Business content is never training data.
Content in team and enterprise workspaces is not used to train models, unless your organisation explicitly agrees. On newly created individual accounts, de-identified content may help improve our speech models — one switch in Settings turns it off; existing accounts are opted out automatically.