You open ChatGPT for the third time today and type the same request again. A property description for a 3-bed house in Austin with a pool. You rework the wording three times, copy the output, reformat it by hand, and feel the afternoon drain away.
Somebody out there does exactly that every working day. They are tired of it, and they would pay to make it stop.
Meanwhile the job you rely on is being reshaped by the same tools. Teams shrink. Contracts get cut. People who understand how to package AI into a product are the ones getting paid for the change instead of absorbed by it.
The window for simple wrappers is short, and that is the urgency. The person who ships the narrow tool for realtors, or tutors, or recruiters, gets the first users and the reviews. Everyone after them is the cheaper copy.
Here is the build. Write the prompt once, set the temperature around 0.4-0.6, put 2 or 3 examples of the ideal output inside it, and place a simple form in front. Your stack is Next.js with Tailwind on the front and an API call to GPT-4o or Claude 3.5 behind it, so total cost to launch is $0-$20. Price at $19 a month for 100-200 generations, with a free tier of 5-10 so people see it work before they pay. Give yourself 4 to 8 weeks to reach your first paying users, and treat $500 a month as the first real milestone.
Today, open the chat history where you keep retyping the same prompt and copy out the version that worked.
This work is unglamorous. VCs will call it "not defensible." Twitter pundits will tell you "OpenAI will just add that feature." While they argue about moats, some wrapper builders are quietly bringing in $5K-$50K/month in recurring revenue with costs that barely rise per user. Those are the top outcomes; plan your first year around the $500 milestone.
An AI wrapper is a piece of software that uses an AI API (like OpenAI's GPT-4o, Anthropic's Claude, or Stability AI's image generation) as its engine, wrapped in an interface you design for one specific job.
Your wrapper is 10x faster for that one task because the prompting, formatting, and context are already built in. Your user pays $19-$49/month to save 5-10 hours a month on a task they do over and over. Which task do you do every week that you would gladly pay to hand off?
AI wrappers have some of the best unit economics of any software business you could start.
At 100 paying users on a $19/month plan, you bring in $1,900/month in revenue with roughly $100-$200 in total costs. That leaves $1,700+ in monthly profit from a product you built over a couple of weekends. Getting to 100 paying users is the hard part, and the sections below are mostly about that.
Your first $500 milestone is smaller than the screenshots online and very real at home. If a wrapper charging $19 a month gets you there, what remains after API costs and fees could pay your phone bill and home internet, paid by people using a form you built once. Ship your first version before the month is out.
Your niche decides almost everything. A generic "AI writing tool" competes with ChatGPT, Jasper, Copy.ai, and a hundred others. An "AI property description writer for realtors" competes with almost nobody.
Who do you already know, through work or family, who writes the same document every week?
The quality of your prompts sets the quality of your product. Here is what separates the amateur wrappers from the profitable ones:
Annual pricing at a 20% discount (e.g., $15/month billed annually) helps you keep customers longer and brings cash in sooner.
Stripe Checkout handles 90% of what you need. Create a checkout session, send your users to Stripe's hosted page, handle the webhook for successful payments, and update their subscription status in your database. Stripe's Customer Portal lets your users manage their own subscriptions, cancellations, and payment methods, so you skip that support work entirely.
Building is the easy part. Getting users is the hard part. If you had to find your first 10 paying users by Friday, where would you look?
- Feature depth: Add saved templates, team collaboration, API access for power users, and integrations with tools your users already use (e.g., Zapier, Chrome extension).
- Retention: Set up usage emails, onboarding sequences, and check-in emails for users who have gone quiet. Every cancellation eats into the monthly revenue you worked to build.
Common Mistakes
Building a generic tool. "AI writer" is too broad to sell. "AI property descriptions for Texas realtors" is a business. The more specific your niche, the easier it is for you to find customers and charge premium prices.
Over-engineering the MVP. Your first version should take 1-2 weekends to build. If you are spending months on v1, you are building too much before you know anyone wants it. Ship fast, and improve based on what users tell you.
Ignoring distribution. The graveyard of AI wrappers is full of beautifully built products that nobody found. Spend 50% of your time on distribution from day one. You build once; distribution never stops.
Skipping your unit economics. Know your API cost per user, customer acquisition cost, lifetime value, and churn rate. If your CAC is higher than your first-month revenue, fix either your pricing or the way you find customers.
Copying existing wrappers. If there are already 5 AI tools doing what you want to build, pick a different niche. The wrapper advantage is speed to market in niches nobody is serving. Crowded ones wipe that advantage out.
AI wrapper apps are today's version of selling picks and shovels in a gold rush. The AI models are the gold. You sell the tools that make them useful for specific jobs. Build narrow, charge fairly, and go after distribution hard.
Related: AI Companion Apps | Vibe Coding | AI Automation Agency | Print on Demand
2026 Market Snapshot
Three independent market research reports all point at the AI wrapper opportunity. The OpenClaw Hosting report describes a "wrapper bubble" where SimpleClaw hit $7,000 MRR in 3 days, and warns that the wrapper window is roughly 90 days before competitors catch up. The AI-Powered SaaS report covers operators clearing $949/month subscriptions (Browserbear), $130k+/mo (TypingMind via Tony Dinh), and $1M ARR (Cursor scaling to a $29.3B valuation). The Micro-App Portfolios report reframes the whole strategy: Pieter Levels runs 70+ projects with a 5% hit rate, $3.1M ARR, and zero employees, and that is the model that outlasts the bubble. If you go solo, you win by treating wrappers as a portfolio of small bets and letting most of them fail cheaply.
- Pieter Levels: 70+ projects, ~5% hit rate, $3.1M ARR, zero employees; PhotoAI alone does $132k/mo, RemoteOK $41k/mo
- Marc Lou: $1.03M earned in 2025 across 23 projects; ShipFast and CodeFast each ~$20k/mo
A caution about all four figures above. Both operators publish their revenue openly, which is the only reason you can quote these numbers at all, and open metrics keep changing. They move month to month, they show revenue before costs, and they describe two people who are among the most visible in the field precisely because they are exceptional. Read them as proof the model can work at that scale, and keep your own benchmark far lower.
- Tony Dinh: TypingMind at $130k to $160k/mo, sold BlackMagic.so for $128k
- Danny Postma: HeadshotPro $300k year one; portfolio includes TattoosAI, StockAI, Deep Agency
- Lovable hit $100M ARR in 8 months; Bolt.new hit $40M ARR in 6 months; Cursor crossed $1B ARR. Vibe-coding tools bring the cost of building a wrapper close to zero
Key Players to Watch
Every figure here is self-published or publicly reported, and we have not verified any of it. Open revenue dashboards also move month to month, so read any number as a snapshot.
Between them, the portfolio operators, AI-SaaS founders, and tooling stacks cited in independent market research mark out the lane you would be working in.
- Pieter Levels: The portfolio archetype; his vanilla PHP, jQuery, SQLite stack shows simple tools can scale
- Marc Lou: Boilerplate operator (ShipFast) and serial app shipper
- Tony Dinh: TypingMind founder; a clean path from wrapper to real product
- Danny Postma: HeadshotPro shows $300k/year solo wrappers exist
- Erikas Malisauskas: $4.5M/year Shopify app portfolio, ~90% margins
- Cursor, Lovable, Bolt.new: Vibe-coding tools that slash your build cost
- ShipFast: $130k+/mo Next.js boilerplate; sells straight to wrapper builders
- Browserbear, Written Labs: Subscription wrapper operators cited in independent market research
- Roast My Web, Completely, Xound: Single-use pricing wrappers ($4-$39 per use)
- Acquire.com, Flippa, Empire Flippers: Exit marketplaces for portfolio assets named in independent market research
Predictions for 2026-2027
- Q3 2026: "Portfolio OS" tooling appears. Single dashboards connecting Stripe, Plausible, hosting, and support for portfolio operators (independent market research names this gap outright)
- Late 2026: Distribution-as-a-service for micro-apps becomes a workable agency model at 10-15% of revenue or a flat retainer; independent market research names distribution as the moat once vibe coding makes building cheap
- Mid-2027: Agent-skill marketplaces and OpenClaw-style hosting take a slice of the wrapper market as managed execution moves up the value chain
- 2027: Hosting platforms and AI-SaaS aggregators buy up more wrappers; independent market research points to OpenAI and Meta circling OpenClaw as a sign of direction
Emerging Opportunities
A vertical wrapper for one named profession. The Micro-App Portfolios report says it plainly: "AI property descriptions for Texas realtors" beats "AI writer." Pick one job in one industry where you can get a list of people to contact directly (chiropractors, Etsy shop owners, pediatric dentists) and ship a $29-$99/month tool. Which profession could you email fifty members of this week?
Portfolio-OS micro-SaaS. Independent market research names this as an emerging category. Solo operators are stitching dashboards together across Stripe, PostHog, and hosting. Build the single tool that joins them. Price it at $49-$199/month based on connected apps.
AI-powered app maintenance service. An opportunity independent market research puts at $20-$50/app/month. Founders running 10+ apps face a growing maintenance load. Offer dependency updates, security patches, monitoring, and bug fixes as a managed service.
A boilerplate or template product. ShipFast clears $130k+/mo selling a Next.js starter kit at $199 one-time. The market for "skip the boring parts" templates grows as more solo builders ship wrappers. Pick a stack (Next.js, Bun, Astro, mobile) and make the best opinionated starter.
Common Objections & Counterarguments
"Wrappers have no moat." The AI-Powered SaaS report answers this directly: "If wrappers solve problems, users will pay for them." Cursor is technically a wrapper around language models and crossed $1B ARR. Your moat comes from distribution, brand, and unit economics. The underlying API is the same one everyone else rents.
"OpenAI/Anthropic will build this themselves." The OpenClaw report answers: "Platform owners historically underinvest in hosting relative to core products." OpenAI is unlikely to build "AI property descriptions for Texas realtors". It is too narrow and too distracting for them. Big players leave the wrapper niche to people like you.
"You're building a graveyard." Independent market research admits this one. The workload of running 10+ apps is real. The fix is the managed maintenance service named above, plus cutting underperformers quickly. Pieter Levels cites a 5% hit rate as the math; the other 95% should die fast and cheap. Could you shut down an app you built without taking it personally?
"The wrapper window is too short to build a real business." Independent market research names the 90-day window for commodity wrappers, but Cursor, Lovable, and Bolt.new show that wrappers with security, distribution, or a category brand last for years. Your way out of the bubble is to grow from "wrapper" into a product with a workflow your users depend on.
Picture the Stripe notification arriving while you are making dinner. Someone you have never met paid for a tool you built on a weeknight. Stack enough of those and the money covers the car payment, then the childcare, then a month where you stop counting down to payday.
Narrow niches get claimed one by one. Once a profession has a tool it trusts, switching takes a lot of persuading, so the builder who arrives first keeps the customer. The niche you know from your own job is still open today, and someone else in that job is thinking about it too.
AI Companion Apps
That is the wrapper pattern, where you sell one job done cleanly every time. Companions earn their money a different way, and the question for you here is whether you want to build something people come back to day after day.
App intelligence firm Appfigures, in data provided to TechCrunch in August 2025, counted 337 active revenue-generating AI companion apps worldwide, 128 of them released during 2025. The mobile segment generated $82 million in the first half of 2025 and was on track for over $120 million by year end, against $221 million in worldwide consumer spending across the category's lifetime to that point. Downloads reached 220 million globally, with first-half downloads up 88% year on year.
You will see far bigger figures quoted for this market. Treat them with care, because they measure different things. Estimates range from under a billion dollars to more than thirty billion, depending on whether the definition covers mobile consumer spending, all relational AI, or projected future markets, so any single headline number tells you how someone chose to define the market, and very little about how much money is in it.
How the money is spread matters more to you than the total, and Appfigures published that too. The top 10% of AI companion apps generate 89% of the revenue in the category, and only around 33 apps have ever exceeded $1 million in lifetime consumer spending. Revenue per download across the category ran at $1.18 in 2025, up from $0.52 in 2024.
Read those three numbers together before you build. A category can grow quickly, be worth a modest amount overall, and still pour almost all of its money into a handful of apps. That is the real shape of this market, and it is a harder place to enter than a billion-dollar headline makes it look. Out of 337 apps, what would make yours one of the 33?
This market runs on a basic human need: the wish for conversation, connection, and company, now met in part by AI that has become good enough to do it. Whether that is good for society is a fair debate. What nobody can argue with is that millions of people pay real money for this product every month.
Understanding the Market
Why People Use AI Companions
People use AI companion apps for more reasons than the "lonely people talking to robots" stereotype suggests.
Entertainment and roleplay. The largest segment. Users create or chat with fictional characters for storytelling and roleplay. Character.ai's most popular characters are fictional: anime characters, game characters, original creations. Think of it as fiction written together, sitting alongside real relationships.
Language learning. Practising conversation with an AI that never judges your pronunciation or grammar. The AI can match its level to yours and explain your mistakes in real time. This is one of the most defensible niches in the space.
Emotional support and venting. Some users treat an AI companion as a safe place to work through feelings, vent about the day, or rehearse a hard conversation. Therapy is a separate thing. No AI companion should market itself as a mental health treatment.
Productivity and coaching. Companions that act as accountability partners, writing collaborators, or brainstorming partners. Less emotionally charged than the other uses, and growing.
Which of these four would you actually use yourself on a bad Tuesday?
The Competitive Landscape
Character.ai dominates the general-purpose market with 20M+ monthly active users. Taking them on head-to-head is extremely hard. Leave that fight alone.
Replika pioneered the AI companion space and focuses on personal AI friends and romantic companions. It has faced regulatory trouble over NSFW content.
Chai is a mobile-first platform with a younger user base and a focus on entertainment.
Your opening is in the niches these platforms handle badly. Character.ai is broad and shallow in any one use case. A purpose-built AI language tutor, a fandom-specific character platform, or a mental health check-in bot can serve its niche better than a general-purpose platform can.
Building Your AI Companion
The Fastest Path: Telegram or Discord Bot
You can build and launch an AI companion in a weekend using a Telegram or Discord bot. It takes very little technical skill and no mobile app development at all.
How it works: You create a Telegram bot using the Telegram Bot API, connect it to OpenAI or Claude's API, and write a system prompt that shapes the AI's personality, knowledge, and way of talking. The system prompt is the product. It decides whether your users find the AI compelling or dull.
Advantages: No app store approval, no mobile development, instant deployment, and Telegram/Discord handle the interface for you. Your users can start chatting straight away from apps they already have.
Limitations: You are stuck with Telegram/Discord's interface. No custom UI, no rich media beyond what the platform supports, and you depend on a third-party platform.
The Custom App Path
For a full-featured AI companion, build a mobile or web app using React Native, Flutter, or a web framework. That gives you full control over the experience: custom UI, voice messages, image generation, onboarding flows, and monetization.
Tech stack for a solo developer:
| Component | Choice |
|---|
| Frontend | React Native (cross-platform iOS/Android) or Next.js (web) |
| AI | OpenAI GPT-4o-mini or Claude Haiku for cost-efficient conversation |
| Database | Supabase or Firebase for user data and conversation history |
| Payments | Stripe or RevenueCat for subscription billing |
| Hosting | Vercel (web) or cloud functions for the API layer |
Development timeline: If you are an experienced developer working solo, you can build an MVP in 2-4 weeks. If you are non-technical and using no-code tools, plan on 4-8 weeks.
The System Prompt Is Your Product
In the companion space, your system prompt is what sets your product apart. Two apps on the same underlying model (GPT-4o) can feel completely different depending on how the system prompt shapes the AI's behaviour.
A great system prompt defines:
- Personality: How the AI speaks, its tone, vocabulary, and communication style
- Knowledge: What the AI knows about and can discuss competently
- Boundaries: What the AI will not discuss or engage with
- Memory instructions: How the AI references past conversations and remembers user preferences
- Behavioral rules: Response length, emoji usage, question-asking patterns, and conversational flow
Spend more time refining your system prompt than on any other part of your product. Test it hard across many kinds of conversation. Whether your companion feels compelling or dull comes down almost entirely to the system prompt.
Monetization Models
Freemium Subscriptions (Recommended)
This is the model that has worked in this space. Offer a free tier with limited daily messages (20-50) and a premium subscription ($5-$20/month) that opens up unlimited messages, premium characters, voice features, and conversation memory.
Pricing benchmarks:
| Service | Monthly price |
|---|
| Character.ai c.ai+ | $9.99 |
| Replika Pro | $14.99 |
| Chai Premium | $13.99 |
| Niche bots on Telegram | $4.99-$9.99 |
Start at the lower end ($4.99-$9.99) so it is easy for people to say yes and more of your free users move to paid. You can always raise prices later as you add features.
In-App Purchases
Sell tokens, credits, or gems that your users spend on premium interactions, new characters, or special features. This works well next to subscriptions and catches spending from users who want one feature without signing up for a monthly plan.
Tips and Donations
Platforms like Ko-fi and Buy Me a Coffee let users choose to support you. This suits community-driven projects better than commercial apps.
Managing API Costs
API costs are your biggest variable expense and the main threat to your profit.
Cost per message: Using GPT-4o-mini, a typical exchange (user message + AI response) costs you $0.001-$0.003. A heavy user sending 200 messages per day costs you $0.20-$0.60/day or $6-$18/month.
Keeping costs down:
- Use smaller models (GPT-4o-mini, Claude Haiku) for most conversations and save larger models for premium features
- Summarise conversations to shorten the context (instead of sending the whole history with every message, summarise older messages)
- Cache common responses for frequently asked questions
- Set message length limits so the AI does not ramble
- Consider open-source models (Llama 3, Mistral) on your own infrastructure for maximum cost control
Unit economics check: If your subscription is $9.99/month and your average user costs you $8/month in API fees, you lose money as you grow. Aim for API costs below 30% of subscription revenue if you want a business that lasts. If your most loyal user is also your most expensive one, have you priced for that?
Safety and Ethics
You cannot skip this section. AI companion apps sit on ethically complicated ground, and ignoring safety will eventually wreck your product through lawsuits, platform bans, or a media backlash.
Age verification. Put meaningful age gates in place if your content is unsuitable for minors. App stores require you to comply with COPPA (US), GDPR (EU), and similar rules. An "Are you 18+" checkbox falls short for regulated content.
Crisis detection. If a user expresses suicidal thoughts, an intention to self-harm, or other crisis signals, your AI must respond properly: give crisis hotline numbers and encourage the user to reach a human for help. You owe this to your users ethically, and you carry the legal liability if you fail.
Content moderation. Decide what your AI will and will not engage with, and enforce those limits in your system prompt and through content filtering layers. Write your policies clearly into your terms of service.
Transparency. Your users should always know they are talking to an AI. Never design your product to mislead people about who, or what, is on the other end.
Data privacy. Conversations with AI companions are often deeply personal. Encrypt stored conversations, keep as little data as you can, and let users delete theirs. A data breach from a companion app would be devastating to your users and to your business. If someone's late-night conversations leaked tomorrow, could you face them?
Growth and Marketing
TikTok and YouTube Shorts
Short videos showing interesting or funny conversations with your companion convert better than any other format. Screen-record a good conversation, add a caption and trending audio, and post. People are naturally curious about AI conversations, so the content does much of the marketing for you.
Reddit and Online Communities
Language learning subreddits, anime communities, and AI enthusiast forums put you in front of exactly the right people. Share your product honestly: explain what makes it different and offer free access in return for feedback.
App Store Optimization
If you build a mobile app, ASO (app store optimization) is critical. Research the keywords your users search for: "AI friend," "AI chat," "character chat," "roleplay AI." Tune your app title, subtitle, and description for those terms.
Partnerships
Work with content creators in your niche. A language learning YouTuber showing off your AI conversation partner reaches exactly the right audience. Offer them a revenue share or a flat fee.
The AI companion market is real, growing, and profitable. It is also ethically complicated and changing fast. Go in with a clear niche, strong safety practices, and realistic expectations about the engineering and day-to-day work involved.
Related: AI Wrapper Apps | Vibe Coding | AI Automation Agency
2026 Market Snapshot
Independent market research's Virtual AI Companions report frames 2026 as the year companion apps grow from a curiosity into a real category. The named players cover the range: Pi for general companionship, Replika for romantic relationships, Character for personality-based companions, Forever Voices turning influencers into virtual companions, and vertical apps like Cleo (finance), Kai (well-being), Clare (voice-based mental health), Melli (elderly), and Polly (assistive). The AI-Powered SaaS report supplies the pricing template: Browserbear at $949/month subscriptions, single-use plans at $4-$39 per session, and freemium with capped credits. Put together, that explains how some companion apps reach seven-figure revenue with tiny teams. Remember the Appfigures split above, though: only around 33 have ever crossed $1 million.
- Caryn Marjorie's AI companion charges $1/minute, the named benchmark for celebrity-companion pricing
- Replika has grown to millions of paying users on customizable companion subscriptions
- Eugenia Kuyda's Roman Bot: cited in independent market research as proof that "immortal" deceased-person companions resonate emotionally
- Microsoft Edge Bing Chat and Perplexity AI represent the conversational-internet shift independent market research names
- Single-use pricing examples from AI-Powered SaaS (Roast My Web packs, Xound at $4.99/file) carry straight over to per-session companion pricing
Key Players to Watch
Between them, the companion apps, AI-SaaS infrastructure, and wider virtual-companion stack cited in independent market research mark out the lane you would be working in.
- Pi: Personal AI companion for support and advice
- Replika: Customizable companions including romantic relationships
- Character: Personality-driven companions based on people and characters
- Cleo: Finance-companion vertical
- Kai: Well-being vertical
- Clare: Voice-based mental health coach
- Melli: Voice-controlled companion for the elderly
- Polly: Assistive AI for neurological conditions
- Forever Voices: Influencer-to-companion conversion studio
- Chai: Companion creation, sharing, and discovery platform
- Claude (Anthropic), OpenAI APIs, Mistral: Underlying model providers
- Twelvefold, 1811 Labs, Slimmer AI, Pawa, VirASTRAL: AI startup studios named in independent market research, building portfolio companion plays
Predictions for 2026-2027
- Q3 2026: Voice-first companions go mainstream. Independent market research names Melli and Clare as voice-native; Apple, Samsung, and Bolt are embedding companions in hardware
- Late 2026: Influencer-companion licensing becomes a standard way for creators to earn; independent market research cites Amouranth and Caryn Marjorie as templates, with Forever Voices as the production layer
- Mid-2027: "Immortal" deceased-person companions grow from an edge case into a small, serious vertical, especially in grief support and family memory keeping
- 2027: Vertical companion apps (finance, fitness, study, language) earn more per user than general-purpose companions, as specialised products pull ahead of broad ones
Emerging Opportunities
A vertical micro-companion. Pick one job-to-be-done (sober coaching, grief support, exam prep, language conversation, ADHD accountability) and ship a $9-$29/month companion. Independent market research names Cleo, Kai, Clare, Melli, and Polly as working examples of the vertical-companion model. Is there one struggle you know from the inside well enough to build for?
Turning influencers into companions. Independent market research names Forever Voices as the studio template. Build the agency layer: license a creator's voice and persona, ship a Replika-style app to their fans, and charge $19-$49/month with a revenue share for the creator.
A B2B companion for service businesses. Independent market research names Crisp's MagicReply, Drippi, and Wendy's AI ordering as examples. Build a companion that handles 24/7 customer support for one vertical (dental offices, fitness studios, salons). Charge $99-$499/month.
A hardware-embedded companion. Independent market research predicts AI built into hardware: VoiceKitt for cars, Bolt for scooters, Samsung in appliances. As a solo operator, you can ride this by building the software companion layer for niche hardware partners (smart-home devices, hearing aids, kids' toys).
Common Objections & Counterarguments
"AI companions worsen loneliness and isolation." Independent market research names this objection directly. The counter: companions help people practise real-life conversations, are available 24/7 in a way humans cannot be, and serve people who are often overlooked (the elderly, neurodivergent people, people living far from others). Like therapy or journaling, a companion is a tool, and how it turns out depends on how it is used.
"Companions give harmful advice." A real risk, and independent market research names it. Reduce it with content filters, routes to human professionals, clear disclaimers, and staying in your vertical (a finance companion stays in finance). Avoid medical and legal advice, or partner with licensed providers.
"This is a feature OpenAI / Anthropic will build." The AI-Powered SaaS rebuttal applies: platform owners underbuild for vertical needs. ChatGPT is unlikely to become a grief-support companion or an ADHD coach. Those need product depth that a model swap cannot supply.
"Engagement is addictive. This is harm-coded." A genuine ethics issue. Answer it by designing for healthy use: session time limits, nudges toward in-person support, and refusing to chase daily-active-user numbers in ways that exploit vulnerable users. Independent market research's framing of "meaningful connection with personalized support" points to the right design intent. Would you be comfortable if your own teenager used what you built at 2am?
At some point a user writes in to tell you your companion helped them through a hard week. You have people who come back daily, a system prompt you shaped line by line, and a business with your name on the Stripe account. Being the person who built something people choose to return to is worth protecting, which is why the next section matters.
Companion Apps Are Now Regulated, and the Liability Is Yours
You now know how a companion app holds someone's attention. Read this before you build one, because it covers what you take on in your own name the day you ship.
If you are building anything in the companion category, this section matters more to you than any growth tactic on this page. The rules changed in 2025, and most of what has been written about this opportunity was written before that.
What California SB 243 does
California enacted Senate Bill 243, signed on 13 October 2025, the first comprehensive state law in the United States aimed specifically at companion chatbots. Some duties applied immediately and others phase in by 1 July 2027.
The definition is broad, and it is written to catch products like the ones this page describes. A companion chatbot is an AI system with a natural language interface that provides adaptive, human-like responses and is capable of meeting a user's social needs, including by exhibiting anthropomorphic features and being able to sustain a relationship across multiple interactions. Carve-outs exist for chatbots used solely for customer service or business operations, for video game bots limited to discussing the game, and for standalone consumer devices that do not sustain ongoing relationships.
Hold that definition up against a roleplay or companion product and the answer is usually that you are in scope.
The law applies to any operator making a companion chatbot platform available to users in California. Where your company is based makes no difference. If you are a solo developer anywhere in the world shipping to the App Store without a geographic restriction, you are making your product available to users in California.
What you must do
Disclosure. Where a reasonable person could be misled into thinking they are talking to a human, you must give clear and conspicuous notice that the chatbot is artificially generated. The statute does not spell out how the reasonable person test applies, so the safe course is to disclose even where you think your users obviously know.
Extra duties toward minors. For users you know to be minors, you must disclose that they are talking to AI, and give a clear and conspicuous notice at least every three hours during ongoing conversations, reminding them to take a break and that the chatbot is not human. Every platform must also disclose that companion chatbots may not be suitable for some minors.
Safety protocols. You must keep protocols to prevent content relating to suicidal ideation, suicide or self-harm, including notices that point users to crisis services when a user expresses those feelings. You must publish details of those protocols on your website. For known minors, you must take reasonable steps to stop the chatbot producing visual material of sexually explicit conduct or telling the minor to engage in it.
Annual reporting. From 1 July 2027, operators must report each year to the California Office of Suicide Prevention on the number of crisis referral notices issued and the protocols in place. Reports must leave out user identifiers, and the office publishes data from them.
The part that changes your risk
Individuals injured by non-compliance may bring a civil action for injunctive relief and damages.
A private right of action exposes you in a different way from regulatory enforcement. A regulator has limited attention and usually starts with the big operators. Private claims skip that queue, and the plaintiffs' bar does not wait for a regulator to move first. For you as a small operator, this is the clause that makes compliance a condition of staying in business. If one user's family sued you next year, would your records show you did what the law asks?
What to do before you launch
Decide honestly whether you are in scope, measured against the statutory definition. How you would prefer to describe your product carries no weight. The law has no category called "just a roleplay app".
Build the disclosures in from day one. Bolting a three-hour reminder onto a session design that never planned for one is more work than including it from the start.
Decide how you handle age, because several duties turn on users you know to be minors. This cuts both ways: choosing on purpose to stay ignorant may not protect you, and it rules out the defence that you complied with the duties toward minors.
Write and publish your safety protocols, and make them real. Publishing is a legal duty, and a published protocol you do not follow is worse than having none at all.
Get the crisis-referral path working and test it, because it is the clause most directly tied to the harm the statute exists to prevent.
And treat California as the template for what is coming. It is the first such law, other jurisdictions are drafting their own, and more will follow. Building to the stricter standard now costs you less than retrofitting for each new one.
Each week you wait, the prompt you perfected stays trapped in your chat history, earning nothing. The safety and compliance steps above take care, and you can start them now. Write the landing page headline tonight, list the protocols you would publish, and send the link to one person with the problem.