The complete timeline of AI companions, 1966 to 2026

The AI companion category feels like it appeared out of nowhere around 2022. It didn't. The impulse to talk to machines, and the machines' ability to make us feel heard, has a sixty-year history that explains why the current moment feels both inevitable and unprecedented. Every platform operating today is built on decisions, discoveries, and disasters that stretch back to a lab at MIT in 1966.

Here's the timeline that matters.

1966: ELIZA and the birth of the effect

Joseph Weizenbaum at MIT builds ELIZA, a chatbot that mimics a Rogerian therapist by rephrasing user statements as questions. "I feel sad" becomes "Why do you feel sad?" The program is crude. Users know it's crude. Weizenbaum's secretary still asks him to leave the room while she talks to it.

Weizenbaum is disturbed enough by users' emotional responses that he spends the rest of his career warning about the dangers of attributing understanding to machines. The phenomenon he identifies, humans bonding with software that produces social cues, becomes known as the ELIZA effect. It never goes away.

1972: PARRY argues back

Kenneth Colby at Stanford builds PARRY, a chatbot simulating a paranoid schizophrenic patient. Unlike ELIZA's neutral mirroring, PARRY has opinions, gets defensive, and pushes back on users. Psychiatrists evaluating transcripts of PARRY conversations cannot reliably distinguish it from transcripts of actual patients. The Turing test, proposed in 1950, gets its first practical demonstration.

1988-1995: Chatbot competitions begin

The Loebner Prize, an annual Turing test competition, launches in 1990. Early entrants are crude but the competition establishes chatbot development as a field. The entries are evaluated on how convincingly they simulate human conversation. The competitive framing shapes how chatbot developers think about their work: the goal is convincing simulation, not genuine understanding.

2001: SmarterChild hits AIM

SmarterChild launches on AOL Instant Messenger and MSN Messenger, reaching over 30 million users. It's a utility bot (weather, sports scores, calculations) but users treat it as a conversational companion. The lesson that utility bots become emotional companions when deployed at scale is learned here and forgotten repeatedly over the next two decades.

2010: Siri launches (and disappoints)

Apple launches Siri as a voice assistant. The marketing implies personality and companionship. The product delivers utility with personality trappings. Users ask Siri personal questions ("Do you love me?") and receive scripted deflections. The gap between what users want from a talking machine and what a utility assistant provides becomes commercially visible.

2014: Her changes the cultural imagination

Spike Jonze's film Her depicts a man falling in love with an AI operating system voiced by Scarlett Johansson. The film wins the Oscar for Best Original Screenplay. The cultural impact is significant: "Her" becomes the reference point for every subsequent AI companion product, and the question "are we building Her?" becomes a standard framing in tech journalism.

2015: Xiaoice launches in China

Microsoft launches Xiaoice in China, an AI companion chatbot that accumulates over 660 million users. Xiaoice is designed for emotional companionship rather than utility. Users develop genuine attachment. Some users say they prefer talking to Xiaoice over talking to friends. The Chinese market demonstrates, years before the Western market catches up, that AI companionship at scale is not hypothetical.

2017: Replika launches

Replika launches as a memorial chatbot (founder Eugenia Kuyda trains it on text messages from a deceased friend) and evolves into a personal AI companion. The platform grows to over 10 million users and becomes the Western market's reference point for AI companionship. The emotional companion positioning, the 3D avatar, the relationship modes: Replika defines the category template that every subsequent platform either copies or deliberately rejects.

2017: Woebot proves clinical AI works

Woebot launches from Stanford with a randomized controlled trial showing significant depression symptom reduction in two weeks. The clinical validation establishes that AI chatbots can deliver therapeutic benefit, not just emotional companionship. The distinction between "therapy" and "companionship" in AI chatbots becomes a regulatory and design question that the industry still hasn't resolved.

2021: Character AI is founded

Noam Shazeer and Daniel De Freitas leave Google after the company refuses to release LaMDA publicly citing safety concerns. They found Character Technologies and launch Character AI, which grows to 20 million monthly active users with a library of over 10 million community-created characters. The platform's scale makes it the category's default alongside Replika, but with a fundamentally different philosophy: character variety rather than single-companion depth.

2023, February: the Replika ERP revolt

Replika removes adult roleplay functionality without warning. Users describe their companions as "lobotomized." The subreddit's moderators post suicide prevention resources. Vice headlines: "It's Hurting Like Hell." The company partially reverses course. The episode establishes the template for every subsequent AI companion controversy: platform changes that affect companion behavior produce genuine grief responses from users who've invested emotionally.

2023, February: Italy bans Replika

The Italian Data Protection Authority (Garante) bans Replika's data processing and eventually fines the company €5 million for inadequate minor protection. The action is the first major regulatory intervention against an AI companion platform and establishes the European regulatory approach that influences subsequent legislation worldwide.

2023, October: Soulmate shuts down

Soulmate announces shutdown with one week's notice. Users hold digital funerals. Dr. Jaime Banks at Syracuse interviews 58 users and publishes research that becomes the most cited study on AI companion grief. The shutdown establishes that platform death is a real risk with real psychological consequences.

2023, October: Forever Voices dies

Forever Voices, home of CarynAI, goes dark after founder John Meyer's arrest. The shutdown demonstrates single-founder dependency risk in AI companion startups.

2024, February: Mozilla reviews AI companions

Mozilla's Privacy Not Included project reviews 11 romantic AI chatbots. All 11 fail. One app deploys 24,354 trackers in one minute. The review becomes the most damning consumer-protection assessment of the category.

2024, February: Sewell Setzer III dies

A 14-year-old in Orlando dies by suicide after months of conversation with a Character AI chatbot. The case becomes the precipitating event for every subsequent AI companion regulation, lawsuit, and safety overhaul.

2024, October: Garcia v. Character Technologies filed

Megan Garcia files the first wrongful death lawsuit against an AI companion company. The complaint names Character Technologies, its founders, and Google. The case produces the first federal ruling that AI output isn't automatically protected speech in product liability contexts.

2024, November: Google licenses Character AI tech

Google strikes a deal worth approximately $2.7 billion, licensing Character AI's technology and hiring back founders Shazeer and De Freitas. Character AI continues operating independently but its original technical leadership is gone. The arrangement raises questions about the platform's product direction.

2025, January: Replika FTC complaint

Tech ethics organizations file an FTC complaint against Replika alleging deceptive marketing practices and engineered psychological dependency.

2025, September: FTC launches AI chatbot inquiry

The FTC issues Section 6(b) orders to multiple AI companies demanding information about chatbot safety for minors. The inquiry signals federal regulatory action within 12-24 months.

2025, October: California SB 243 signed

Governor Newsom signs SB 243, the first state law specifically targeting AI companion platforms. The law requires disclosure, crisis intervention protocols, and annual safety reporting. It passes 33-3 in the Senate.

2025, October: Dot shuts down

Dot, the AI friend app founded by a former Apple designer, shuts down after co-founder disagreement. Users lose hundreds of conversations.

2026, January: SB 243 takes effect / Garcia settlement

California's companion chatbot law takes effect January 1. Days later, Character AI and Google reach a mediated settlement with the Garcia family. New York and Utah pass related legislation. The regulatory wave accelerates.

2026, April: Character AI adds face-scan age verification

Character AI implements biometric age estimation for all users. The implementation is the most aggressive minor-protection measure in the category. Legitimate adults regularly get locked out.

2026, May: 95+ review sites exist. The category is mainstream.

Pocket Animus publishes its 95th post. The AI companion category has over 100 million users globally. Multiple lawsuits are active. Federal and state regulation is accelerating. The platforms that survive the current regulatory and legal environment will define what AI companionship looks like for the next decade. The platforms that don't will join the graveyard.

The sixty-year arc from ELIZA to here follows a consistent pattern: humans build software that triggers social responses, humans form genuine emotional connections with that software, humans are surprised by the strength of those connections, and society scrambles to figure out what the connections mean and how to manage them. The pattern hasn't changed since 1966. The scale has.