In 30 seconds

When you send text to an AI product, that text goes to the vendor. Some vendors use what you send to improve their models. If that text contains customer details, contracts, or anything confidential, "used for training" can mean your private information becomes part of a system you do not control. This is the first thing to check, before volume, price, or features.

What "training on your data" actually means

An AI model learns from examples. If a vendor trains on your inputs, your data becomes one of those examples. In practice this rarely means your exact text pops out elsewhere, but it does mean:

  • Your confidential content has left your control and entered someone else's training process.
  • You cannot easily pull it back out once it is in.
  • You may be breaking a promise you made to your own customers about how their data is handled.

The distinction that matters

Most reputable business products separate two things:

  • Processing your request, which every AI must do to answer at all. Unavoidable and usually fine.
  • Retaining and training on your request, which is a choice, and one you can often turn off.

Ask specifically about the second. "We process your data securely" answers a different question than "we do not train on your data."

How to check

  • Read the data-use section of the terms, not just the marketing page.
  • Ask directly: "Do you use our inputs to train your models, and can we opt out?"
  • Look for a business or enterprise setting that disables training and shortens retention.
  • Confirm the answer covers any model provider sitting behind the vendor's product, not just the vendor.

What this does not solve

A "no training" promise is necessary, not sufficient. You still control what you send. Some data should never go to an AI regardless of the vendor's terms, and you still need to keep customer data safe in how the workflow is built.

Good enough to move on when

  • You know, in writing, whether the vendor trains on your inputs.
  • You know how long they keep your data and whether you can delete it.
  • The answer covers everyone in the chain, including the underlying model provider.