Google AI Studio Apps: Who Pays for Usage?

Your app is ready to share with friends, but who pays when they use AI? It depends on the project and account the app uses to call services. This article separates AI costs, hosting costs, and trial entitlements, with calculations you can adapt to your app.

Multiple users are connected to a single server and the app owner's coin wallet. It conveys the cost of the service being made available to others.

Check which account is charged when users call the API

When using the app owner's Gemini API key, end users trigger requests, but usage counts against the key owner's project. Signing in with a personal Google account does not automatically transfer AI charges to that user. Check actual API calls and the linked billing account.

Google's Build mode documentation says shared-app API requests count against the creator's usage limits and paid models may incur charges. Apps deployed to Cloud Run through AI Studio use the creator's key for all users' Gemini requests. Sharing for testing and deploying hosting are separate stages requiring cost review. References Build mode documentation

Charges depend on the task and account allowance. Some usage may fall within free allowances or available credits; some models and capabilities require payment. Sharing does not always trigger a charge, and a free trial does not guarantee free use indefinitely.

Separate development, AI-call, and website-access costs

Before estimating costs, separate app creation, AI calls, and hosting. A free allowance for one part may not cover another. The following table is an account-review framework proposed by KrobClick, not a Google pricing package.

What to checkWhen it happensWhat should I ask?
Creating and editing appsAsk AI to write or revise code in AI StudioWhat are the account permissions and creation limits?
Calling Gemini in the appThe app sends data to the model for a task, such as summarizationWhich model, whose key, and which billing unit?
Hosting and add-on servicesHave the server receive requests, store data, or call other servicesAre Cloud Run, databases, file storage, or other APIs involved?

Google offers additional AI Studio benefits to Google AI Pro and Ultra members, but distinguishes experimentation benefits from production API use. Membership does not establish that all costs of a public app are covered. Read the membership benefits announcement

For AI costs, check pricing for the selected model and capabilities. Units vary, including input and output tokens or audio and video units. Additional tools may have separate charges. Check Gemini API pricing

Cloud Run charges by resources and usage patterns, such as CPU, memory, and requests. Free allowances and rates also depend on service details and region. If you add a database or file storage, review those service costs separately. Multiplying website visitors by token prices does not capture the full cost. Check Cloud Run pricing

Estimate AI requests, not just website visits

Suppose you build an app to summarize a paragraph. Count how often users click Summarize and how many model requests each click triggers. A review pass or error retry may mean more requests than clicks, depending on the actual code.

This hypothetical example illustrates the calculation. These are not Google prices or test results. Suppose 100 users each perform 3 tasks per month, with 2 AI requests per task: 100 × 3 × 2 = 600 requests. Suppose 100 users each run 3 tasks in a month, with 2 AI requests per task. Total requests are 100 × 3 × 2 = 600 requests

Assume each request uses 1,500 input tokens and 500 output tokens. Tokens are processing units, not a fixed number of characters or Thai words. For illustration, use rates of $1 per million input tokens and $4 per million output tokens:

At 1,000 users with other assumptions unchanged, this component would cost $21. In practice, response length, conversation rounds, and resent history may change. Measure averages from real use; do not immediately adopt example figures as your app budget.

This formula deliberately excludes credits, caching discounts, batch processing, add-on tools, and model-specific units. Hosting, taxes, exchange rates, and payment fees are also excluded. Substitute request counts, token volumes, and actual rates from the Pricing page, then check for additional items. For calculations in baht, state the exchange rate and date as well.

Credits, quotas, and budget alerts serve different purposes

Current Billing documentation describes Prepay and Postpay with account-specific conditions. Check your own Billing page first. Prepay deducts API charges from topped-up credits and may allow automatic top-ups. These credits cover Gemini API, not every other Google Cloud service. Read billing terms

Google says balance processing is delayed, so usage may exceed credit before the system stops. Credit is not always an immediate hard cap. With auto-reload enabled, check refill amounts, thresholds, and monthly limits. These limits exclude manual credit purchases and do not cap hosting costs.

An Alerts only budget sends notifications; it does not automatically stop spending. Google Cloud separately documents spend-cap budgets for supported services. Check the budget type and service coverage before treating it as a usage cutoff. The Alerts only budget has a warning function. It doesn't automatically stop expenses. Google Cloud documentation separately describes spend-cap budgets for supported services. Read the actual budget type and service scope before assuming it will stop usage. View budget differences and notifications

KrobClick recommends server-side per-user task and input-size limits, API and hosting usage monitoring, and AI shutoff at defined thresholds. Hiding buttons is insufficient if server requests remain possible. Request quotas limit usage, not a guaranteed monetary budget independent of data volume and model.

Check accounts, costs, and access before sharing the app

  1. Which part of the app calls AI? Does simply opening the page trigger calls? How many requests occur per click, and are there retries?
  2. Whose account is the request linked to? Check the API key's project against the project hosting the app. Do not assume they always belong to the same account.
  3. Which allowances apply? Check models, quotas, credits, plans, Billing, and subscription terms in your actual account. Check models, quotas, credits, plans, Billing, and subscription terms in your actual account
  4. Are there any costs besides AI? List hosting, databases, files, and external services, identifying who pays each bill
  5. How to prevent overuse? Test server-side limits and how to turn off AI tasks with a small group before releasing to the public.
  6. What happens when an allowance runs out? Explain temporary unavailability and avoid endless retries. Prepare a message explaining temporary unavailability without endlessly retrying

Build mode documents say keys are server-side secrets. Exported apps need correct destination-environment configuration. Keep keys outside public code and webpage files. Hidden keys alone do not prevent overspending; control the AI-calling endpoint too.

An AI-assisted app need not call AI every time it runs

Separate “Was AI used to write the code?” from “Does the code call AI at runtime?” A bill-splitting tool can calculate in the browser even if AI helped write it; calculations need not all go to Gemini. Receipt reading or text explanations may call AI separately only when requested.

KrobClick favors running calculations and suitable tasks in the browser, adding AI where it offers a clear benefit. This makes costs and data flows easier to trace. Hosting or other services may still cost money; inspect the actual system.

Before sharing, establish what each task calls, whose account it uses, and how usage stops at a limit. Then expand beyond trial users and revise estimates using actual usage. What does one task call, whose account does it use, and how does it stop at the limit? Once these three matters are clear, expand beyond trial users and periodically update estimates from actual usage

Sources

Reviewed official documentation on October 14, 2026, Thailand time. This article is KrobClick’s explanation of the documentation, not a test of actual billing. All example figures are hypothetical, not Google rates. Model terms, quotas, payment plans, and account access can change. Check Billing and Pricing before launch.