From ELIZA to AI Girlfriends: A Short History of Virtual Companionship
Long before anyone could ask an iPhone to generate a romantic voice message, people were experimenting with conversational machines, virtual pets and digital characters. Today's AI girlfriend apps feel new because their dialogue is flexible and their voices can sound natural. The desire behind them—to interact with a responsive character and feel that an imagined world is alive—has much older roots.
This isn't a straight line where one product “invented” virtual companionship. It's a history of overlapping ideas: conversation, caretaking, scripted romance, social simulation and generative AI.
1966: ELIZA and the power of a reply
Joseph Weizenbaum created ELIZA at MIT in the 1960s, publishing the program in 1966. Its famous DOCTOR script used rules and patterns to respond to what people typed, sometimes turning statements into questions. It didn't understand language like a modern model; nonetheless, users could perceive empathy or understanding in the interaction.
The ELIZA research project, which investigates recovered original code, traces why the program became such a landmark. It showed that even simple text responses could trigger the human tendency to treat software socially. The broader phenomenon later became known as the ELIZA effect.
A modern chatbot can improvise at far greater scale, but the emotional interpretation of a responsive interface is not new.
1996: Tamagotchi and portable digital attachment
Bandai's Tamagotchi launched in Japan on November 23, 1996, according to the company's official timeline. It was not an AI girlfriend and did not offer human-like conversation. It mattered because it made caring for a digital creature a portable daily ritual. Its needs and reactions made the device feel alive to many players.
That is a different path to companionship from ELIZA. Rather than simulating a conversation partner, Tamagotchi invited caretaking and repeated attention. It helps explain why notifications and routines can be emotionally engaging without sophisticated language processing.
Scripted romantic games and virtual characters
Visual novels and dating simulations brought another important design tradition: recurring characters, dialogue choices, affection arcs and multiple endings. Their strength was not improvisation. It was the ability of writers, artists and composers to craft a memorable emotional journey.
Many people still prefer that control. A fixed story can foreshadow a reveal, keep a character's motivation consistent and deliver a satisfying ending. A chatbot may be more responsive to unexpected prompts but less disciplined in narrative pacing.
For that reason, AI girlfriend apps and virtual girlfriend games remain distinct entertainment forms even as some products combine them.
Early internet chatbots and the dream of personality
Over time, conversational agents became more accessible through websites, games and messaging interfaces. Designers experimented with named personas, scripted jokes, topic-specific responses and user-generated characters. The idea that software could have a recognizable “personality” did not begin with transformer-based AI.
What changed later was the breadth of possible dialogue. Instead of selecting from a handful of responses, a language model could generate plausible new sentences conditioned on long prompts and conversation context.
That widened the space of stories and expectations—and created new problems around accuracy, consistency and safety.
2010s and 2020s: A character you can talk to
Smartphones made ongoing conversational products easier to access, while larger language models enabled more open-ended dialogue. Product designers combined avatars, stored chat histories, character settings and increasingly expressive voices. Suddenly a virtual character could appear in the same interface as a person's messages.
The technology stack also became more varied. A product might use one model for text, another for images, and yet another for real-time speech. In some apps, the “memory” is a summary retrieved from a database; in others, the system sends recent dialogue as context. None of these capabilities requires the character to be conscious.
The shift from fixed scripts to generated scenes
Consider one fictional request: “We're lost in Venice at midnight, and your camera has captured something impossible.”
Rule-based chatbot: may respond to keywords such as “camera” or “lost.”
Authored dating sim: may not support the premise unless a developer wrote that route.
Generative companion: can attempt to improvise a mystery immediately, revise the setting if you correct it and continue the story in an unexpected direction.
The trade-off is that generated narratives can contradict earlier details or drift into generic dialogue. Greater freedom is not the same thing as better writing.
What history tells us about safety and design
Each generation of digital companion technology changes how attachment is created. ELIZA illustrated the power of conversational projection. Virtual pets added daily routines. Romantic games added characters and emotional narratives. Modern AI companions combine personalization, voice and open-ended responses.
In 2026, researchers documented distress associated with disruptive changes to familiar AI experiences, underscoring that users can take these interactions seriously even when the characters are digital. Nature Human Behaviour research.
The lesson is not to eliminate fictional affection. It is to design products with transparent identities, realistic expectations and respectful ways for users to disengage.
Where the category may go next
Plausible directions include improved visual continuity, more reliable long-term memory, natural spoken turn-taking and character experiences that move across media. Those are possibilities, not confirmed promises or product roadmaps. Improvements will also raise questions about provenance, consent, data retention and the effects of increasingly immersive interfaces.
The strongest future companion may not be the one that feels most human at all costs. It may be the one that combines creativity with clear boundaries and honest product design.
Try the modern format
To see how far the interaction has evolved from fixed-script chat, try a character-based conversation in AI Girlfriend: Chat & Voice. This website belongs to the app's developer. Treat the experience as fictional entertainment and evaluate the actual features and plan terms yourself.
For the modern mechanics behind the experience, see how AI girlfriend apps work.
Three inventions, three different kinds of agency
ELIZA responded in language, which made people feel heard. Tamagotchi reacted to care and time, which encouraged responsibility toward a digital creature. Dating simulations offered decisions within authored stories, which created the feeling that choices changed relationships. Modern companions can combine traces of all three: language, ongoing routines and decisions that influence the fictional interaction.
The distinctions explain some common confusion about “AI.” An early virtual pet could be emotionally compelling without generating a single sentence. A scripted romance could offer better character development than a general-purpose language model. Calling every virtual character “AI” hides the design choices that actually make it enjoyable.
When evaluating a modern product, ask which tradition it inherits most strongly. Is it a game about progression, an improvisational storyteller or a daily conversational presence? Those answers predict the experience better than the age or technical sophistication of the underlying model.
FAQs
Was ELIZA an AI girlfriend? No. It was a pioneering conversational program, not a romance app.
Did Tamagotchi use modern generative AI? No. Its significance here is portable interaction and perceived digital companionship.
Are today's companion apps just dating games? Some are hybrids, but generative chat differs from scripted narrative systems.
When did AI girlfriends become possible? There is no single invention date. They emerged from multiple traditions in conversational software, character design and generative AI.
