Client files and AI tools

Some clients will allow a named tool. Some will forbid anything that trains on their material or stores it outside a region. Ask in the kickoff. Write the answer on the account.

A ban means a ban, including the "quick" side tool on a freelancer's laptop. Your vendors are your problem.

A draft tool inside the client workspace, with access limited to that client, is easier to explain than a personal account with a pile of brands in one history.

Still read the retention terms. "Inside the product" does not automatically mean "private enough for this contract." If you have not published the exact terms, do not guess in a security review. Say you will confirm.

Campaigns under embargo, financials, and personal stories need the strictest setting you have, or no model at all. The embarrassment of a leak is larger than the embarrassment of a slower caption.

Watermark nothing you would not want repeated. Assume prompts are readable by someone else someday.

If a client asks whether a model touched the draft, say yes and say what a human changed. Hiding it turns a process question into a trust problem.

You do not need to narrate the tool in every caption. You do need to be able to answer.