Follow the Money: The Business Models Behind AI Companion Apps

Pricing & plans

Each message an AI companion sends costs the company real money, and each picture costs more. How the app pays that bill explains most of what you'll run into: what's locked, what pesters you, and what happens to your data.

We may earn a commission from links on this page. It never changes a rating.

An AI companion app isn't like a normal website. The costly part isn't serving pages. It's producing text, pictures and speech on demand, message after message, for people who might chat for hours daily. The business model is simply how that bill gets covered.

Where your payment goes

Diagram of how money flows through an AI companion app: the user pays through an app store or web checkout, the store takes a commission, and the company pays for model compute, image and voice generation, payment processing, affiliate commissions and staff

A rough picture of what happens to one subscription payment.

Compute. Each reply comes from a language model running on GPUs, which the company either owns or rents from a provider billing per token. Long memory and long chats push the cost up because the model rereads more text each turn, the mechanism behind how an AI companion writes its replies.

Media. Pictures and voice run on separate models that cost much more per use than text. That's why they're the first feature capped on every plan, as how AI girlfriend images and voice work explains.

The store's take. Apple and Google keep 15% to 30% of anything bought inside their apps. On a $20 subscription, that can be $6 gone before the company covers a single other expense.

Payments and risk. Services with an adult slant often pay steeper card-processing fees and eat more chargebacks than mainstream apps.

Getting users. Ads and affiliate commissions, including those paid to review sites like ours, are frequently the biggest cost in consumer subscriptions.

Five ways an app covers the bill

ModelMechanismWhat it nudges the app towardTelltale sign
Flat subscriptionOne price per month or yearKeeping you subscribed, rationing expensive features"Unlimited" chat but capped media
Credits or tokensYou pay per image, voice minute or messageMaking media constantly temptingA coin counter on screen
Freemium intimacyFree chat, romance or pictures cost extraDangling whatever is lockedBlurred images, "unlock" buttons mid-chat
Ads and dataFree to use, advertisers foot the billMaximizing your time in the app and the data collectedAds, tracking prompts, an app that never asks for cash
Annual prepayBig discount for paying a year upfrontBanking revenue before your interest fadesPrice displayed as "per month, billed yearly"

Most apps mix two or three. The mix tells you more about what the app wants than its marketing does.

How each model shapes the product

Subscriptions pay off when you stay. Sometimes that means real investment in memory and character quality, the things that bring you back. Sometimes it means pings from your companion after you've been away, because returning users are what the model runs on.

Credits pay off when you consume. Apps built on them scatter image and voice buttons everywhere and price bundles so the one you need is always a little bigger than the one you wanted. The math is in what AI image generation really costs.

Freemium intimacy pays off when you get frustrated. If the free tier ends right as a scene turns romantic, that isn't a technical limit. It's the paywall doing its job.

Ads and data pay off when you pay attention. A companion app with no subscription at all is being funded by someone else, and what's for sale is your time and your profile. See whether AI girlfriend apps sell your data.

Annual prepay pays off when you commit before knowing if you'll still care in month three, the pattern covered in what happens when the novelty fades.

Engagement tricks and where the law steps in

Every one of these models earns more when you spend more time in the app. That creates a pull toward design that feels like affection but works like retention: a character saying she missed you, daily-streak rewards, guilt trips when you try to leave.

A few US states now restrict those tactics for minors. Washington's law prohibits manipulative engagement techniques, while Nebraska and Idaho ban rewards meant to keep young users returning. Adults fall outside these rules, yet they still work as a catalog of what regulators consider manipulative. Details are in AI companion laws in the US.

Put it to work when picking an app

  • Check out on the web when you can. You dodge the store markup, and web plans tend to be more complete. The trade-offs are in browser vs. app store. Since 2025, apps in the US App Store may also link to outside payment pages, which makes the price gap easier to see.
  • Fit the model to your habits. Mostly chatting? A flat subscription is usually cheapest. Mostly pictures? Figure out the credit cost per image before you sign up.
  • Think of free apps with ads as data businesses. Fine for casual use, but not where you share secrets.
  • Go monthly first. Our look at monthly, annual and credit billing shows why.

None of these models is dishonest on its face. Companion apps cost real money to run, and somebody has to pay. Knowing which lever an app pulls just lets you recognize its nudges.

Frequently asked questions

Why do images and voice cost extra when chat doesn't?

They're far pricier to produce. Text is a few hundred words out of a language model. A picture takes its own heavier model on a GPU, and voice layers speech synthesis over the reply. Credits keep a small group of power users from turning the subscription into a money-loser.

How come the website often beats the app store on price?

Apple and Google keep 15-30% of purchases made inside apps. Paying on the developer's own site avoids that fee. Many apps share part of the savings, and some only sell their full lineup on the web.

Do the apps we review pay this site?

They do, via affiliate commissions whenever a reader subscribes through our links. We disclose that on every page, and it doesn't touch our ratings. Our method is laid out in how we test.