Can I use my users' data for something new?
What decides whether data you already collected can power a new AI feature, benchmarks, partnerships, or marketing, including your original promises, FTC limits on retroactive policy changes, and state purpose-limitation rules.
Short answer
Often, if the new use fits what users were told and would reasonably expect when they gave you the data. When it doesn't, the safest path is usually to tell users before the change and, for material changes, get their consent, applying the new use only to data collected afterward or from users who agree. Quietly updating the privacy policy to cover a new use of data you already hold is the approach regulators object to most. Deciding this before you build lets you design the feature around the data you can use, instead of reworking it later.
What decides it
What you already promised
Start with what users were told when the data was collected: the privacy policy in effect at the time, consent screens, app store privacy labels, enterprise contracts, and marketing. If a new use contradicts those promises, changing the policy today doesn't change what users agreed to then.
The FTC on changing your terms
In February 2024, the FTC warned AI companies and others that adopting more permissive data practices, such as sharing data with third parties or using it for AI training, and telling users only through a quiet, retroactive change to the terms of service or privacy policy could be unfair or deceptive. It pointed to earlier cases in which companies changed privacy policies to share data without notifying users or getting their consent. The practical takeaway is to give real notice before a material change and to get consent before applying it to data you already hold.
State purpose-limitation rules
- California. The CCPA regulations require that collecting and using personal information be consistent with the reasonable expectations of the consumers involved, or compatible with the purpose it was collected for. A use outside that generally needs the consumer's consent. What your privacy policy says helps set expectations but doesn't decide them on its own.
- Other states. Several state privacy laws, including Virginia's and Colorado's, require consent before using personal data for purposes that aren't reasonably necessary to, or compatible with, the purposes you disclosed.
- Sensitive data. Health, biometric, precise location, and children's data usually require opt-in consent for new uses. Maryland's law, in effect since October 2025, goes further: sensitive data may be processed only when strictly necessary to provide what the user asked for, and it can't be sold. See Does Washington's My Health My Data Act apply to my app?
Common AI cases
- A new AI feature on existing data. Summarizing a user's own documents or history for that same user usually fits the original purpose. Using one user's data to serve other users, or sending it to a new AI provider, may not. See Can I send user data to an LLM provider?
- Training and fine-tuning. This is the use regulators watch most closely. See Can I train or fine-tune a model on my users' data?
- Benchmarks and aggregate insights. Properly de-identified or aggregated data is generally treated differently from personal data, but the standard is demanding. De-identified data has to stay that way, with commitments not to re-identify it and contracts that pass those commitments on.
- Sharing with partners. Giving data to a partner, or letting it improve the partner's product, can count as a sale or sharing under state laws and needs its own disclosures and opt-outs.
- Marketing and advertising. Using product data, especially conversation content, to target ads or marketing is a new purpose for most AI products, and it often triggers state opt-out rights.
- Enterprise customer data. Business customers' contracts usually limit use to providing the service. A new use generally needs the customer's agreement, not just a policy update.
How to make the change well
- Tell users before the change takes effect, in the product or by email, not only in a revised policy.
- Get consent for material changes, with a clear choice, not a pre-checked box.
- Apply the new use going forward, or only to data from users who agree.
- Use less data. Choose the minimum needed, and de-identify or aggregate it where you can.
- Keep records of each policy version, when it took effect, and who consented to what.
- Update every disclosure: the privacy policy, app store privacy labels, and data processing agreements.
Checklist an agent can run against the codebase
- Find where the new feature or pipeline reads existing user data, and list the data fields it uses.
- Check when that data was collected, and which privacy policy version was in effect at the time.
- Find consent records and flags, and check whether the new use is gated on them.
- Check whether the new use sends data to a new vendor, AI provider, or partner.
- Find sensitive data in the fields used: health, biometrics, precise location, children's data, and financial data.
- Check whether data said to be de-identified or aggregated still carries user IDs or other identifiers.
- Check whether enterprise customers' data is excluded, or gated on their agreement.
- Compare the new use with the current privacy policy and app store privacy labels.
Sources
If this applies to you
An AI feature review covers a new use of existing data before you build it: what your earlier promises allow, the notice and consent the change needs, and updated privacy documents. For one planned use, such as a new feature built on existing data, a pre-build feature check covers that decision before it ships.