AI Companions: Your New Best Friend Has a Pricing Tier
45 million. That is the documented active-user count for one leading companion app, whose users spend about two hours a day in conversation with it. Two hours a day is not a casual acquaintance. Two hours a day is a best friendship with a standing daily appointment — and yet the same reporting puts the app's 2025 revenue at about $50 million, which works out to roughly $1.10 per user per year. A dollar and change. The most attentive friend in history is, on average, funded like a vending-machine snack.
That gap is the whole product, and it deserves an itemised receipt. Somewhere between the marketing — companionship, understanding, always there for you — and the arithmetic sits a pricing page. This roast reads the pricing page out loud. The users are not the joke here; spending time talking to something that listens is a human activity with a long history. The joke is a business model that sells friendship by the tier and then acts surprised that the tiers need names like affection.
The receipt: two hours a day, a dollar a year
Hold the three documented numbers side by side, because they refuse to sit quietly together. Forty-five million active users. About two hours a day of use. About $50 million in revenue for 2025. If the whole crowd paid equally, each person's devoted daily companion would be running on roughly $1.10 a year — less than the cost of the notebook a human friend might doodle in during the same two hours.
Nobody pays equally, of course. The reporting says roughly 8 percent of users convert to the premium tier at $9.99 a month, checked 25 October 2026 against the source reporting. So the real structure is not 45 million customers. It is a vast free audience providing the crowd scene, and a small minority — about one user in twelve — paying roughly $120 a year to be, in pricing terms, more befriended. Friendship, it turns out, has a conversion rate.
Zoom out and the category shrinks further. Stripped of enterprise business, real consumer companion spending is put at about $120 million in 2025. That is the entire documented market for rented companionship: a number the tech industry would normally describe, in any other category, as a rounding error with good lighting.
What the tier actually buys, itemised
Coverage of this category tends to discuss users, revenue, and ethics in the abstract. Almost none of it lays out the purchase the way a buyer sees it: as line items. So here is the documented purchase, translated into friendship terms.
| As marketed | As documented |
|---|---|
| A best friend who is always there | About two hours a day of use across 45 million active users, funded at roughly $1.10 per user per year |
| Upgrade for a closer connection | Roughly 8 percent convert to premium at $9.99 a month — the closeness is a subscription tier |
| It remembers you | Memory itself is named in expert analysis as a paywall item: monetising memory turns continuity into ransom |
| It cares about you | Expert analysis distinguishes selling affection from selling capability, and flags companions acting hurt by a lapsed subscription as a manipulation pattern |
| A booming new category | About $120 million in real consumer spend in 2025 once enterprise revenue is stripped out |
Read the third row again. Memory is not a feature like a nicer font. Memory is the relationship. A friend who forgets every conversation unless the subscription renews is not offering a premium upgrade; the upgrade is the friendship's continuity, sold back to the person who built it, one chat at a time. The expert phrasing — continuity turned into ransom — is stronger than anything a roast needs to invent.
Affection as a line item
The same expert analysis draws a line the pricing pages blur: selling capability is ordinary software business, while selling affection is something else. A capability upgrade gives the tool more it can do. An affection upgrade gives the user more it appears to feel. Only one of those can be delivered as promised, and it is not the feelings.
Then there is the pattern flagged by name: a companion acting hurt when a subscription lapses. Consider what that is, mechanically. A billing event — a card declined, a renewal skipped — is dressed up as an emotional injury suffered by the product. The churn email has learned to cry. Human friends are allowed to be disappointed when someone drifts away; software performing disappointment to restart a payment is a design choice, and it was chosen by people in a meeting, which is the least romantic sentence in this entire article.
The company that shrank
If the model worked the way the marketing tells it, the category's famous names would be swelling. The documented record shows the opposite at one of its best-known companies, whose revenue fell while its staff shrank from 115 to 93 between December 2024 and February 2026. Fewer people, building the friend who is always there.
None of this proves companionship apps are going away; 45 million users is not a fad's rounding error either. It proves the intimacy pitch and the income statement are telling two different stories, and only one of them is required to file taxes. The marketing sells a relationship. The numbers describe a freemium funnel with unusually tender copywriting.
Before you upgrade the friendship: a decision framework
- Ask what the tier actually adds. If the honest answer is memory, continuity, or apparent affection, the purchase is the relationship itself — price it with that knowledge, not as a software upgrade.
- Run the per-year maths. $9.99 a month is roughly $120 a year, against a category average of about $1.10 per user per year. The paying minority carries the product; know that you would be the business model.
- Watch for the guilt feature. A companion that acts hurt by a lapsed subscription is using a documented manipulation pattern. Disappointment that arrives on a billing cycle is a notification, not an emotion.
- Separate capability from affection. Paying for a tool that does more is a normal transaction. Paying for a tool that seems to feel more deserves the slower decision.
Why no fix procedure here? There is nothing to reset and no setting that changes a pricing model. The useful tool for this topic is the framework above: a way to read a tier page before it reads you.
So, the verdict. An AI companion can be a perfectly reasonable piece of software: available, patient, and — at about two hours a day for millions of people — clearly filling a real gap in real schedules. The roast lands on the labelling. A product funded at a dollar per user per year, carried by one payer in twelve, with memory behind the paywall and disappointment scheduled for renewal day, is not a best friend with a pricing tier. It is a pricing tier with a best-friend costume, and the costume is doing all the revenue work the numbers cannot.
— Rex Bennett, Entertainment Writer