A thousand subscribers paying $15 a month sounds like a pleasant business. That’s $15,000 arriving every month, presumably while you work on something more interesting than chasing invoices.
Then somebody starts sending your character hundreds of messages a day. Another customer generates images all evening. The model bill arrives, several subscribers cancel, and suddenly the word “unlimited” on your pricing page looks expensive.
There is money in AI companionship. Building a business around it means deciding what people will keep paying for and what you can afford to deliver. Those decisions belong in the first prototype, before you commission fifty characters.
The market is already collecting real money
Sensor Tower’s June 2026 analysis of AI apps put AI companion mobile-app revenue at $150 million in the first quarter of 2026, more than twelve times its level in the first quarter of 2023.
The broader generative-AI mobile category generated $6.1 billion in in-app purchase revenue between Q2 2025 and Q1 2026, up 232% year over year.
Keep the definitions attached to those figures. They describe measured mobile-app revenue, not the entire AI economy. They don’t tell you how much operators spent, and they aren’t an estimate of industry profit. Web subscriptions and business contracts are different revenue channels.
For a small founder, the interesting part is that customers already pay for conversational entertainment. You don’t have to invent that habit. You do need a reason for someone to choose your product and return after the novelty wears off.
“An AI friend for everyone” gives you an enormous audience to advertise to and almost nothing specific to say.
Start with a character people can recognize
Consider a narrower proposal: an adult interactive-fiction subscription with a recurring detective character. Subscribers investigate a fictional case through conversation, with new story material released regularly.
That gives the customer something to buy beyond access to a chat window. They’re paying to spend time in a particular fictional world. It also gives the founder concrete work: write the setting, establish what the detective knows and test whether conversations remain consistent.
You can prototype with an existing language-model API, a simple web interface and a small set of original characters. Training your own foundation model would be an extraordinary place to start this experiment.
Keep the first version small. One character, one setting and a limited number of test subscribers will expose plenty of problems. Ask a tester to contradict the detective, restart a conversation and refer to something said yesterday. Watch what breaks.
If the character forgets the case, a beautiful portrait won’t save the evening.
BizBoostPro’s guide to building paid services with AI assistance puts a specific offer near the beginning of the process. For companionship products, that offer needs an equally clear description: what subscribers get, how often it changes and which features cost extra.
Give the visual work a budget, too
Character images help establish the world before a subscriber writes the first message. A detective’s portrait should match the personality in the dialogue; otherwise, you’ve effectively hired two actors for the same part.
For an adult fictional-character project, the image generator on JOI is an example of a service for exploring character appearances and visual styles. Its image-generation page illustrates the visual side of this category; it doesn’t establish a business model or demonstrate anyone’s profitability.
A founder still needs to check the terms governing intended use, choose original characters and decide which images belong in the product. Generating twenty attractive faces can be an afternoon’s work. Keeping one character recognizable across a continuing story is a different production problem.
And your prototype needs ordinary product work: billing that makes sense, a way to delete conversation history and support when something fails. Customers notice those things immediately.
What could the profit actually look like?
Here’s an illustrative monthly budget, not a forecast or a statement about any existing service. Assume 1,000 paying subscribers, each charged $15, with no sales tax included in that price.
|
Monthly item |
Assumed amount |
|
Subscription revenue |
$15,000 |
|
AI generation, hosting and storage |
−$4,000 |
|
Payment fees and refunds |
−$1,000 |
|
Customer acquisition |
−$3,000 |
|
Support and ongoing story production |
−$2,000 |
|
Other software and administration |
−$1,000 |
|
Surplus before founder compensation and tax |
$4,000 |
|
Founder compensation |
−$3,000 |
|
Remaining operating surplus before tax |
$1,000 |
That leaves about 6.7% of revenue after the assumed founder compensation. A $1,000 increase in generation costs would wipe it out.
These are planning placeholders. Actual costs depend on the model, conversation length, image usage, payment channel and staffing. Free users also consume resources; this example hasn’t allocated them a separate allowance. Setup costs and financing would require additional provision.
The lesson is practical: meter expensive features before you promise them. A subscription could include a defined image allowance, with optional additional credits. Test the allowance against actual usage rather than guessing how restrained your most enthusiastic subscriber will be.
The second payment deserves your attention
RevenueCat’s 2026 subscription-app report found that AI apps generated 41% more realized revenue per payer over a year than non-AI apps at the median. It also reported weaker retention. These are broad AI-app benchmarks, not companion-only results.
That combination makes sense as a warning for founders. A striking demo can persuade someone to subscribe. The product then has to survive an ordinary Tuesday.
In the hypothetical budget above, suppose 10% of subscribers leave each month. Maintaining 1,000 subscribers means replacing 100 of them. At an assumed $30 acquisition cost per paying customer, replacement alone consumes the entire $3,000 marketing allowance. Growth would need additional spending or cheaper acquisition.
Measure where departures happen. After the first conversation? After a memory failure? When subscribers discover the image allowance? Those answers suggest different repairs.
Run a small paid pilot long enough to observe renewals. Talk to the people who leave, and resist explaining every cancellation away as a bad customer.
An AI companion business can earn money through a specific, well-maintained fictional experience. But before expanding the cast, look at the subscriber who came back without a discount. Find out what brought them back. That’s the part worth building next.