How to Monetize an AI App or Chatbot (Free Users Too)

How to monetize an AI app or chatbot

You monetize an AI app or chatbot with a paid plan (a subscription, prepaid credit packs, or usage-based pricing) plus a free tier with limits, and you cover part of the free tier’s cost with ads, rewarded credits, or affiliate recommendations. Teams and developers can also pay for API access or business plans. The difference from other apps is that every free message costs you real money in model inference, so the free tier has to be priced and capped as carefully as the paid one.

For app monetization in general, read How to monetize an app first.

What a free user costs you

Most AI apps pay a model provider per token: input tokens (system prompt, chat history, and the user’s message) and output tokens (the answer). A typical app pays almost nothing to serve one more free user. An AI app pays for every message.

Prices vary a lot between models. On Anthropic’s API pricing page, Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, while Claude Haiku 5.5 costs $0.10 and $0.50 (for prompts up to 100,000 tokens). Google’s Gemini API pricing page lists Gemini 2.5 Flash-Lite at $0.10 input and $0.40 output per million text tokens on the paid tier. Prices checked on October 22, 2026; they change often, so use the live pages.

Illustrative example: cost per user vs revenue per user

Sample figures only. Assume an average message uses 2,000 input tokens and 400 output tokens. At the Anthropic prices above, one message costs about $0.008 on Claude Sonnet 5.5 and about $0.0004 on Claude Haiku 5.5. The subscription is $9.99 a month, and you keep $8.49 after a 15% store fee. The $1.00 offer payout is also an assumption.

Three lessons follow. Serve the free tier on a cheap model: the same 300 messages cost 20 times more on the larger one. Put a fair-use cap or a fallback model on paid plans, because one heavy subscriber can wipe out the margin of several normal ones. And price free credits (from offers, ads, or referrals) so their inference cost stays below what the action pays you.

Chat history grows with every turn, so input tokens often outweigh output. Anthropic’s page also notes that Claude 4.7 and later models use a tokenizer that produces about 30% more tokens for the same text, so measure tokens per message in your own logs rather than guessing. Prompt caching helps: on that page, a cache hit costs 10% of the base input price on most Claude models (5% on Sonnet 5.5 and Opus 5.5), which cuts the cost of resending a long system prompt.

AI app monetization models compared

Most AI apps that earn well run two or three of these at once. Hybrid monetization explains how to give each model its own job without one cannibalizing another.

Subscriptions, credit packs and usage-based pricing

Subscriptions

A flat monthly price is easy to sell, so most consumer chat apps lead with one. The catch is the heavy user from the table above. Protect the plan with a stated monthly allowance, a slower or smaller fallback model after the allowance, or a higher tier for power users. Track churn rate and inference cost per subscriber together; a plan that keeps users but loses money on each one is not working.

Credit packs

Credits work well when actions differ in cost, for example when an image costs you as much as 50 chat messages. Sell packs, charge each action a fixed number, and show the balance everywhere. Keep the unit simple: 1 credit = 1 standard message beats a token count. If you sell credits through Apple’s in-app purchase, they may not expire (see the app store section below).

Usage-based pricing

Charging per message, per minute of audio, or per 1,000 tokens matches your cost exactly, which suits developers and businesses. Consumers dislike open-ended bills, so most consumer apps use prepaid credits instead. Billing platforms support both: Stripe’s usage-based billing docs cover recording usage, billing credits for prepaid products, and usage alerts.

Setting free limits that still grow the app

The free tier is your marketing: good enough that people come back, capped enough that it does not eat your budget. Levers that work:

• Model routing. Free users get a small, fast model; paid users get the larger one. Show the difference so upgrading feels worth it.
• Daily caps, not monthly. A daily limit brings users back every day and stops one user from spending a month’s budget in an afternoon.
• Feature gates on costly actions. Long documents, file uploads, image generation, and voice cost far more than short text; put these behind paid plans or credits.
• Context limits. Shorter history windows for free chats cut input tokens a lot.
• Accounts first. Require sign-in before the first message, so one person cannot reset limits by clearing cookies.

How to monetize free users covers the wider question of earning from people who never pay.

Ads in a chat interface: what works and what does not

Ads can work in AI products, and large players are testing them. In February 2026, NBC News reported that OpenAI started testing ads on ChatGPT’s free and $8 Go tiers, clearly marked and visually separated from answers, and not shown to users under 18. For smaller apps, Google told Bloomberg in 2025 that “AdSense for Search is available for websites that want to show relevant ads in their conversational AI experiences”, as TechCrunch reported; that is an existing Google program, so you apply and follow its policies like any other publisher.

The limits are real, though:

• Few impressions. A chat is often one page open for many turns, so banners earn little per session compared with its inference cost.
• Trust. Users ask chatbots for advice. An ad that looks like part of the answer, or an answer that seems shaped by an advertiser, damages the product. Keep ads outside the answer and label them.
• Context you do not control. The model may write things next to an ad that the advertiser would not want, and topics like health or personal problems are poor places for ads.
• Store rules in mobile apps. Full-screen ads between messages count as interstitials, and Google Play limits disruptive ads; Offerwall vs rewarded video compares opt-in formats that avoid that problem.

Ads fit best in high-volume tools on cheap models, such as translation or search-style bots.

Rewarded credits: free users earn more usage

Rewarded credits turn a free user’s time into usage. When a user hits the daily limit, you show an option to earn extra credits by completing an offer from an advertiser, such as trying an app, signing up for a service, or answering a survey, through an offerwall. The advertiser pays when the result is confirmed, you earn per completed offer, and the user gets credits in your app. Apple’s guideline 3.2.2(x) says apps “may otherwise incentivize users to take specific actions within apps (such as by completing a level or watching an ad)”, as long as nothing forces users to download other apps to use yours.

How to set it up so it pays:

1. Price credits against inference. Decide how many credits you give per $1 you earn, then check what those credits cost you in tokens. In the example above, 60 large-model messages cost $0.48 against a $1.00 offer. The virtual currency exchange rate guide and the virtual currency calculator help with the math.
2. Place it where the limit bites. The best moment is the “you have used today’s free messages” screen, next to the upgrade button, plus a permanent “Earn credits” entry in the menu. Offerwall placement covers more spots.
3. Credit only on the server. Add credits when a signed postback reaches your backend, never from the app or browser, and remove them when a result is reversed. Postback security explains signing, and preventing duplicate reward credits covers retries.
4. Handle used credits. AI credits are often spent within minutes, before an offer can be reversed. Let the balance go negative on a reversal, or hold large rewards for a while; reward hold periods covers the trade-off.
5. Watch for farms. Free AI usage attracts multi-accounting. Stop multi-accounting and VPN abuse lists the checks.

Affiliate recommendations and API or B2B plans

Affiliate recommendations

Shopping, travel, finance, and product research bots recommend things all day, and affiliate links pay a commission when a user buys. Two rules matter. First, disclose: the US Federal Trade Commission’s Endorsement Guides FAQ says “the closer the disclosure is to your recommendation, the better”, calls “Paid link” next to the link adequate, and says a disclosure only on your home page is not enough. In a chat, that means a label on the link inside the answer. Second, keep answers honest: rank products on what fits the user, not on commission rate, or users will stop trusting the bot. Affiliate disclosure rules covers the details, and Offerwall vs affiliate links compares the two models.

API access and B2B plans

If your app has something others want (a tuned pipeline, a niche data set, or a workflow), sell it to businesses as team seats, an API with per-request pricing, or a white-label version. Business buyers pay more and churn less, but expect invoices, uptime, and support. Meter usage from day one.

App store rules for selling AI credits and plans

Checked on October 22, 2026. These are summaries; read the live text before you submit.

• Apple requires in-app purchase for digital features. Apple’s App Review Guidelines (3.1.1) say that to sell features or functionality in your app, including subscriptions and in-game currencies, you must use in-app purchase, and that “any credits or in-game currencies purchased via in-app purchase may not expire.” Monthly credit packs that vanish at the end of the month are a problem on iOS; a subscription with a monthly allowance is the usual way around it.
• Links to web payments depend on the storefront. Under 3.1.1(a), apps on the United States storefront may include buttons and links to other ways to pay; other storefronts need an entitlement or do not allow it.
• Web purchases can carry over. Guideline 3.1.3(b) lets multiplatform apps give users access to what they bought on your website, provided those items are also available as in-app purchases in the app.
• Disclose third-party AI. Guideline 5.1.2(i) says you must clearly disclose where personal data will be shared with third parties, “including with third-party AI”, and get explicit permission first. That covers sending user prompts to a model provider.
• Google Play requires Play Billing for digital goods. Google’s Payments policy covers app functionality, digital content, virtual currencies, subscriptions, and “cloud software and services”, with exceptions such as physical goods. Google runs opt-in alternative billing and external link programs in eligible regions; its US alternative billing page says enrolled developers report transactions and pay service fees starting October 1, 2026.

Earned credits from offers are not a sale, so they do not go through in-app purchase, but anything you sell still does. Can you put an offerwall in an iOS app? covers Apple’s rules for rewarded walls, and Google Play’s incentivized install policy covers the Android side. Store fees by tier are in the fee table in How to monetize an app.

Frequently asked questions