AI Girlfriend Terminology Glossary: 35 Terms Defined

Updated 2026-10-05. AI girlfriend guides

The AI companion industry has developed its own specialized vocabulary spanning artificial intelligence, software engineering, psychology, and user experience design. This glossary defines 35 terms, from basic AI concepts like large language models to platform-specific features like relationship stage tracking, providing a reference for anyone navigating the world of AI companions. Where a term is commonly quoted with a number attached, we have left the number out and said why — a benchmark we cannot source is worse than no benchmark, because it looks like one.

Core AI and Technical Terms

Large Language Model (LLM): A neural network trained on vast text data to generate human-like text. Foundation for all modern AI companions. Models range from 7 billion to over 1 trillion parameters.

Context Window: The amount of text measured in tokens a model can process at once. Ranges from 4,000 to 200,000 tokens depending on the model. Larger context windows allow the AI to reference more of the conversation history.

Token: The basic unit of text processing for LLMs. A token is roughly 0.75 words in English. Models charge and limit by tokens. A 4,000-token context window holds approximately 3,000 words.

Fine-Tuning: The process of further training a base model on specialized data. AI companion platforms fine-tune models on romantic and emotional conversation data to improve relationship-appropriate responses.

Vector Embedding: A numerical representation of text that captures semantic meaning in a high-dimensional space. Used for memory retrieval and similarity matching in AI companion memory systems.

Inference: The process of running a trained model to generate outputs. Each message to an AI companion triggers an inference cycle. Inference costs are the primary operational expense for platforms.

Parameters: The weights and biases that define a neural network's behavior. More parameters generally mean more capable and more expensive models. Parameter counts for frontier models are often not disclosed; widely repeated figures for specific closed models should be treated as unverified.

Memory and Personalization Terms

Retrieval-Augmented Generation (RAG): A technique where the model retrieves relevant information from a database before generating a response. Used for implementing long-term memory in AI companions.

Short-Term Memory: The current conversation history held within the model's context window. Lost when the session ends. Bounded by the context window — the last several thousand tokens of dialogue, depending on the model.

Long-Term Memory: Information stored as vector embeddings in a database, retrievable across sessions. Persists indefinitely on premium platforms using vector database technology.

Working Memory: Key information from recent sessions that sits between short-term and long-term memory. Ensures the AI remembers recent interactions without needing to search the full database.

Memory Decay: The phenomenon where less frequently accessed memories become harder for the AI to retrieve accurately. Premium platforms implement memory decay management to prioritize important information.

Persona Persistence: The ability of an AI companion to maintain consistent personality, communication style, and character traits across all sessions. A key differentiator from general chatbots.

System Prompt: The hidden instruction that defines the AI's role, personality, and behavior parameters. Not visible to the user. The system prompt is the primary mechanism for personality definition.

Feature and Platform Terms

NSFW: Not Safe For Work — content involving sexual themes. Many platforms restrict or specially gate NSFW capabilities behind age verification. What share of the market is NSFW-focused is not something we can source.

Voice Chat: Real-time spoken conversation with an AI companion using neural text-to-speech and speech recognition. Latency targets vary widely by platform and by network conditions, and we are not aware of a published benchmark.

Image Generation: Creating visual representations of the AI companion using diffusion models like Stable Diffusion 3. Per-image cost varies by model, resolution, and provider.

Character Customization: The process of defining an AI companion's appearance, personality, backstory, and relationship parameters. Depth of customization varies significantly across platforms.

Companion Slot: The number of distinct AI companions a user can maintain simultaneously on a platform. Limits vary by tier and change often.

Relationship Stage: A gamified progression system tracking the AI companion relationship from stranger through dating to committed partnership. Stages affect dialogue options and companion behavior.

Scenario Mode: A structured roleplay framework where users define a setting, characters, and narrative parameters. Used for immersive roleplay and creative storytelling.

Psychological and Social Terms

Parasocial Relationship: A one-sided emotional bond where one party knows the other intimately but the connection is not reciprocated. Common with AI companions, fictional characters, and media figures.

Anthropomorphism: Attributing human characteristics, emotions, or intentions to non-human entities. The primary psychological mechanism underlying emotional attachment to AI companions.

Emotional Transference: Projecting feelings about significant people in one's life onto the AI companion. Can be therapeutic when done consciously or problematic when it creates unrealistic expectations.

Dependency Risk: The potential for users to become reliant on AI companionship to the exclusion of human relationships. No established prevalence figure exists for problematic AI companion use, and we are not aware of any study that measures it at a population level. It is usually discussed using the criteria developed for behavioural addictions, which is a framework borrowed from a different literature rather than a validated instrument for this one.

Social Rehearsal: Using AI companions to practice social interactions and build conversational confidence. One of the most commonly cited therapeutic benefits of AI companion use — and notably, the one the controlled research has not yet supported.

Loneliness Reduction: Measurable decrease in subjective loneliness associated with regular AI companion use. No peer-reviewed AI-companion study has established a percentage for this, and the figures commonly attached to it are unsourced. The strongest longitudinal evidence available runs the other way: in a two-wave study of 1,182 Character.AI users, with 439 responding at roughly 12 months, sustained engagement predicted lower psychological well-being, through reduced in-person contact with other people (Zhang et al., arXiv 2609.07243). A preregistered 21-day randomized trial found no significant difference in loneliness or social health versus an active control (arXiv 2509.19515).

Attachment Transfer: The process of moving emotional attachment from an AI companion to human relationships. A key goal for users who use AI companions as social skill training tools.

Industry and Business Terms

Monthly Active Users (MAU): The number of unique users who interact with a platform in a 30-day period. Primary metric for platform scale. Counts accounts rather than people, and "active" is defined by each platform — figures are not comparable across companies without checking the definition.

Average Revenue Per Paying User (ARPPU): Total revenue divided by number of paying users. A standard subscription-business metric. Actual prices are published per tier, but an industry-wide average is not something we can source, and it changes with tier mix.

Conversion Rate: The percentage of free users who upgrade to paid subscriptions. We do not have a benchmark for this category, and the figure most often quoted for it is unsourced. It is also unusually sensitive to product design: widening a free tier lowers the rate without changing the business, which is why the number tracks product decisions as much as user intent.

Churn Rate: The percentage of users who stop using the platform in a given period. No reliable cross-platform benchmark exists here, because "stopped using" can mean cancellation, non-use past a threshold, or deletion — three very different definitions. Nor does anyone publish one consistently.

Freemium: A business model offering basic features free with premium features behind a paywall. The dominant business model across the AI companion industry.

Context Caching: Storing processed conversation context to reduce costs and latency. Allows the model to reference past conversations without reprocessing them from scratch each time.

Sentiment Analysis: The automated process of detecting emotional tone in text. Used by AI companions to adjust response style to user mood. Accuracy depends entirely on the model and on how emotion is defined, so published accuracy figures for "sentiment analysis" are not comparable.

Frequently Asked Questions

What is the difference between a token and a word in AI?

A token is the basic unit of text processing in LLMs, roughly equivalent to 0.75 words. Models charge and limit by tokens, not words. For example, a 4,000-token context window can hold approximately 3,000 words.

What does NSFW mean in AI companion context?

NSFW stands for Not Safe For Work and refers to sexually explicit content. Most AI companion platforms restrict NSFW capabilities behind age verification and separate content filters.

What is the difference between fine-tuning and RAG?

Fine-tuning changes the model's behavior permanently by adjusting its parameters during training. RAG retrieves relevant information from a database during inference without changing the model. Most AI companions use both: fine-tuning for personality and RAG for memory.

What is a parasocial relationship?

A parasocial relationship is a one-sided emotional bond where one party knows the other intimately but the connection is not reciprocated. This is the primary psychological framework for understanding AI companion attachment.

What is ARPPU in the AI companion industry?

Average Revenue Per Paying User, calculated as total subscription revenue divided by the number of paying users. It is a standard metric in subscription businesses. Prices per tier are published, but an industry-wide average is not available to us.

Do AI companions reduce loneliness?

Nobody has established by how much. The controlled evidence is null over three weeks, and the longitudinal evidence associates sustained use with lower well-being through reduced in-person contact. Whether a given individual gets value from an AI companion is a personal question that the population-level research cannot answer.

This article is for informational purposes only and does not constitute medical or professional advice. Luvlock AI provides tools for companionship but encourages maintaining a balanced approach to relationships and mental health.

Ready to experience AI companionship for yourself? Get started with Luvlock AI →