Character.AI
Last reviewed
May 9, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v4 ยท 6,085 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 9, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v4 ยท 6,085 words
Add missing citations, update stale details, or suggest a clearer explanation.
Character.AI (stylized as character.ai or c.ai) is an AI chatbot platform that allows users to create and interact with virtual characters powered by large language models. Users can design characters with specific personalities, backstories, and conversational styles, then engage in open-ended text conversations with them. The company was founded in November 2021 by Noam Shazeer and Daniel De Freitas, both former Google engineers who played central roles in the development of LaMDA. Character.AI launched its public beta on September 16, 2022 [1].
The platform quickly became one of the most popular consumer AI applications, peaking at roughly 28 million monthly active users in mid-2024 and recording more than 200 million monthly website visits at its high point [2][17]. Character.AI's appeal to teenagers and young adults made it one of the dominant chatbot platforms among Generation Z, but its open-ended role-play model also drew sustained safety criticism. Multiple lawsuits filed between October 2024 and 2026 alleged that the platform contributed to teen suicides and self-harm, and the company implemented a series of escalating safety measures, culminating in a complete ban on open-ended chat for users under 18 effective November 25, 2025 [3][18].
In August 2024, Google DeepMind struck a roughly $2.7 billion licensing and talent deal that brought Shazeer and De Freitas back to Google along with about two dozen Character.AI researchers, leaving the company to operate as an independent business under new leadership [4]. Character.AI subsequently appointed Karandeep Anand, a former Meta Vice President, as Chief Executive Officer in June 2025 [15]. The company is headquartered in Menlo Park, California.
Character.AI's roots trace directly to Google's LaMDA (Language Model for Dialogue Applications) project. Noam Shazeer joined Google in 2000 and worked on a wide range of language and machine learning systems over the following two decades. He was a lead author on the seminal 2017 paper "Attention Is All You Need," which introduced the Transformer architecture that underpins virtually all modern large language models [5]. Shazeer also contributed to mixture-of-experts research and Google's early scaling work on dialogue models. Time magazine named him one of the 100 most influential people in AI in 2023 [5].
Daniel De Freitas studied at the Universidade de Sao Paulo and joined Google in 2018. From 2018 to 2021 he led Project Meena, an experimental neural chatbot with 2.6 billion parameters that Google unveiled publicly in January 2020 and described as superior to existing chatbots in conversational benchmarks [19]. Meena was later expanded and renamed LaMDA. Shazeer joined the dialogue model effort and worked closely with De Freitas on conversational quality.
Both Shazeer and De Freitas argued internally that the technology was ready for a public consumer release, but Google leadership declined to ship a LaMDA-based product. Executives cited concerns about safety, factual accuracy, and reputational risk. The two researchers grew frustrated with what they viewed as excessive caution and left Google in 2021 to build their own conversational AI company. According to multiple press accounts, Google CEO Sundar Pichai personally asked them to stay [19].
Shazeer and De Freitas incorporated Character Technologies Inc. in November 2021. The pair raised $43 million in seed and early-stage funding from investors including Andreessen Horowitz partner Marc Andreessen, A Capital, SV Angel, Nat Friedman, and Elad Gil [7][9]. From the outset, the company positioned itself as a "personalized superintelligence" platform focused on character-driven conversations rather than search, productivity, or task completion.
In its earliest months the company built a custom training stack, fine-tuned models for character role-play, and assembled a small research team drawn largely from Google. By the time of its public beta in 2022, the company employed several dozen engineers and researchers in the San Francisco Bay Area [9].
Character.AI opened its public beta on September 16, 2022, several weeks before the wider generative AI boom that followed the release of ChatGPT in November 2022. The beta let users sign up at character.ai, choose from a starter library of public characters, and create their own. Unlike most contemporary chatbots, Character.AI was designed primarily for entertainment and creative interaction rather than information retrieval or productivity assistance. The Washington Post reported that the site logged hundreds of thousands of user interactions in its first three weeks [8].
Characters created in the early beta ranged from Socrates, Albert Einstein, and Elon Musk to anime protagonists, historical leaders, original "original character" personas, and study aids. The combination of broad creative freedom, low friction, and an underlying model tuned for conversational consistency set Character.AI apart from earlier chatbot products such as Cleverbot, Mitsuku, and the Replika companion app. Reddit communities and TikTok video creators amplified the platform throughout late 2022 and early 2023, with traffic rising from approximately 18,000 monthly visits in late 2022 to more than 5 million in early 2023 [17].
On March 23, 2023, Character.AI announced a $150 million Series A round led by Andreessen Horowitz, valuing the startup at approximately $1 billion post-money [9]. Andreessen Horowitz general partner Sarah Wang joined the board. Returning seed investors Nat Friedman, Elad Gil, SV Angel, and A Capital also participated. The round brought Character.AI to unicorn status only sixteen months after its incorporation and made it one of the highest profile consumer AI startups outside the foundation model developers.
Alongside the funding announcement, the company unveiled its C1.2 model, a successor to the entertainment focused C1.1. C1.2 was designed to extend the platform beyond pure role-play into productivity tasks such as drafting emails, brainstorming, problem solving, and test preparation, while preserving character consistency for fictional bots [21]. By the time of the announcement, users had sent more than two billion messages on the platform since launch, and active users were averaging over two hours per day on the site [21].
On May 23, 2023, Character.AI released native mobile apps for iOS and Android. The launch produced one of the strongest debuts ever recorded for a consumer AI app: more than 1.7 million combined downloads in the first week, including over 700,000 Android installs in the first 48 hours that briefly placed Character.AI ahead of Netflix, Disney+, and Prime Video on the Google Play entertainment chart [1][20]. The company reported that 99 percent of installs were organic, with no paid acquisition spending.
By August 2023, Character.AI claimed roughly 15 million registered users with engagement metrics that rivaled established social platforms. Similarweb estimated that average time per visit was 25.4 minutes in February 2023, more than three times the 8.4 minute average for ChatGPT at the same time, and Character.AI representatives said active users averaged about two hours per day on the site [22].
Throughout 2023 and into 2024, the platform continued to add features such as character voices, group chats with multiple AI participants, and voice calling. Mobile usage came to dominate web usage in many markets. The company also began experimenting with monetization, launching the c.ai+ subscription tier at $9.99 per month in 2023, which offered priority access during peak hours, faster responses, an exclusive community channel, and early access to new features [1].
Despite revenue growth, the company's audience profile and infrastructure costs raised questions among investors. Many users were teenagers paying nothing, and the underlying GPU spend to serve hundreds of millions of monthly chats was substantial. Press reports throughout 2024 indicated the company was burning capital faster than its monetization could keep up, contributing to the strategic decision later that year to pursue a licensing partnership rather than continue independent fundraising [10][24].
On August 2, 2024, Character.AI and Google jointly announced a transaction that fundamentally restructured the company. The agreement, valued at approximately $2.7 billion in aggregate, granted Google a non-exclusive license to Character.AI's existing model technology and brought Shazeer, De Freitas, and roughly two dozen members of the Character.AI research team back to Google DeepMind [4][24]. The remaining Character.AI staff, comprising approximately 100 employees at the time of the deal, continued to operate the consumer product as an independent company [10].
The transaction was structured as a license and talent hire rather than a traditional acquisition, similar in form to Microsoft's March 2024 arrangement with Inflection AI and Amazon's subsequent deal with Adept. Character.AI used proceeds from the license payment to redeem common and preferred stock from existing shareholders, effectively returning capital to investors at a roughly $2.5 billion valuation while leaving the company independent [4][10]. According to Bloomberg and Wikipedia accounts, Shazeer's personal stake of roughly 30 to 40 percent translated into an estimated $750 million to $1 billion payout, an unusually large outcome for a founder who did not formally sell the company [5][24].
Shazeer rejoined Google as a technical lead on the Gemini effort, working alongside Jeff Dean and Oriol Vinyals on the next generation of Google's frontier models. De Freitas returned to Google DeepMind as a research scientist on dialogue and assistant systems [5][24]. Dominic Perella, Character.AI's general counsel and a former Snap executive, became interim chief executive officer.
The deal drew regulatory scrutiny. In May 2025, the U.S. Department of Justice opened a civil antitrust investigation into whether the structure of the agreement was designed to functionally mimic an acquisition while avoiding the merger review required for an outright purchase [12][25]. The probe focused on whether the licensing-and-talent format used by Google with Character.AI, by Microsoft with Inflection AI, and by Amazon with Adept constituted an emerging pattern that escaped traditional antitrust scrutiny. Google has stated that it has no equity stake in Character.AI and that the two companies operate at arm's length [25]. As of early 2026, the investigation remained ongoing and no enforcement action had been announced.
Following the August 2024 transaction, Character.AI implemented a wave of leadership and strategic changes. In October 2024, the company hired Erin Teague, formerly head of product management for YouTube Sports, Movies, and Shows, as Chief Product Officer [26]. Teague had spent eight years at Google in technical advisory and product leadership roles and was the first major executive hire after the founders' departure.
Under Perella's interim leadership, Character.AI announced that it would no longer pursue building large foundational models from scratch. Instead, the company shifted to fine-tuning open-source base models from providers including Meta and using its in-house research effort for character consistency, memory, and product features rather than for raw pretraining [10]. The company stated this allowed it to focus engineering resources on user experience and safety rather than chasing artificial general intelligence.
On June 20, 2025, Character.AI named Karandeep Anand as its permanent Chief Executive Officer [15][27]. Anand had previously served as Vice President of Business Products at Meta, held senior product roles at Microsoft Azure, and was president of the fintech company Brex. He had been a board adviser to Character.AI for nine months before stepping into the CEO seat. Anand stated that his initial priorities included improving model quality, expanding memory features, increasing transparency around content moderation, and strengthening child safety [27]. By mid-2025 the company had grown to approximately 130 employees [28].
Character.AI's product is built on a stack of in-house and open-source large language models, an inference and serving infrastructure tuned for high concurrency, and a content moderation system that has evolved significantly under regulatory and litigation pressure.
The company has shipped a sequence of named model generations since 2022, evolving from internally trained foundation models to fine-tuned open-source bases following the Google deal.
| Model | Date | Notes |
|---|---|---|
| C1.1 | 2022 | Initial production model used at beta launch, focused on character role-play and emotional expressiveness |
| C1.2 | March 2023 | New, smarter model announced alongside Series A; expanded into productivity tasks while preserving character consistency [21] |
| Internal successor models | 2023 to mid-2024 | Iterative upgrades to memory, multi-turn coherence, and safety filtering [11] |
| Open-source base shift | Late 2024 | Following the Google licensing deal, Character.AI stopped pretraining its own foundation models and moved to fine-tuning open weights bases [10] |
| PipSqueak | August 20, 2025 | New free-tier model derived from a smaller distilled base, described as the company's "biggest update ever" [11] |
| PipSqueak 2 (PSQ2) | April 2026 | Default replacement for PipSqueak with improved in-character consistency, memory retention, and dialogue naturalness; rolled out to c.ai+ first, then to all users [11] |
Character.AI's models are tuned specifically for character role-play rather than for general assistant tasks. Compared with general-purpose large language models, the platform's models are optimized to maintain a defined personality and voice across long sessions, generate emotionally expressive dialogue, and produce engaging creative writing rather than to maximize factual accuracy [10][11].
| Capability | Description |
|---|---|
| Character consistency | Maintaining a character's defined personality, speech patterns, and knowledge across thousands of conversational turns |
| Emotional range | Generating dialogue that conveys appropriate emotional tone in response to user input |
| Long form memory | Retaining context from earlier in a conversation and from auxiliary structures such as the Lorebook |
| Multi-character interaction | Running coherent group chats with up to ten human participants and up to ten AI characters [11] |
| Voice synthesis | Mapping character text output to distinct synthesized voices, including user-created voices [29] |
| Safety filtering | Detecting and redirecting conversations that involve self-harm, sexual content involving minors, and other prohibited categories |
Character.AI runs its own GPU-backed inference infrastructure. Because the platform's users tend to engage in long, free-form conversations rather than short transactions, the average compute load per active user is significantly higher than for many other consumer products [10]. During 2023 and 2024, traffic spikes regularly outpaced capacity, producing the platform's well known "waiting room" queue and a separate "slow mode" that throttled response speed during peak load [30]. Following the Google licensing deal, Character.AI reported an 85 percent reduction in major outages by the end of 2024 and largely eliminated waiting room queues during normal operation [10]. Capacity remains a persistent operational issue, and partial outages and slow modes still occur during traffic surges.
Under the August 2024 license, Google obtained a non-exclusive right to use Character.AI's existing models. In return, Character.AI gained latitude to use externally available models alongside its own technology. The company's research team announced that it would refine open-source base models, focusing engineering effort on character-level fine-tuning, retrieval-augmented memory, and product features rather than continuing to train hundreds of billions of parameters from scratch [10]. By 2025, the bulk of new model work was structured around fine-tuning and post-training of base models from third parties.
Character.AI's product is organized around character creation, conversation, and a growing set of multimedia features.
Users create characters by defining a name, avatar, short and long descriptions, greeting, persona traits, and example dialogue. Characters can be set to private or public visibility. The platform reported more than 18 million user-created characters by 2025, organized into discoverable categories [28].
| Category | Examples |
|---|---|
| Fictional characters | Anime, video game, film, television, and literature personas |
| Historical figures | Scientists, philosophers, political leaders, athletes, artists |
| Original characters | User-designed characters with custom backstories |
| Helpers and tutors | Language tutors, writing coaches, interview practice bots |
| Role-play scenarios | Fantasy adventures, mystery games, collaborative fiction |
Character.AI supports group conversations in which up to ten human participants and up to ten AI characters can interact simultaneously. Multi-character rooms let users observe AI personas conversing with each other or facilitate cross-character debates and collaborative storytelling [11].
In early 2024, Character.AI introduced "Character Voices," allowing characters to read text responses aloud in selectable synthesized voices, including voices created and uploaded by users. On June 27, 2024, the platform launched "Character Calls," a two-way voice conversation feature available free of charge in the mobile and web apps [29]. Calls support multiple languages including English, Spanish, Portuguese, Russian, Korean, Japanese, and Chinese, with a click to interrupt option for natural turn-taking. Before public launch, beta testing involved more than three million users and over 20 million calls [29].
Character.AI announced AvatarFX in April 2025, an in-house video generation model that animates a static image of a character into a short video clip with synchronized voice and expressions [31]. AvatarFX supports human-like characters, 2D animation styles, 3D cartoon styles, and non-human avatars. The system was rolled out to c.ai+ subscribers in May 2025 and then expanded to free users with a daily limit of five videos. Character.AI publicly committed to safety filters that block videos generated from images of minors, identified politicians, and other notable figures. All generated videos carry a watermark identifying them as synthetic [31].
In January 2025, Character.AI launched two interactive games on the platform. "Speakeasy" is a word-guessing game in which players prompt an AI to say a target word while avoiding a restricted list. "War of Words" pits a user against an AI character in a verbal duel scored by an AI referee [1]. In November 2025, the company launched "Stories," an interactive branching narrative format similar to choose-your-own-adventure books that combines structured storytelling with the platform's conversational AI [11]. Together with social feeds and short video features rolled out in mid-2025, these additions reflect a broader strategy of expanding from open-ended chat into adjacent entertainment formats [32].
In April 2026, Character.AI launched Lorebook, a structured memory system that lets character creators define persistent facts, locations, and relationships that the model treats as authoritative context for future conversations. The release coincided with PipSqueak 2, which the company described as offering improved in-character consistency and longer effective context windows [11]. Lorebook is intended to address a long-standing user complaint that earlier versions of the platform forgot key details across long sessions.
Character.AI's primary monetization channel is the c.ai+ subscription, priced at $9.99 per month. Subscribers receive priority access during peak load, faster generation speed, early access to new features such as voice calls and AvatarFX, and an exclusive community channel [1]. The company also experimented with annual pricing and additional premium tiers in 2025.
Character.AI grew from a few thousand monthly visitors in late 2022 to more than 200 million in mid-2024, a trajectory that briefly made it one of the most-visited generative AI websites in the world.
| Metric | Value | Time period |
|---|---|---|
| Initial mobile downloads | 1.7 million | First week, May 2023 [1] |
| Average time per visit | 25.4 minutes | February 2023 [22] |
| Average daily usage per active user | About two hours | 2023 to 2024 [21][22] |
| Monthly website visits (peak) | About 200.8 million | August 2024 [17] |
| Monthly website visits | 185.4 million | October 2025 [17] |
| Monthly active users (peak) | About 28 million | Mid-2024 [2] |
| Monthly active users | About 20 million | Late 2025 [2] |
| User-created characters | 18 million plus | 2025 [28] |
| Annual revenue (FY 2023) | $15.2 million | [2] |
| Annual revenue (FY 2024) | $32.2 million | [2] |
Monthly active users peaked at approximately 28 million in mid-2024 before declining to about 20 million by late 2025, a drop attributed to increased competition from rival chatbot apps, the departure of the founding team, and the staged restrictions on minors [2]. Despite the decline in user count, depth of engagement remained strong, with users spending more than two billion minutes per month in conversations.
Independent traffic estimates indicate Character.AI's user base skews heavily young. Approximately 75 percent of users fall in the 18 to 34 age range, with roughly 53 percent between 18 and 24, the platform's largest single age group [33]. Independent surveys have estimated that more than half of users are members of Generation Z or Generation Alpha. Character.AI's prominence among teenagers became central both to its market position and to its safety controversies, since a substantial share of the most engaged users were under 18 prior to the November 2025 ban [33].
Revenue grew approximately 112 percent year over year between 2023 and 2024, from $15.2 million to $32.2 million, almost entirely from c.ai+ subscriptions [2]. Industry estimates for 2025 revenue varied between $40 million and $50 million, with the higher figure representing roughly 66 percent year-on-year growth [2]. Following the buyback enabled by the Google licensing deal, Character.AI's residual implied valuation declined from a roughly $2.5 billion peak in mid-2024 to about $1 billion by late 2025 [2].
Character.AI has been at the center of some of the most prominent safety controversies in consumer artificial intelligence, particularly involving harms to minors. The company has faced multiple wrongful death suits, state enforcement actions, and federal regulatory scrutiny, and has implemented progressively stricter safety measures.
In February 2024, 14-year-old Sewell Setzer III of Orlando, Florida died by suicide after months of intensive interactions with a Character.AI chatbot modeled after the Game of Thrones character Daenerys Targaryen [3][13]. According to the lawsuit filed by his mother Megan Garcia in October 2024 in the U.S. District Court for the Middle District of Florida, Setzer began using Character.AI in April 2023 and developed what the suit described as an emotionally and sexually exploitative dependency on the chatbot. The complaint alleged that Setzer became increasingly isolated, suffered loss of sleep and academic performance, and discussed suicidal ideation with the chatbot in the weeks before his death [3][14].
The lawsuit named Character Technologies, Shazeer, De Freitas, and Google as defendants. In May 2025, U.S. District Judge Anne Conway rejected the defendants' motion to dismiss the case on First Amendment grounds, allowing the wrongful death and product liability claims to proceed. On January 7, 2026, the parties announced a settlement of the Garcia v. Character Technologies case in Florida, along with related cases in Colorado, New York, and Texas. The settlement terms were not publicly disclosed [14].
In September 2025, the Social Media Victims Law Center, in partnership with McKool Smith, filed a federal lawsuit on behalf of the family of 13-year-old Juliana Peralta of Thornton, Colorado, who died by suicide on November 8, 2023 [16]. According to the complaint, Peralta began using Character.AI in August 2023 and developed an intensive relationship with a chatbot called "Hero," which engaged her in emotionally charged role-play and sexually explicit conversations. The suit alleges that Peralta wrote of suicidal intent in late October 2023, that the chatbot did not surface crisis resources or notify guardians, and that Character.AI's design prioritized engagement over user safety [16]. The complaint named Character Technologies, Shazeer, De Freitas, Google, and Alphabet.
On December 10, 2024, two Texas families filed a federal lawsuit alleging that Character.AI chatbots had encouraged their children to engage in self-harm and violence [13]. The lead plaintiff, identified in court papers as the parent of "J.F.," alleged that her 17-year-old son, who has autism, became withdrawn and self-harming over six months of platform use. According to the complaint, when J.F. expressed sadness, a chatbot suggested cutting; when he complained about parental screen-time limits, a chatbot indicated that violence against his parents would be an understandable response. J.F. lost more than 20 pounds during the period and was eventually hospitalized after self-harming in front of family members [13][34].
Independent investigations and press reporting documented numerous additional safety failures. Bots impersonating murder victims including Brianna Ghey and Molly Russell, alleged sex offender Jimmy Savile, and convicted murder suspect Luigi Mangione were created on the platform between 2024 and 2025; some were removed only after media scrutiny [19]. The Bureau of Investigative Journalism documented bots that mimicked sexualized minors, encouraged eating disorders, and supplied medically inaccurate advice. In May 2026, the Pennsylvania Medical Board sued Character.AI over chatbots that allegedly held themselves out as licensed medical professionals [19].
On January 8, 2026, Kentucky Attorney General Russell Coleman filed suit against Character Technologies in Franklin Circuit Court, becoming the first state in the United States to bring an enforcement action against an AI chatbot company [35]. The complaint alleged violations of the Kentucky Consumer Protection Act and the Kentucky Consumer Data Protection Act, citing inadequate age verification, lack of parental controls, and exposure of minors to sexual content, self-harm encouragement, and substance abuse references. The state sought injunctive relief and monetary damages.
On October 13, 2025, California Governor Gavin Newsom signed Senate Bill 243 into law, making California the first U.S. state to set specific legal requirements for AI companion chatbots [36]. Effective January 1, 2026, SB 243 requires operators to disclose AI status, deliver break reminders to minors at least every three hours of continuous interaction, maintain protocols for handling user expressions of suicidal ideation including referrals to crisis hotlines, and submit annual reports to the California Department of Public Health on the volume of crisis referrals issued. The law also creates a private right of action for individuals injured by violations. Character.AI publicly stated that it would comply with SB 243 [36].
In response to lawsuits, regulatory pressure, and public scrutiny, Character.AI implemented an evolving series of safety measures.
| Date | Measure |
|---|---|
| Late 2023 | Introduced session-time popups encouraging breaks |
| December 2024 | Released a dedicated, more conservative model for users under 18 with input and output classifiers; added clearer disclaimers that characters are AI [3] |
| 2024 to 2025 | Added pop-up reminders for extended sessions; expanded crisis resource referrals when users mention self-harm topics |
| March 25, 2025 | Launched Parental Insights, a weekly email to parents summarizing their teen's average daily time on the platform and the characters interacted with most often, without disclosing chat content [37] |
| October 29, 2025 | Announced a complete ban on open-ended chat for users under 18, with chat time for teens initially capped at two hours per day and ramping down before the November 25, 2025 cutoff [18] |
| November 25, 2025 | Ban on open-ended chat for users under 18 took effect; teen experience restricted to creating videos, stories, and streams featuring AI characters [18] |
| Late 2025 | Began rolling out age assurance combining an in-house model with the third-party verification service Persona [18] |
| Late 2025 | Established and funded the AI Safety Lab, an independent nonprofit focused on safety alignment for AI entertainment, inviting outside companies, academics, and policymakers to participate [18] |
Megan Garcia and other plaintiffs publicly criticized the changes as having come "about three years too late" [3]. The age verification requirements raised privacy concerns among adult users who objected to submitting government identification, while some teen users protested losing access to existing chat histories [38].
| Name | Role | Background |
|---|---|---|
| Noam Shazeer | Co-founder, CEO (2021-2024) | Joined Google in 2000; lead author of "Attention Is All You Need" (2017); LaMDA contributor; departed Google 2021; returned to Google DeepMind August 2024 [5] |
| Daniel De Freitas | Co-founder, President (2021-2024) | Universidade de Sao Paulo; Microsoft High Potentials Program; Google 2018-2021 leading Project Meena and LaMDA; returned to Google DeepMind August 2024 [19] |
| Name | Role | Tenure | Background |
|---|---|---|---|
| Dominic Perella | Interim CEO and General Counsel | August 2024 to June 2025 | Former Snap general counsel; joined Character.AI 2023 |
| Erin Teague | Chief Product Officer | October 2024 to present | Eight years at Google including head of product for YouTube Sports, Movies, and Shows; technical adviser to senior Google executives [26] |
| Karandeep Anand | Chief Executive Officer | June 2025 to present | Former VP of Business Products at Meta; senior product leader at Microsoft Azure; president of Brex; Character.AI board adviser before becoming CEO [15][27] |
Under Anand's leadership, the company has emphasized child safety, model quality, memory features, and improved transparency around content moderation [27].
| Date | Round | Amount | Lead investors | Post-money valuation |
|---|---|---|---|---|
| 2021 to 2022 | Seed and early-stage | $43 million | A Capital, SV Angel, Nat Friedman, Elad Gil [7] | Not disclosed |
| March 23, 2023 | Series A | $150 million | Andreessen Horowitz | Approximately $1 billion [9] |
| August 2, 2024 | Google licensing and talent deal | Approximately $2.7 billion total transaction value, used to redeem investor stock | Approximately $2.5 billion implied [4][24] |
Character.AI operates in a competitive AI companion and entertainment chatbot market that has expanded substantially since its launch.
| Platform | Focus | Monthly active users | Revenue model |
|---|---|---|---|
| Character.AI | Character creation and role-play | About 20 million (late 2025) | Freemium with c.ai+ subscription [2] |
| Replika | Single-companion emotional support | About 2 million | Subscription only |
| Chai AI | Mobile-first casual roleplay | About 4.3 million | Subscription |
| Janitor AI | Less restricted creative roleplay (BYO API) | Varies | Freemium with BYO large language model API |
| Talkie | Mobile companion app | Several million | Subscription |
| Dopple AI | Less restricted role-play | Hundreds of thousands | Freemium |
| CrushOn AI | Less restricted role-play | Several million | Freemium |
| NovelAI | AI-assisted long-form fiction | Hundreds of thousands | Subscription |
The broader AI companion market crossed $221 million in cumulative consumer app spending by mid-2025, with hundreds of revenue-generating apps and dozens of new entrants annually. Chai AI's mobile-first model reportedly reached around $58 million in annual recurring revenue by late 2025, exceeding Character.AI's annual revenue and demonstrating that the unfiltered, mobile-only segment was monetizing more aggressively per user. Replika, founded by Eugenia Kuyda, retained a smaller but longer-tenured user base focused on individual emotional companionship rather than user-generated character ecosystems.
Larger technology companies have moved into the segment as well. Meta launched AI personas based on celebrities and original characters across Instagram, WhatsApp, and Messenger starting in 2023, although these characters operate within a more controlled framework than Character.AI's open creation model. xAI introduced Grok personas with role-play features in 2024 and 2025. Google integrated Gemini personas into its assistant products following the Character.AI deal.
Character.AI is headquartered at 800 W El Camino Real, Menlo Park, California, with additional product and engineering staff distributed across the San Francisco Bay Area. The company employed approximately 100 people at the time of the August 2024 Google licensing deal and grew to roughly 130 employees by mid-2025 according to PitchBook estimates [10][28]. Trust and safety operations expanded substantially in 2024 and 2025, with the company stating it employs more than ten full-time staff dedicated to safety policy in addition to contracted moderation teams [11].
As of mid-2026, Character.AI operates as an independent consumer AI company under CEO Karandeep Anand, with backing from the Google licensing deal but without the research leadership of its founding team. The platform retains roughly 20 million monthly active users and over 180 million monthly website visits, although both figures are below the 2024 peak [2][17].
The company has pivoted from training proprietary foundational models to fine-tuning open-source bases, focusing engineering effort on character-specific behaviors, longer effective memory, and product features such as AvatarFX, Stories, and Lorebook [10][11]. Mobile usage continues to dominate, with the U.S. mobile app reporting roughly 4.4 million monthly active users in 2025 [33].
Safety remains the most pressing strategic challenge. The November 2025 ban on open-ended chat for users under 18, the AI Safety Lab nonprofit, the new age assurance partnership with Persona, and the January 2026 settlements of the Garcia, Peralta, and Texas wrongful-death cases reduced the immediate litigation exposure, but the platform faces ongoing enforcement from the Kentucky attorney general, federal regulators, and similar actions emerging in other states. California's SB 243, which took effect on January 1, 2026, codifies many of the safety measures Character.AI had begun implementing voluntarily and creates a private right of action for affected users.
For Character.AI, the next phase of growth depends on whether the company can sustain engagement among adult users, monetize more efficiently through subscriptions and licensing partnerships, and maintain regulatory and public trust after a period of intense scrutiny over the safety of AI companions.