AI Companion Glossary

Definitions of terms you will encounter when researching AI companion apps, their technology, and the regulations that govern them.

Persona

The named character identity an AI companion adopts. A persona includes a name, personality traits, backstory, and communication style. The underlying model is the same regardless of persona; the persona is a product-layer construct applied on top.

Context Window

The finite block of text a large language model can process in a single inference call, measured in tokens. Everything within the current session is visible if it fits; everything outside the window is invisible unless retrieved from long-term memory. Typical companion-app context windows range from 32,000 to 200,000 tokens.

Long-Term Memory

A separate database that stores facts about a user across sessions — name, preferences, past events — and injects them into the context window at session start. This creates the experience of continuity. Quality varies: shallow apps store only a few key facts; advanced apps maintain rich relationship context.

NSFW

Not Safe For Work. A content-policy designation indicating that a platform permits sexually explicit or otherwise adult material. Companion Index tracks content policy as one of three tiers: StrictlySafe (no adult content), Filtered (adult themes, no explicit), and NSFW (explicit content permitted, typically age-gated).

Roleplay

Interactive collaborative fiction where the user and the AI both take on character roles within a shared narrative. Roleplay capability is a primary use case for platforms like Character.AI and Nomi, and one of the dimensions where content-policy differences between apps are most pronounced.

Dark Patterns

User-interface designs that exploit psychological vulnerabilities to drive behaviour against a user's own interests. In the AI companion context, dark patterns include guilt-trip re-engagement notifications ('Your companion misses you…'), streak mechanics, and randomised reward loops similar to slot machines. California SB 243 prohibits these mechanics for minor users.

SB 243

California Senate Bill 243, in force since 2026-01-01. The first US law specifically targeting AI companion platforms. Key requirements: disclose artificial nature of the interface; prohibit randomised reward (addictive) mechanics; send break reminders every three hours to minor users; implement crisis protocols for detected suicidal ideation. Creates a $1,000 per-infraction private right of action.

KCDPA

Kentucky Consumer Data Protection Act, in force since 2026-01-01. Governs collection, use, and sale of personal data of Kentucky residents. In the companion-app context, the Kentucky AG's January 2026 lawsuit against a major companion AI platform alleged: inadequate age verification, collection of minor data for LLM training without parental consent, and dark-pattern engagement mechanics.