Founding-member pricing: $2,999 once, then $299 a month once your site is live. Claim your spot

Beginner guide

How do AI companion apps work? The parts, in plain English

AI companion apps work by sending each message, plus a written character persona and saved memories, to a large language model that writes the reply in that character's style. Around that model sit image generation for photos, safety filters and age checks, and billing for memberships, tokens and unlocks. The owner's main job is bringing in members.

  • By the WLF University team
  • Updated
  • 9 min read

What happens when you send an AI companion a message?

From the member's side, an AI companion app feels simple: you pick a character, type a message, and she answers like someone who knows you. Behind that screen, a few separate systems work together on every message. If you are thinking of starting an AI companion business, it helps to know what each one does, because each is either something you build and pay for yourself or something a provider runs for you.

Here are the seven parts, in the order a message passes through them:

PartWhat it doesWhat the member notices
1. Chat interfaceThe website or app where members log in, pick a companion and typeThe look and feel of your brand
2. Character personaWritten instructions that define who the companion isHer personality and the way she talks
3. MemoryStores and recalls details from past chatsShe remembers their name and their day
4. Language modelThe AI that writes each replyNatural, fast conversation
5. Image generationCreates photos of the companionPhotos and unlockable images
6. Safety and age checksFilters content, verifies age, shows required noticesAn age gate and clear rules
7. BillingCharges for memberships, tokens, tips and unlocksA checkout that works

The sections below take each part in turn, then show which parts you would handle yourself and which a done-for-you build handles.

What does the AI language model actually do?

The language model is the engine that writes replies. A large language model, or LLM, is trained on a huge amount of text to do one job very well: predict what comes next. As Georgetown's Center for Security and Emerging Technology explains in its guide to next word prediction, a model like this generates a reply by predicting the next word again and again, and that simple process turns out to be surprisingly powerful. Ask it a question and the most likely continuation is an answer; give it a character to play and the most likely continuation is that character talking.

The model itself has no personality and no memory of your members. On its own it is a general writer. Everything that makes it feel like a specific companion comes from what the app sends along with each message, which is why the next two parts matter so much.

How does an AI companion get her personality?

Every companion has a persona: a block of written instructions the app sends to the model with each message, usually called a system prompt or character card. It describes who she is, how she talks and what she will and will not do. Anthropic's own prompting documentation notes that setting a role in the system prompt focuses the model's behavior and tone, and that even a single sentence makes a difference. Companion apps take that idea much further, with a full profile for each character.

A typical persona covers:

  • Identity. Name, age (always an adult), job, hometown and a short backstory.
  • Personality. Core traits, such as playful, shy, confident or sarcastic.
  • Speech style. Short or long messages, slang, emoji, how she greets someone.
  • Boundaries. Topics she avoids and rules she never breaks.
  • Niche fit. Interests that match the audience the site serves, such as gaming, country life or fitness.

This is why niche sites work. A site built for one audience can give every companion a persona that fits it, so a visitor who arrives from a gaming community meets characters who actually talk about games. Our guide to the best AI companion niches to start with covers how to pick that audience.

How do AI companions remember you?

Memory comes in two layers. The first is the model's context window: the amount of text it can consider at one time. IBM describes the context window as the equivalent of a model's working memory, which decides how long a conversation it can carry on without forgetting details from earlier in the exchange. Recent messages are sent along with each new one, so the companion can follow the thread of tonight's chat.

The second layer is long-term memory, for things a member said last week. A common approach is to save short summaries of past conversations in a searchable database, then pull the relevant ones back in when a new message arrives. Pinecone, a vector database company, walks through this pattern in its write-up on memory for open source LLMs: each conversation is summarized, stored as a new memory, and retrieved and added to the prompt when it becomes relevant again.

For a business owner, memory is not a technical detail. It is a large part of why members come back. A companion who remembers a member's name, job and bad week feels personal, and a personal experience is what members pay to continue.

How do AI companion apps create photos?

Companion photos come from image generation models, most of them built on a method called diffusion. IBM's explainer on diffusion models describes how they learn by adding random noise to images until they look like TV static, then learning to reverse that process, so they can turn noise into a new, detailed picture from a text description.

The hard part for a companion site is consistency. Members expect the same face in every photo, in different outfits and settings, so companion images are made from a fixed reference look for each character. Two rules apply on any responsible site: every companion is an adult, and no companion uses the likeness of a real person. Photos are where several revenue streams come from, including locked images members pay to open and custom photo requests.

Safety is its own layer, and it runs on every message and image. It usually includes:

  • Content moderation. Automated filters check what goes in and out. OpenAI, for example, offers a moderation endpoint that classifies harmful content in text and images and is free to use.
  • Age verification. Adult companion sites must keep minors out, both because the law requires it and because payment processors do.
  • AI disclosure. Several US states now require companion apps to tell users they are talking to AI. Washington's law, which takes effect on January 1, 2027, requires a disclosure at the start of a chat and at least every three hours, according to a Mayer Brown legal update.
  • Crisis protocols. Some of those laws also require a plan for referring users who mention self-harm to crisis resources.

Our guide to AI companion laws covers these rules state by state. It is general information, not legal advice, and the same applies here.

Skip the build. WLF University builds your branded AI companion site and runs the AI chat, companions, billing and compliance for you, then teaches you to bring in members. $2,999 once, then $299 a month once your site is live, and you keep up to 60% of total site revenue. See what is included.

How do AI companion apps make money?

The last part is billing, and it is where the app turns attention into revenue. Most companion sites stack several streams rather than relying on one:

  • Memberships for recurring access to chat and members-only content.
  • Tokens that members buy and spend on messages, photos and extras.
  • Tips to a favorite companion.
  • Photo unlocks for locked images.
  • Custom requests for a specific photo.

Billing is also one of the hardest parts to set up yourself. Adult-oriented companion sites are treated as high risk, and card networks set strict conditions. When Mastercard tightened its adult content rules, it required the banks behind adult merchants to confirm controls such as age and identity verification and content review. The AI girlfriend business model guide explains how the streams combine into revenue per member.

Which parts do you build, and which can be run for you?

If you build an AI companion app yourself, every one of the seven parts is your job: choosing and paying for a language model, writing personas, building memory, generating consistent photos, moderation, age checks, legal notices and a high-risk payment setup. With a done-for-you AI companion business, the provider runs the technology and you own the brand. Here is how that splits with WLF University:

PartBuild it yourselfWLF University
Branded site and chat interfaceYou hire a developerBuilt for you, in your niche and look
Companions, personas and photosYou create and manage themLoaded at launch, with hundreds of new companions added every week
AI chat and memoryYou pay for and maintain the AIBuilt in, with no separate AI usage bill
Safety, age checks and complianceYour jobHandled
Billing, payments and chargebacksYou apply for a high-risk processorHandled
Bringing in membersYour jobStill your job, taught in the course

The course's second module, Your Site, Decoded, is a tour of exactly these parts on your own dashboard: your companions, the AI chat and every revenue stream. The AI girlfriend business course guide lists all five modules. For the cost of building each part yourself, see the AI companion business cost breakdown.

What does this mean if you want to start one?

Knowing how AI companion apps work makes one thing clear: the technology is a stack of moving parts that needs constant upkeep, and none of it brings in a single member. Members arrive only when someone sends them, through social media, communities, search or paid ads. That is true whether you build the app or have it built for you, and it is why results depend on the traffic and effort you put in. No setup guarantees income.

So the real choice for a beginner is where to spend your time. You can spend months assembling the seven parts, or start with them running and put your time into marketing from the first week. Our pillar guide on how to start an AI companion business walks through the full path. If you already run traffic and want the operator view of the same platform, WhiteLabelFans explains its white label AI companion platform for affiliates and media buyers.

Frequently asked questions

How do AI companions work?

An AI companion sends each message you write to a large language model, together with a written persona for the character and any saved memories about you. The model writes a reply in that character's style. Separate systems handle photos, safety filters, age checks and payments.

What are AI companion apps?

AI companion apps are websites or apps where people chat with AI characters that have their own personality, look and memory of past conversations. Members usually pay through memberships, tokens, tips and photo unlocks. Many are built around a specific audience or interest.

Are AI companions real people?

No. AI companions are AI characters, not real people, and several US states now require companion apps to tell users so. On a responsible site every companion is an adult character and none uses a real person's likeness.

Is an AI companion the same as a chatbot?

An AI companion is a type of chatbot built for an ongoing relationship rather than answering one-off questions. It adds a fixed personality, memory of past chats and often photos, so the conversation feels personal and continues over time.

Do I need to know how to code to start an AI companion app?

Not if you use a done-for-you provider. Building one yourself means hiring developers for the model, memory, images, safety and billing. With WLF University the site and technology are built and run for you, and your job is bringing in members.

Skip the build. Start the business.

We build and run your AI companion site, live within 48 hours of onboarding. You learn to drive traffic and keep up to 60%.