Assistants have become quietly load-bearing at work: draft contracts, patient notes, salary discussions, half-finished strategy documents. Most of that text leaves the device. Understanding where it goes takes ten minutes with the provider's own documentation, and the answer differs by account type more than by brand.
Four separate questions
- Retention: how long is the conversation stored after you delete it from the interface?
- Human review: can staff or contractors read samples for safety, abuse or quality work?
- Training: is your content used to improve future models, and can that be switched off?
- Sub-processors: which other companies handle the data, and in which jurisdictions?
Consumer and business defaults differ
Paid business and enterprise tiers commonly exclude customer content from model training by default, while free consumer tiers commonly do not. If colleagues use personal accounts for work, the organisation's policy is whatever those free tiers say.
Practical rules that survive product changes
- Never paste credentials, keys or identifiable personal data of a third party into a general assistant.
- Check the training toggle again after any major product update; defaults get reset.
- For regulated work, insist on a data-processing agreement naming retention periods and sub-processors.
- Assume deletion in the interface removes it from your view, not necessarily from backups, until the documentation says otherwise.
This article names no provider and rates none. Policies change frequently enough that the only reliable answer is the current documentation for the specific account you use.
