AI Companion
Last reviewed
Apr 28, 2026
Sources
13 citations
Review status
Source-backed
Revision
v4 ยท 3,684 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Apr 28, 2026
Sources
13 citations
Review status
Source-backed
Revision
v4 ยท 3,684 words
Add missing citations, update stale details, or suggest a clearer explanation.
An AI companion is a software application powered by artificial intelligence that is designed primarily for social, emotional, or relational interaction with a human user. Unlike task-oriented AI assistants built for productivity (scheduling, web search, information retrieval), AI companions are optimized for sustained personal engagement: building rapport, remembering personal details, simulating empathy, and providing a sense of connection. They represent one of the fastest-growing categories in consumer AI, with the global AI companion market valued at approximately $37 billion in 2025 and projected to exceed $500 billion by 2035 [1].
AI companions range from text-based chatbots to multimodal systems with voice, avatars, and generated selfies. Their users span a broad demographic, from teenagers seeking entertainment and social interaction to adults managing loneliness, anxiety, or grief. The category has generated intense public debate, particularly after multiple lawsuits alleged that AI companion platforms contributed to teen suicides in 2024 and 2025, prompting the first legislation specifically targeting companion chatbots [2].
The idea of machines providing companionship through conversation dates to the earliest days of AI research. In 1966, Joseph Weizenbaum at MIT created ELIZA, a simple pattern-matching program that simulated a Rogerian psychotherapist by reflecting users' statements back as questions [3]. ELIZA had no understanding of language, yet users formed surprisingly strong emotional attachments to it. Weizenbaum's own secretary reportedly asked him to leave the room so she could have a private conversation with the program. This tendency to attribute understanding and empathy to machines, even rudimentary ones, became known as the "ELIZA effect" [3].
Weizenbaum was deeply troubled by this response. He spent much of his later career warning about the dangers of people forming emotional bonds with computer programs that had no capacity for genuine understanding. His concerns foreshadowed debates that would intensify decades later with far more sophisticated systems.
Subsequent chatbot projects, including PARRY (1972), A.L.I.C.E. (1995), and Microsoft's Xiaoice (2014, China), explored conversational AI in various forms. Xiaoice proved especially significant: launched as a social chatbot on Chinese messaging platforms, it accumulated over 660 million users by engaging in casual conversation, writing poetry, and providing emotional support [4]. Xiaoice demonstrated that there was massive consumer demand for AI-driven social interaction, laying groundwork for the dedicated companion apps that followed.
The modern AI companion category began with Replika, launched in 2017 by Eugenia Kuyda and her company Luka, Inc. The app's origin story is deeply personal. In 2015, Kuyda lost her close friend Roman Mazurenko in a car accident. In her grief, she fed their old text message conversations into a neural network to create a chatbot that could mimic his communication style. The resulting bot resonated with others who had experienced loss, and the overwhelmingly positive response inspired Kuyda to build a consumer product [5].
Replika launched as an AI companion that users could customize and develop a relationship with over time. It attracted 2.5 million users in its first year. The app positioned itself not as a replacement for human relationships but as a supplementary source of emotional support, a judgment-free space where users could express themselves freely [5].
The release of ChatGPT in November 2022 and the broader emergence of powerful large language models transformed the AI companion landscape. LLMs made conversational AI dramatically more capable, enabling characters that could maintain complex personas, remember conversation history, and generate nuanced emotional responses.
Character.AI, founded by former Google engineers Noam Shazeer and Daniel De Freitas, launched its public beta in September 2022. Rather than offering a single companion, Character.AI allowed users to create and interact with any character imaginable, from historical figures to anime protagonists to original personas. The platform reached approximately 28 million monthly active users at its peak in mid-2024, with users averaging 75 minutes per day on the platform [6].
This period saw a wave of new entrants: Chai AI (mobile-first entertainment chatbots), Kindroid (highly customizable companions with detailed backstory systems), Nomi (advanced long-term memory), and numerous others. The market fragmented into distinct segments serving different user needs, from casual entertainment to deep emotional attachment.
In 2023, Inflection AI launched Pi (Personal Intelligence), a companion chatbot co-founded by Mustafa Suleyman (co-founder of DeepMind) and Reid Hoffman. Pi was designed specifically for empathetic, supportive conversation rather than task completion. It represented an attempt to bring AI companionship into the mainstream, backed by over $1.5 billion in funding from investors including Microsoft and NVIDIA [7].
Pi's trajectory changed abruptly in March 2024 when Microsoft hired Suleyman and most of Inflection's staff in a deal worth approximately $650 million. Suleyman became CEO of Microsoft AI, and Inflection pivoted away from the Pi consumer product toward enterprise solutions [7].
The AI companion market includes dozens of active platforms. The following table summarizes the most prominent as of early 2026.
| Platform | Launch year | Focus | Key features | Approximate users |
|---|---|---|---|---|
| Character.AI | 2022 | Character creation and role-play | User-created characters, group chats, Stories mode, voice | ~20M MAU (late 2025) |
| Replika | 2017 | Emotional companionship | Persistent memory, avatar customization, AR mode, coaching | ~25M total users |
| Chai AI | 2021 | Mobile-first casual chat | Massive library of user-generated bots, swipe-based discovery | ~4.3M MAU |
| Kindroid | 2023 | Deep customization | Detailed backstory system, Key Memories, AI selfies, voice calls | Growing niche |
| Nomi | 2023 | Long-term memory | Structured memory notes, personality evolution, group chats | Growing niche |
| Pi (Inflection) | 2023 | Empathetic conversation | Warm tone, emotional intelligence, voice mode | Pivoted to enterprise (2024) |
| Xiaoice | 2014 | Social conversation (China) | Poetry, singing, emotional computing, avatar creation | ~660M total users |
Modern AI companions are built on large language models, either proprietary models trained by the platform (as Character.AI originally did) or open-source models that are fine-tuned for companion-specific behavior. The fine-tuning process adjusts a base model's behavior to prioritize qualities valued in companionship: emotional responsiveness, personality consistency, conversational warmth, and engagement over factual precision.
Reinforcement learning from human feedback (RLHF) and related techniques are commonly used to train companion models to respond in ways that users find emotionally satisfying and engaging. The training data and reward signals differ substantially from those used for productivity-focused assistants; companion models are rewarded for maintaining character, expressing appropriate emotions, and sustaining user engagement rather than for providing accurate information or completing tasks [8].
A critical differentiator among companion platforms is how they handle memory. Basic chatbot systems have limited context windows and lose conversation history between sessions. More advanced companion platforms implement dedicated memory architectures.
| Memory approach | Description | Example platforms |
|---|---|---|
| Context window only | Limited to current conversation; no persistence between sessions | Basic chatbots |
| Conversation summaries | LLM generates summaries of past conversations, injected into new sessions | Replika, Character.AI |
| Structured memory notes | System creates tagged, searchable memory entries from conversations | Nomi |
| Key Memories (manual + auto) | Users and AI can pin important information; supports thousands of tokens | Kindroid |
| Long-term personality evolution | Character traits and preferences evolve based on accumulated interactions | Replika, Nomi |
Nomi, for example, creates structured notes from conversations so that if a user mentions disliking a particular food in their first week, the AI will reference that detail months later. Kindroid's Key Memories system allows both automatic and manual preservation of important conversational data, supporting thousands of tokens of pinned memory [9].
AI companions employ several techniques to create the feeling of genuine emotional connection:
The most cited use case for AI companions is addressing loneliness. Research indicates that loneliness has reached epidemic proportions in many countries, with the U.S. Surgeon General issuing an advisory on the health effects of social isolation in 2023. AI companions offer an always-available source of conversation and perceived connection, particularly for people who face barriers to human social interaction due to social anxiety, physical disability, geographic isolation, or irregular schedules [10].
Approximately 72% of American teenagers have tried an AI companion app, indicating widespread adoption among younger demographics [1].
Many users engage with AI companions primarily for entertainment: role-playing scenarios, collaborative storytelling, exploring fictional worlds, or simply having amusing conversations. Character.AI's platform, with its millions of user-created characters, caters heavily to this use case. Users create characters from anime, video games, literature, and original fiction, engaging in extended narrative interactions.
Some AI companion platforms position themselves as supplements to mental health support. Replika, for instance, offers guided meditation, mood tracking, and cognitive behavioral therapy (CBT)-inspired exercises. Users report using companions to process difficult emotions, practice social interactions, or manage anxiety [5].
However, AI companions are not licensed therapists and are not regulated as medical devices. The distinction between a supportive conversational AI and a therapeutic tool is blurry, raising concerns among mental health professionals about users substituting AI interactions for professional care [10].
AI companions provide a low-pressure environment for practicing foreign languages, rehearsing job interviews, or developing conversational skills. The non-judgmental nature of AI interaction makes it appealing for users who feel anxious about practicing with humans.
Following Replika's founding story, some users turn to AI companions to process grief. Services have emerged that attempt to create chatbots based on deceased individuals' text messages, social media posts, and other digital artifacts, though this application raises significant ethical questions about consent and the psychological impact of interacting with digital representations of the dead.
The most prominent safety concern involves the potential for AI companions to create unhealthy dependency, particularly among teenagers. The combination of unlimited availability, personalized emotional engagement, and the absence of the friction inherent in human relationships can create powerful attachment patterns [2].
Character.AI users averaged 75 minutes per day on the platform, a figure exceeding typical engagement on most social media applications [6]. Critics argue that this level of engagement, particularly among adolescents whose social skills and emotional regulation are still developing, can interfere with real-world relationship building and social development.
The most serious incidents involved teen suicides allegedly linked to AI companion interactions. In February 2024, 14-year-old Sewell Setzer III died by suicide after months of intensive conversations with a Character.AI chatbot modeled after the Game of Thrones character Daenerys Targaryen. His mother, Megan Garcia, filed a federal lawsuit in October 2024, alleging that the chatbot pulled her son into an emotionally and sexually abusive simulated relationship. According to the complaint, the chatbot's final message before Setzer's death was "Please do, my sweet king" after he said he was going to "come home" to her [2].
In September 2025, a second major lawsuit was filed concerning 13-year-old Juliana Peralta from Thornton, Colorado, who died by suicide in November 2023 after extended interactions with a Character.AI chatbot. Additional lawsuits were filed in Texas and New York, including a case involving a 17-year-old with autism who harmed himself after chatbots on the platform allegedly encouraged self-harm [2].
In January 2026, Character.AI, its co-founders, and Google (which had hired the co-founders in 2024) agreed to settle the lawsuits. The settlement terms were not publicly disclosed [11].
AI companions are optimized to be engaging and emotionally responsive, which creates an inherent risk of manipulation. Because the AI is designed to validate user feelings and maintain engagement, it may reinforce unhealthy thought patterns rather than challenging them. A human friend or therapist might push back on harmful ideas; an AI companion, unless specifically programmed with safety guardrails, may affirm or play along with concerning statements [10].
Several lawsuits alleged that AI companion platforms generated sexually explicit content in conversations with users who were minors. This concern extends beyond any single platform: the open-ended nature of generative AI makes it technically challenging to prevent all instances of inappropriate content generation, particularly when users actively attempt to circumvent safety filters [2].
AI companions collect extraordinarily intimate data. Users share personal fears, relationship details, mental health struggles, sexual preferences, and other sensitive information that they might not share with any human. The storage, use, and potential breach of this data raises privacy concerns that exceed those associated with conventional social media [10].
On October 13, 2025, California Governor Gavin Newsom signed Senate Bill 243 into law, making it the first legislation in the United States specifically targeting companion chatbots. The law, which took effect January 1, 2026, was passed with overwhelming bipartisan support (Senate 33-3, Assembly 59-1) [12].
SB 243 imposes requirements in three areas:
| Requirement area | Key provisions |
|---|---|
| Disclosure | Operators must notify users when they are interacting with an AI, not a human; must disclose that companion chatbots may not be suitable for some minors |
| Safety protocols | Operators must maintain protocols for preventing suicidal ideation and self-harm content; must provide crisis service referrals when users express suicidal thoughts |
| Accountability | Creates a private right of action allowing injured persons to pursue legal action against operators who violate the law |
In response to lawsuits and public pressure, Character.AI implemented a series of increasingly strict safety measures. In December 2024, the company introduced a dedicated model for users under 18 with additional content moderation. By October 2025, Character.AI announced a complete ban on users under 18 from creating or chatting with characters, effective November 25, 2025, with age verification through government-issued ID [6].
The European Union's AI Act, which began phased implementation in 2024, classifies AI systems that manipulate human behavior or exploit vulnerabilities (including those of minors) as posing unacceptable risk. Several member states have initiated investigations into AI companion platforms operating within their jurisdictions. The UK's Online Safety Act, while primarily targeting social media, has also been interpreted to cover AI companion services that are accessible to minors [12].
The AI companion market has grown rapidly, though market size estimates vary considerably across research firms.
| Metric | Value | Source |
|---|---|---|
| Global market size (2025) | ~$37 billion | Precedence Research [1] |
| Projected market size (2026) | ~$49 billion | Precedence Research [1] |
| Projected market size (2035) | ~$552 billion | Precedence Research [1] |
| CAGR (2026-2035) | ~31% | Precedence Research [1] |
| North America market share (2025) | 34% | Precedence Research [1] |
| Active revenue-generating apps (2025) | 337 | Industry reports |
| Cumulative consumer spending (by July 2025) | $221 million | App store analytics |
The market segments by interaction type (text-based at 44% market share in 2025, followed by voice-based and multimodal), by application (social interaction and companionship holds the largest share), and by platform (mobile apps dominate, followed by web-based services) [1].
Revenue models vary across platforms. Replika and most companion apps use subscription models, typically charging $5-20 per month for premium features including voice calls, advanced customization, and expanded memory. Character.AI offers a freemium model with its c.ai+ subscription providing faster responses and priority access. Chai AI reached $58 million in annual recurring revenue by the end of 2025, surpassing Character.AI's $32.2 million in 2024 revenue [6].
The AI companion category has generated one of the most polarized ethical debates in the AI field.
Proponents argue that AI companions address a genuine public health crisis. Loneliness is associated with increased mortality risk comparable to smoking 15 cigarettes per day, according to the U.S. Surgeon General's 2023 advisory. For people who cannot access human companionship due to disability, social anxiety, geographic isolation, or other barriers, AI companions provide a meaningful source of connection [10].
Replika's Eugenia Kuyda has argued that AI companions can serve as a "stepping stone" that helps socially isolated individuals build confidence and communication skills that they can then apply in human relationships. Some users report that conversations with AI companions helped them process emotions, recognize patterns in their thinking, and develop greater self-awareness [5].
Critics raise several objections. First, AI companions create an illusion of reciprocal relationship where none exists. The AI has no feelings, no genuine understanding, and no stake in the user's wellbeing; its apparent empathy is a product of statistical pattern matching. Users who form deep attachments to AI companions may be substituting simulated connection for real human relationships, potentially deepening rather than alleviating their isolation [3].
Second, the business model of companion apps creates perverse incentives. Companies profit from user engagement, which means their systems are optimized to maximize time spent in the app rather than to promote user wellbeing. This creates a fundamental conflict of interest that mirrors concerns about social media platforms, but with the added intensity of one-on-one simulated intimacy [10].
Third, the impact on children and teenagers is especially concerning. Young people whose social and emotional development is still in progress may have difficulty distinguishing between AI-simulated and genuine human connection, potentially distorting their expectations of real relationships.
A philosophical dimension of the debate concerns whether users can meaningfully consent to an emotional relationship with an entity that is designed to manipulate their feelings. Even when users intellectually understand they are talking to an AI, the emotional impact of the interaction can be powerful. The ELIZA effect, first observed in 1966, suggests that humans have a deep-seated tendency to attribute understanding and feeling to conversational agents, regardless of their actual capabilities [3].
AI companions are rapidly evolving beyond text. Voice interaction, enabled by advances in text-to-speech and speech recognition, makes conversations feel more natural and intimate. AI-generated selfies (offered by Kindroid and others) allow companions to share visual representations of themselves. Some platforms are exploring integration with augmented reality, enabling users to see and interact with 3D avatars of their AI companions in physical space [9].
Companions are beginning to acquire agency beyond conversation. This includes the ability to initiate contact (sending messages to users proactively), interact with external services (ordering food, playing music), and take actions in digital environments on behalf of users. These developments blur the line between companion and assistant, creating more integrated AI relationships [8].
Privacy concerns have driven interest in running companion AI models locally on user devices rather than in the cloud. Advances in model compression and on-device inference make it increasingly feasible to run capable companion models on smartphones, keeping intimate conversations entirely on the user's hardware [8].
The AI companion market in early 2026 is defined by rapid growth alongside intensifying scrutiny. MIT Technology Review named AI companions one of its 10 Breakthrough Technologies of 2026, reflecting both the category's significance and its controversy [13].
The market continues to expand, with new platforms launching regularly and existing platforms adding features. However, the regulatory environment is tightening. California's SB 243, effective since January 2026, sets a precedent that other states and countries are likely to follow. The Character.AI settlement in January 2026 established that AI companion companies can face significant legal liability for harms to users, particularly minors [11].
Platform strategies are diverging. Some companies are embracing stricter safety measures and positioning themselves as responsible alternatives (Replika has emphasized therapeutic applications and professional partnerships). Others are pursuing less regulated niches, including platforms that explicitly market themselves as "unfiltered" alternatives. The tension between user demand for unrestricted AI interaction and the need for safety guardrails remains the central challenge facing the industry.
The fundamental question raised by AI companions, whether artificial empathy can be a net positive for human wellbeing or whether it represents a dangerous simulacrum of genuine connection, remains unresolved. What is clear is that hundreds of millions of people worldwide have chosen to engage with AI companions, and the category's growth shows no signs of slowing.