Free AI API Tiers: Your Prompts May Train the Model

Free vs paid AI API data use, compared

Each provider's terms decide what happens to the text, files and images your app sends. The table summarises the main ones as of 19 October 2026. Terms change, so re-read the linked pages before you build.

Google's terms are the clearest example. For unpaid services, Google says it uses the content you submit and the responses "to provide, improve, and develop Google products and services and machine learning technologies", that human reviewers may read, annotate and process API input and output after it is disconnected from your account, and: "Do not submit sensitive, confidential, or personal information to the Unpaid Services."

Why Singapore businesses are not covered by the EU exception

Google's terms say that if you are in the European Economic Area, Switzerland or the UK, the paid-service data rules apply to all services, including free ones. Developers sometimes read blog posts written for those regions and assume free usage is safe everywhere. Singapore is not in that list, so the unpaid rules apply to a Singapore business on the free tier.

The same terms define when you are "paid": Gemini API access is a paid service only when you call it through a Google Cloud project with an active billing account. A key created in AI Studio on a project without billing is on the unpaid terms, however many tokens you use.

How free prototypes end up leaking customer data

The leak is rarely a decision. It usually happens like this: a staff member or freelancer builds a quick tool with a free key, it works, people start using it, and nobody revisits the key or the account it belongs to.

• A WhatsApp or website chatbot answers customer questions, so customers' names, phone numbers and complaints flow through the free tier.
• An email summariser reads the shared inbox, including invoices and contract terms.
• A form or OCR tool extracts details from uploaded documents such as IDs, payslips or medical memos.
• The free key sits in a freelancer's personal account, so the business can't see the settings, logs or bill.

Under the PDPA, your business remains responsible for personal data it sends to a provider, and your privacy notice should describe how it is used. The PDPC's Advisory Guidelines on Use of Personal Data in Generative AI (published 20 July 2026) treat a business that uses an AI system as a deployer with primary responsibility for PDPA compliance; our summary of the guidelines explains what that means. Sending customer messages to a service whose terms allow product improvement and human review is hard to square with that.

How to check which tier your tools are on

If anyone in the business has built something with an AI API, spend 30 minutes finding out what it runs on.

1. 1

   List every AI integration

   Ask staff and freelancers which chatbots, automations, Zapier or Make steps, plugins and scripts call an AI model. Include website chat widgets.

2. 2

   Find the key and its account

   For each one, find which provider and which account the API key belongs to. Keys in a personal account are a problem in themselves.

3. 3

   Check billing status

   For Gemini, check the key's Cloud project has an active billing account. For Mistral, check the Admin panel's 'Anonymous improvement data' toggle is off. For OpenAI, check Settings, Data controls, Sharing is disabled for the project.

4. 4

   Check what data goes in

   Look at a few real requests or logs. If they contain names, contact details, IDs, health or financial details, the tool must be on paid, no-training terms.

5. 5

   Write it down

   Record the provider, account owner, plan, data types and data-use setting in your IT asset or vendor register.

Switching a prototype to paid, no-training terms

Moving to paid terms is usually cheap for small volumes. Our LLM API pricing comparison shows per-token prices; for a small chatbot the bill is often a few dollars a month. The switch itself is mostly account housekeeping.

If you need contractual data-residency or enterprise terms, compare the direct APIs with cloud-hosted versions; our Azure OpenAI vs OpenAI API guide covers that choice. For staff using chat apps rather than APIs, see our AI training opt-out settings guide.

When a free tier is fine

Free tiers are useful for learning and testing with made-up or public data: drafting prompts, trying models, building a demo with sample questions, or summarising your own published web pages. The rule is simple: if you would be comfortable posting the input on your website, the free tier is fine. If not, use paid terms.

Keep test and production separate. Use one project or key for experiments with dummy data, and a different, paid one for anything real, so a test key never ends up in the live tool.

Do it yourself or get help?

If you built the tool yourself, the audit and the switch to a paid account take an afternoon with this guide. Get help when you don't know what your freelancer built, when keys live in someone else's account, or when the tool handles IDs, health or financial details.

Our Custom Tool or Dashboard work (from S$1,990, fixed price in writing) rebuilds AI prototypes on business-owned, paid API accounts with logging, budget limits and data-use settings documented.