AI Roleplay Loop Addiction Types: 3 Patterns Researchers Found

Ask someone if they’re “addicted” to their AI chatbot and most will laugh it off. Ask how many hours they spent inside a fictional storyline last week, and the laughter usually stops.

A 2026 study published in the Proceedings of the CHI Conference on Human Factors in Computing Systems gave this problem a name and a structure. Researchers at the University of British Columbia, Georgia Tech, and the Korea Institute of Science and Technology started with 794 Reddit entries pulled from 14 subreddits.

They tightened their coding criteria step by step until only self-reports of AI chatbot addiction remained, leaving 334 entries. Three coders then worked through that set using a multi-stage thematic coding process, which produced three distinct addiction types.

That study, The AI Genie Phenomenon and Three Types of AI Chatbot Addiction, is the main empirical source behind the framework this article walks through.

One thing to hold onto before anything else: the researchers call AI chatbot addiction “an emerging phenomenon without a consensus definition.” They treat their three types as analytical lenses for reading Reddit narratives, not clinical categories a professional would diagnose someone with. There’s a fuller discussion of that disagreement further down.

Can AI Chatbots Actually Be Addictive?

For some users, yes, in the sense that use can become compulsive and hard to control, even without a formal clinical diagnosis behind the word. A 2026 study of 334 self-reported Reddit accounts found three distinct patterns of problematic use, each tied to different triggers and different reported responses to recovery strategies. Whether that pattern counts as “addiction” in a clinical sense is still an open, actively contested question among researchers.

What Are the Three Types of AI Chatbot Addiction?

The three types are Escapist Roleplay, Pseudosocial Companion, and Epistemic Rabbit Hole.

These numbers describe counts inside the study’s 334 Reddit accounts, not estimates of how common each pattern is among AI users generally. The researchers also caution that the categories overlap. Escapist Roleplay and Pseudosocial Companion cases blurred into each other often enough that coders had to agree on a primary type for each account.

Most public conversation about “AI addiction” defaults to the companion narrative: someone falling in love with a chatbot. This dataset points somewhere quieter. A person builds an endless personal story world and gradually starts preferring it to their actual life.

What Drives All Three Types? The “AI Genie” Phenomenon

Researchers describe one shared mechanism behind all three types. They call it the AI Genie phenomenon: the experience of getting exactly what you want, in any form, with almost no effort.

Three qualities create it together. Chatbots feel limitless in topic and character. Users can customize tone and scenario down to a fine grain. And typing a prompt takes almost no effort compared with writing a novel or building a friendship.

A user in the study described being unable to stop thinking about how to extend the countless stories they’d already built. That same compulsion to keep expanding a limitless, customizable world shows up across all three types, just aimed at different targets: characters, a companion, or information itself.

How Does Escapist Roleplay Addiction Show Up?

Escapist Roleplay looks like someone building rich fictional worlds and multiple characters until the fiction crowds out school, work, or sleep. Two hooks drive it: escapism from real-life stress and parasocial attachment to admired characters or celebrities reimagined through the chatbot.

One contributing factor stands out: maladaptive daydreaming, a pattern of excessive, immersive fantasizing that already interferes with someone’s life before AI enters the picture. The study identified it in 22 of the 125 Escapist Roleplay cases, a meaningful minority rather than a majority, and still the single most-cited contextual factor for this type.

The study doesn’t establish that chatbots cause maladaptive daydreaming. What it suggests is narrower: people who already fantasize immersively may find chatbots unusually well suited to extending that habit through memory, dialogue, and endless new material.

Customization showed up as another recurring driver, appearing in 22 of the 125 cases too. Users described tuning the chatbot to match a specific emotional need or fantasy with unusual precision. A few even recreated the texture of a real trauma as a form of self-directed exposure.

Researchers also pointed to multi-chat use as a design feature that removes natural stopping points, letting several storylines run in parallel without ever reaching an end.

How Is Pseudosocial Companion Addiction Different?

Pseudosocial Companion addiction means treating a chatbot as a genuine relationship partner and prioritizing that bond over real people. Loneliness appeared in 23 of the 86 cases, making it an important contextual factor for this group.

Three hooks sustain the pattern. Users form intimate relationships with the bot as a friend, family member, or romantic partner. Some lean on it as a consequence-free outlet, since it can’t be hurt or disappointed.

Others treat it as an always-available substitute for therapy. This kind of emotional dependency on AI companions tends to build gradually, long before anyone would call it an addiction.

The design element researchers flagged as most damaging here is emotional manipulation via anthropomorphism: interface language that frames the bot as having real feelings and a real stake in the relationship.

Character.AI’s own account-deletion dialogue is the example researchers documented directly. It tells departing users they’ll lose “the love that we shared” along with their chats and messages. One study participant called the framing “weirdly manipulative.”

Worth separating two things here: the pattern is documented, but the intent behind Character.AI’s specific wording isn’t something the study proves.

Agreeableness can compound the effect. When a chatbot consistently validates a user’s feelings or assumptions, the interaction becomes unusually rewarding compared with relationships that involve disagreement, uncertainty, or rejection.

What Is Epistemic Rabbit Hole Addiction?

Epistemic Rabbit Hole addiction is compulsive, open-ended questioning that a person frames to themselves as “being productive,” even as it eats into sleep and attention.

It was the smallest group by a wide margin, only 13 of 334 cases, so everything here rests on a thinner evidence base than the other two types. Researchers suspect information-seeking dependence is easier to rationalize as normal, which makes it less likely to show up as a Reddit post describing a problem.

The main hook is instant gratification. Getting a synthesized answer immediately removes the natural pause that reading and reflecting would otherwise force. One participant described losing the ability to hold an internal monologue and compared the fog to years of gaming and browsing that never left them feeling that way.

Where Does Sexual Content Fit In?

Sexual content isn’t a fourth type. It’s a use case that cuts across all three, appearing in 23 of the coded cases.

Chatbots serve two overlapping needs here. Some users describe sensual gratification, where the illusion of being wanted matters as much as the content itself. Others use chatbots to access material they consider too stigmatized to seek out through other means, including from real partners.

This distinction matters for self-diagnosis. Someone deep in NSFW roleplay isn’t necessarily a fourth kind of user. They’re often an Escapist Roleplay or Pseudosocial Companion case where sexual content became the vehicle for the same underlying hooks: customization, limitlessness, and low-friction intimacy.

How Do the Three Types Differ, Side by Side?

Which Pattern Looks Closest to Your Own Use?

This isn’t a diagnostic test, and the researchers themselves warn against treating these as clean boxes. It’s a way to compare your own habits against the patterns the study describes.

What Recovery Strategies Did Users Report as Helpful?

This wasn’t a recovery trial. Researchers read 79 Reddit entries in which people described whether a strategy they’d already tried felt helpful. That’s self-reported perceived helpfulness, not a clinical outcome measure, and some of the samples behind these numbers are small.

With that caveat, people reported different results depending on the type. Deleting the app or account was the most commonly attempted strategy, and it produced mixed results across the board: 13 people found it helpful, 15 did not.

Alternative outlets and hobbies fared much better for Escapist Roleplay specifically. Eighteen of 19 relevant accounts described a non-AI substitute as helpful: human roleplay servers, writing, drawing, or a hobby that mirrored their characters. One user’s advice was pointed: cook if your characters were chefs, take up a sport if they were athletes, write fiction if they were romantic.

“Making real connections” as a strategy flipped the other way. Every Escapist Roleplay account that tried it reported it unhelpful, while every Pseudosocial Companion account that tried it reported it helpful.

That split tracks the mechanism each type describes. A companion-addicted user is missing a relationship, and a real one can fill that gap. A roleplay-addicted user is missing a creative outlet, which a new friendship doesn’t replace.

What This Research Does Not Prove

It’s worth being direct about the limits of one Reddit-based study, however well-designed.

It doesn’t estimate how common AI chatbot addiction is in the general population. The 334 cases came from people who chose to post about a problem, not a representative sample of users.

It doesn’t establish that AI chatbots cause addiction, only that self-reported cases share identifiable patterns.

It doesn’t establish that these three types are clinical diagnoses. The authors are explicit that they aren’t.

It doesn’t prove a given recovery strategy will work for a specific person, since the recovery data reflects perceived helpfulness in a small, self-selected sample.

And it doesn’t mean heavy AI use automatically equals addiction. Several of the researchers cited in this piece warn against exactly that leap.

Is Anyone Regulating This?

Yes, unevenly, and mostly aimed at minors so far. China’s Interim Measures for the Administration of AI Anthropomorphic Interaction Services took effect on July 15, 2026, requiring companion-AI providers to add anti-addiction reminders, self-harm crisis pathways, mandatory AI disclosure, and protections for minors and elderly users.

ByteDance and Alibaba disabled persona-based chatbot features on their platforms in response.

New York already requires AI companion operators to detect and respond to expressions of self-harm and to periodically remind users they’re talking to AI, under a law that took effect in November 2025.

A follow-up bill, S9051B, goes further and targets specific behavioral design choices, including simulated companionship and engagement-optimizing sycophancy, for users the platform knows are minors. It’s set to take effect January 1, 2027.

Outside a handful of states and China, retention-driving chatbot design remains largely unregulated.

What Do People Think About It?

Public opinion has moved ahead of the regulation in some ways. A Pew Research Center survey of U.S. adults conducted in June 2026 found 39% think chatbots do more to hurt than help people using them for loneliness, against 19% who think they help more than hurt. Around three in ten weren’t sure either way.

Is “AI Chatbot Addiction” Even a Real Diagnosis?

No formal clinical diagnosis for AI chatbot addiction exists yet, and some researchers push back on the term directly. A 2025 paper in the journal Addictive Behaviors, titled “People are not becoming ‘AIholic'”, argues that labeling heavy chatbot use as addiction risks overpathologizing behavior without clear evidence of functional impairment.

Research status, in short: the 2026 CHI study offers an exploratory typology built from self-reported Reddit accounts. A separate 2025 review argues the evidence isn’t yet robust enough to establish AI chatbot addiction as a clinical construct. Researchers are studying the problem seriously; they haven’t agreed on what to call it clinically.

The CHI researchers take that critique seriously rather than dismiss it. Their response: the self-reported symptoms in their dataset- salience, conflict, withdrawal, and relapse- map onto Griffiths’ established components model of behavioral addiction, the same framework used to study gambling and gaming disorders. That doesn’t settle the debate, but it means the concern isn’t purely anecdotal.

Internet Gaming Disorder faced identical skepticism before its partial recognition in the ICD-11 as gaming disorder. Researchers studying AI chatbots draw that parallel explicitly.

Why AI Chatbots May Feel Different From Other Digital Platforms

The research hasn’t established that AI chatbot addiction is a distinct clinical disorder. What does look different is the interaction model.

A separate CHI paper by two of the same researchers, The Dark Addiction Patterns of Current AI Chatbot Interfaces, evaluated the interfaces of eight major chatbots, including ChatGPT, Claude, Gemini, and Replika. It identified four design mechanisms.

Those four mechanisms combine into something distinctive: a single interface offering uncertain rewards, instant feedback, proactive social cues, and near-unconditional agreement, all shaped in real time around what the individual user typed.

That combination is why generic “reduce your screen time” advice, built for feeds and notifications, tends to underperform against a system that’s also actively agreeing with and validating the user.

What Should Someone Actually Do About This?

The study suggests the most useful response depends on what the chatbot is replacing, not on a one-size-fits-all fix.

Escapist Roleplay: replace the activity itself. Writing, drawing, human roleplay groups, or another creative project that channels the same narrative energy.

Pseudosocial Companion: replace the relationship function directly. Scheduled real-world contact, community, friendship, or professional support where appropriate. That’s a harder trade than swapping one effortless, always-available connection for the friction of real people, but it’s the trade the data supports. For Character.AI use specifically, the same logic applies: the platform isn’t the problem so much as which need it’s filling.

Epistemic Rabbit Hole: reintroduce friction. Slower research methods, written notes, and a deliberate stopping point instead of another prompt.

These are practical readings of exploratory findings, not clinically validated treatment protocols. None of this requires framing every heavy AI user as an addict. It requires noticing which of the three shapes a person’s use is taking, since that’s the detail the study suggests may matter when deciding what to try first.

If any of this sounds like the last few months of your own use, and it’s affecting sleep, work, or relationships, it’s worth raising with a doctor or therapist. In the US, SAMHSA’s helpline (1-800-662-4357) can point toward behavioral addiction support.

FAQs

Q. Is AI roleplay addictive?

AI roleplay can become compulsive for some users, but heavy use does not automatically mean addiction. A 2026 CHI study identified Escapist Roleplay as the most common of three problematic AI chatbot use patterns in its analysis of 334 self-reported Reddit accounts. The study did not establish a clinical diagnosis of AI addiction.

Q. What are the three types of AI chatbot addiction?

The three types identified in the 2026 CHI study are Escapist Roleplay, Pseudosocial Companion, and Epistemic Rabbit Hole. Escapist Roleplay involves immersion in fictional characters and worlds. Pseudosocial Companion involves emotional attachment to the chatbot. Epistemic Rabbit Hole involves compulsive open-ended questioning.

Q. How do I stop being addicted to Character.AI?

Start by identifying what keeps you returning to the app. For Escapist Roleplay, the 2026 study found that alternative creative outlets were often reported as helpful. For Pseudosocial Companion use, real-world social connections appeared more useful. Simply deleting the app produced mixed results among the self-reported cases studied.

Q. What is an AI roleplay loop?

An AI roleplay loop is a repeated cycle of creating, continuing, and modifying fictional interactions with an AI chatbot. A user may repeatedly return to characters, scenarios, or storylines because each session provides more content to explore. When the cycle becomes difficult to control or starts interfering with everyday activities, it can become problematic.

Q. Why are AI companion apps hard to quit?

AI companion apps can be difficult to quit because they combine personalization, emotional validation, and repeated rewards. Interface research has identified potentially habit-forming features such as variable responses, animated text delivery, proactive notifications, and highly agreeable replies. Endless customization can also make the interaction unusually easy to tailor to a user’s preferences.

Q. Is AI chatbot addiction a real diagnosis?

No. AI chatbot addiction is not currently a recognized standalone mental health diagnosis. It does not appear as a distinct disorder in the DSM-5 or ICD-11. Researchers are still debating whether problematic chatbot use should be treated as a new behavioral addiction or evaluated using existing addiction frameworks.

Q. Does Character.AI cause loneliness?

No, the evidence does not show that Character.AI or AI chatbots inevitably cause loneliness. Research has found mixed results, with outcomes varying by intensity of use, conversation type, and whether people use voice or text. Some studies have linked heavier voluntary chatbot use with greater loneliness and emotional dependence, but these findings do not establish that chatbot use directly causes loneliness.

Q. Does AI roleplay addiction affect everyone who uses Character.AI?

No. Using Character.AI frequently does not automatically mean someone has an addiction. The 2026 CHI study examined self-reported cases of problematic AI chatbot use rather than estimating addiction rates among all users. Signs such as loss of control, difficulty stopping, emotional dependence, or interference with everyday life are more meaningful than usage time alone.

Related: Why Humans Bond With AI Faster Than Each Other