two separate claims as one.

The sample was gathered by two undergraduate research assistants who were familiar with Character.ai and identified as fans of various fictional worlds and performers but were naïve to the purposes of the study.

“The sample was gathered by two undergraduate research assistants who were familiar with Character.ai and identified as fans of various fictional worlds and performers but were naïve to the purposes of the study.”

• AssumptionThe authors state this as an assumption: Research assistants who were both fans and naïve to the study's rationale would be able to identify conversations among fans the way any fan investigating the question might.

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Persistent use of social chatbots over time may lead to increased feelings of loneliness or isolation.

“At the same time, persistent use over time may lead to increased feelings of loneliness or isolation (Folk & Dunn, 2026).”

• cites Folk & Dunn, 2026 sourceCheck pending

Probably not a substantial claim. The paper’s conclusions may not depend on it.

A chatbot partner that provides a facsimile of a relationship with a favorite character could make fan creation more enjoyable.

“Fans extend and interpret texts through fan fiction and the creation of fan art (Jenkins, 1992); a chatbot partner that provides a facsimile of a relationship with a favorite character could make such creation more enjoyable.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

823 comments were initially identified and recorded, but ten comments found in Discord threads could not be verified as public and were discarded.

“823 comments were initially identified and recorded, but we were unable to verify the public character of ten of the comments, specifically those found in Discord (Discord, Inc., San Francisco, CA, USA) threads. These were therefore discarded.”

It may state two separate claims as one.

The findings highlight how social chatbots may be reshaping fan practices, parasocial engagement, and digitally mediated companionship.

“The findings highlight how social chatbots may be reshaping fan practices, parasocial engagement, and digitally mediated companionship.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Self-disclosure plays a pertinent role in advancing a connection with another individual, as it allows a setting where intimacy and amity can be cultivated.

“Self-disclosure plays a pertinent role in advancing a connection with another individual, as it allows a setting where intimacy and amity can be cultivated (Jiang et al., 2011).”

• cites Jiang et al., 2011 sourceNot checked — no full text of the source

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Social factors seem to drive self-disclosure in human-chatbot communication.

“In fact, it is these social factors that seem to drive self-disclosure in human–chatbot communication (Meng & Liu, 2025).”

• cites Meng & Liu, 2025 sourceNot checked — no full text of the source

It may state two separate claims as one.

The open-ended nature of chatbot discussions allows for a private and relatively unrestricted environment for users to discuss challenging topics without the threat of judgment, and experience emotional support in the form of positive affective encouragement and facilitating a sense of belonging, especially when it is not readily available in one's real-life.

“The open-ended nature of chatbot discussions allows for a private and relatively unrestricted environment for users to discuss challenging topics without the threat of judgment, and experience emotional support in the form of positive affective encouragement and facilitating a sense of belonging, especially when it is not readily available in one’s real-life (Ta et al., 2020).”

• cites Ta et al., 2020Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim.

The CASA paradigm suggests that humans utilize human-human communication heuristics and, to a certain extent, norms when participating in human-computer interaction.

“This can be understood with an appeal to the computers are social actors (CASA) paradigm (Nass et al., 1994; Nass & Moon, 2000), which suggests that humans utilize human–human communication heuristics and, to a certain extent, norms when participating in human–computer interaction.”

• cites Nass et al., 1994Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Nass & Moon, 2000 sourceNot checked — no full text of the source

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Social chatbots are designed in ways that make these agents capable of eliciting emotional bonds with their users.

“While many chatbots are embedded with skills that facilitate natural social interaction, social chatbots are designed in ways that make these agents capable of eliciting emotional bonds with their users (Henkel et al., 2020).”

• cites Henkel et al., 2020 sourceCheck pending

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

Concerns have been raised about the possibility of addiction to social chatbots, including by users who report experiencing addiction.

“In addition, concerns have been raised about the possibility of addiction to social chatbots, including by users who report experiencing addiction (Shen et al., 2026).”

• cites Shen et al., 2026 sourceCheck pending

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Taylor's (2015) observation that the overwhelming majority of fans of fictional texts are focused on engagement with the text, not engagement with communities of other fans.

“This is consistent with Taylor’s (2015) observation that the overwhelming majority of fans of fictional texts are focused on engagement with the text, not engagement with communities of other fans.”

• cites Taylor’s (2015) sourceNot checked — no full text of the source

It may state two separate claims as one.

The distinct lack of human interaction acts as both a protected, secure setting and a potential promoter of further social isolation.

“Whether this arises in the entertainment context of role-playing, or the social compensation context of searching for romance, the distinct lack of human interaction acts as both a protected, secure setting and a potential promoter of further social isolation.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

In reporting results, select words in quotations have been replaced with near synonyms to protect commenter privacy; what are presented as quotations are near-quotations.

“In order to protect the pseudonymous identities participants used in these conversations, select words in these quotations have been replaced with near synonyms. What we present as quotations are, in fact, near-quotations.”

Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

The authors identify as scholars who research fans from the perspective of psychology and communication, and as fans of a variety of fictional narratives; one has strong connections to fan communities and practices, the other is more of a casual fan.

“We, the authors, identify as scholars who research fans from the perspective of psychology and communication. We also identify as fans of a variety of fictional narratives. One of us has strong connections to fan communities and practices; the other is, in fan terms, more of a casual fan.”

It may state two separate claims as one.

The sample was not designed to be representative, so the results are not generalizable.

“Another key limitation is that, as the sample was not designed to be a representative one, the results are not generalizable; instead, the results should be viewed as offering insights for further, more systematic investigation.”

Probably not a substantial claim.

Jenkins (1992) tends to frame fan creative activity as essentially communal.

“Jenkins (1992) tends to frame fan creative activity as essentially communal—a way to engage with one’s fandom as fan fiction is written and read, fan art created and attended to, and so on.”

• cites Jenkins (1992)Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim. It may state two separate claims as one.

Given the inherently social nature of parasocial relationships, comparisons to real-life social relationships and connections were likely to be present in descriptions of connections with media figures, prompting the need for the inclusion of analogy language within this analysis.

“Given the inherently social nature of parasocial relationships, we reasoned that comparisons to real-life social relationships and connections were likely to be present in descriptions of connections with media figures, prompting the need for the inclusion of analogy language within this analysis (e.g., “he’s like the brother I never had,” “she feels like just another friend from my school”).”

• AssumptionRests on an assumption: Patterns of language in the comments indicate the categories of motives (parasocial uses, utilitarian uses, fan engagement).

Probably not a substantial claim. The paper’s conclusions may not depend on it.

All remaining comments were still publicly viewable at the time of writing.

“All remaining comments were still publicly viewable at the time of writing.”

Probably not a substantial claim. The paper’s conclusions may not depend on it.

The term "high school" follows the sub-theme of nostalgia.

“Additional terms, such as “high school”, follow the sub-theme of nostalgia, where those who utilize C.ai do so to experience a specific narrative (e.g., “I’m 18 myself, and I often find myself nostalgic for that era of my life, because I never really got to experience that perfect “high school dream” that so many stories romanticize”).”

• AssumptionRests on an assumption: The n-gram collocation analysis's top collocates are meaningful indicators of user experiences or motives.

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Empath can create novel dictionary categories based on provided seed words.

“In addition to Empath’s built-in categories, the NLP has the ability to create novel dictionary categories based on a provided set of seed words (Fast et al., 2016).”

• cites Fast et al., 2016 sourceNot checked yet

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Each dictionary was examined by the authors to ensure accurate representation of seed words.

“Each was then subject to examination by the authors to ensure accurate representation of seed words.”

Probably not a substantial claim.

The expectancy-value model of media gratifications holds that media use is motivated by needs and wants as informed by perceptions and beliefs about the medium.

“The expectancy-value model of media gratifications holds that media use is motivated by needs and wants as informed by perceptions and beliefs about the medium (Palmgreen & Rayburn, 1982).”

• cites Palmgreen & Rayburn, 1982 sourceNot checked — no full text of the source

Probably not a substantial claim. The paper’s conclusions may not depend on it.

Collocates such as "friend show", "group chat", and "social media" follow the IRL Social Contact category.

“Collocates such as “friend show”, “group chat”, and “social media” adequately follow the throughline of the IRL Social Contact category, with referrals from real-life friends, social networks, and introduction through external media exhibited within these terms.”

• AssumptionRests on an assumption: The n-gram collocation analysis's top collocates are meaningful indicators of user experiences or motives.

Probably not a substantial claim.

Recognizing the subjectivity involved in TA, the authors elected to complement that work with computational approaches to identifying themes in the data.

“Recognizing the subjectivity involved in TA, we elected to complement that work with computational approaches to identifying themes in the data.”

Probably not a substantial claim. The paper may give no reason why it holds.

The language used to describe parasocial relationships is the same language often used to describe social relationships.

“This is largely because the language used to describe parasocial relationships is the same language, often used to describe social relationships—romance, closeness, friendship, affinity, favorite, and so on.”