Do AI APIs Train on Your Data? OpenAI, Anthropic and Gemini Compared
OpenAI and Anthropic say they do not train on API data by default, while Google's Gemini API does use unpaid-tier content and does not use paid-tier prompts to improve products. Retention windows differ too, so the answer depends on the vendor, the tier and the setting.
Key takeaways
- OpenAI has not trained on API data by default since 1 March 2023 unless you opt in.
- Anthropic does not train on commercial product data by default, including the API.
- Gemini uses unpaid-tier content to improve products, but paid-tier prompts are not used to improve products.
- Retention is around 30 days at OpenAI and Anthropic, while Gemini logs default to up to 55 days.
- Zero data retention exists at OpenAI and Anthropic, but it needs approval or a contract.
On this page 9
The short answer: OpenAI and Anthropic do not train on paid API data by default, and Google does not train on paid Gemini API prompts either. The free tiers are different: Google's unpaid Gemini usage is used to improve products, and human reviewers may read it (Gemini API terms). If a customer asks whether their data trains a model, the answer starts with the vendor and tier.
TL;DR
- OpenAI has not trained on API data by default since 1 March 2023, unless you opt in.
- Anthropic does not train on commercial product data by default, including the API.
- Gemini uses unpaid-tier content to improve products, but paid-tier prompts are not used to improve products.
- Retention is around 30 days at OpenAI and Anthropic, while Gemini logs default to up to 55 days.
- Zero data retention exists, but it needs approval or a contract, not a checkbox.
Disclosure: ZeroRaised publishes this analysis and is a launch site for bootstrapped products. Facts come from the linked sources as checked on 31 Oct 2026; ZeroRaised has no commercial relationship with the companies named.
Zero data retention (ZDR) is an arrangement in which a vendor processes a request without storing the prompt or response afterwards, beyond what is needed to return the answer.
Does OpenAI train on API data?#
No, not by default. Since 1 March 2023, data sent to the OpenAI API is not used to train or improve OpenAI models unless you explicitly opt in (OpenAI data controls). That is a strong default, and it has been stable for years.
Storage is a separate question. Abuse-monitoring logs may include prompts, responses and derived metadata, and they are generated for all API usage and retained for up to 30 days by default, unless the law or service protection requires longer (OpenAI data controls). Eligible customers can get Zero Data Retention after approval (OpenAI data controls).
Does Anthropic train on API data?#
By default, Anthropic does not use inputs or outputs from commercial products, including Claude for Work, the Anthropic API and Claude Gov, to train models (Anthropic Privacy Center). It may if you explicitly report feedback, such as a rating, or otherwise opt in (Anthropic Privacy Center). Consumer plans on Free, Pro and Max are covered by separate terms (Anthropic Privacy Center).
Deletion has a clear default: inputs and outputs are automatically deleted on the backend within 30 days (Anthropic Privacy Center). The exceptions are features with longer retention under your control, such as the Files API, a zero data retention agreement, Usage Policy enforcement, or legal requirements (Anthropic Privacy Center).
Does Google train on Gemini API data?#
It depends entirely on the tier. Under the Unpaid Services terms, which cover Google AI Studio and the unpaid quota of the Gemini API, Google uses submitted content and responses to provide, improve and develop products and machine learning technologies, human reviewers may read them, and Google says not to submit sensitive, confidential or personal information (Gemini API terms).
Under Paid Services, meaning the Gemini API through a Cloud project with active billing, Google does not use prompts or responses to improve products and logs them for a limited period solely to detect Prohibited Use Policy violations and for legal disclosures (Gemini API terms).
One regional note: in the EEA, Switzerland and the UK, the Paid Services terms apply to all Services, including free AI Studio and the unpaid quota (Gemini API terms).
Why is retention different from training?#
Founders often collapse two questions into one: training changes the model, retention keeps a copy. A provider can answer no to the first and still hold logs for a month.
Gemini's paid logging is the clearest example. New paid projects store Generate Content API requests by default, and existing projects keep their current setting (Gemini API changelog). Logs are retained for a default maximum of 55 days, adjustable to 7, 14, 28 or 55 (Gemini API docs). Those logs are not used for product improvement by default, but that changes if you share datasets with Google (Gemini API docs).
So the Gemini paid tier is not a training risk by default, but it is a retention story to configure. Teams that need full control instead choose open-weight models. The breakdown of Gemini's default logging covers the settings in detail.
What is zero data retention, and who offers it?#
ZDR means the provider does not keep the prompt or response after serving the request. OpenAI offers it to eligible customers after approval (OpenAI data controls). Anthropic offers it through a zero data retention agreement, a listed exception to its 30 day deletion default (Anthropic Privacy Center).
Google's Contributor Tier runs the other way. It is an invite-only preview offering up to 50% off token costs on select models in exchange for allowing Google to use data from those models under the Unpaid Services terms, and Google says not to submit sensitive or personal information (Gemini API docs). That is data sharing as a discount, the opposite of ZDR.
A checklist for your privacy policy#
Verify each answer against the console and the contract.
- Name the vendor and the tier. "We use a leading AI provider" is not a disclosure. "We use the paid Gemini API" is.
- State the training position. Say whether prompts train or improve models, and name the opt-ins you have enabled.
- State the retention window. Give the number, including log retention and any datasets you build.
- Describe ZDR if you have it. Name the arrangement and which routes it covers.
- Disclose feedback loops. If ratings feed training, say so.
- Review when you change tiers. A free prototype and a paid product are different deals.
The LLM API comparison guide helps you weigh these terms alongside price. If cost is the pressure, the LLM cost guide shows where savings come from.
What this means for bootstrapped founders#
- Do not use a free tier for customer data. It is where the training answer flips against you.
- Separate prototype and production projects. It keeps free-tier experimentation away from paid-tier customer content.
- Write the data sentence before the feature ships. It forces the tier decision early, when it is cheap.
- Check the logging default, not just the training policy. Retention can contradict a promise even when training does not.
Where ZeroRaised fits#
ZeroRaised is a launch site for products built without institutional venture capital or accelerator equity. If you have a product people can try, you can submit your product. The pricing page explains the current options, and there are no promises of traffic or results.
If you want more analysis like this, join the ZeroRaised newsletter. Data terms change quietly; read them before you rewrite a policy.
Sources#
Frequently asked questions
Does OpenAI train on my API data?
No, not by default. Since 1 March 2023, data sent to the OpenAI API is not used to train or improve models unless you explicitly opt in. Abuse-monitoring logs are still generated for all API usage and retained for up to 30 days by default.
Does Anthropic train on API data?
By default, Anthropic does not use inputs or outputs from commercial products, including the Anthropic API, to train models. It may if you explicitly report feedback or opt in. Inputs and outputs are automatically deleted on the backend within 30 days, with some exceptions.
Does Google train on Gemini API data?
It depends on the tier. Unpaid Services, which covers Google AI Studio and the unpaid quota, use submitted content to improve and develop products, and human reviewers may read it. Paid Services, through a Cloud project with active billing, do not use prompts or responses to improve products.
What is zero data retention?
Zero data retention is an arrangement where the vendor processes a request without keeping the prompt or response afterwards. OpenAI offers it to eligible customers after approval. Anthropic offers it through a zero data retention agreement.
Does a paid tier mean my data is never stored?
No. Paid tiers usually mean your data is not used to train or improve models, but providers still keep logs for abuse monitoring, legal compliance or service operation. Gemini's paid Generate Content API also logs requests by default for new projects.