Free questionnaire
AI vendor privacy questionnaire
Twelve questions to ask an AI vendor about data privacy before you sign. Type the vendor's answer under each one, mark it good, weak or open, and print the sheet for your procurement file.
The conversation
AI vendor privacy questionnaire
12 questions
01Which companies touch our data between the input and the answer?
A good answer: A list naming every company, what it does and where it runs.
02In which country is the data processed and stored, and does the contract say so?
A good answer: The region for storage and for the model calls, and the DPA clause that fixes it.
03Do our inputs and outputs train or improve any model, yours or your model provider's?
A good answer: No, neither the vendor nor the model provider, and both contracts say so.
04Does a person at your company or at the model provider ever read our prompts and outputs?
A good answer: Who can read what, for how long, and whether you can switch it off.
05How long do you keep prompts, outputs, logs and backups, each counted separately?
A good answer: A table with a period for each location and the job that deletes it.
06When a customer asks us to delete their data, how do we do that in your system?
A good answer: Deletion by customer ID that covers logs and any search index, with a confirmation.
07Which of your employees can open our data, and is each access recorded?
A good answer: Named roles, access tied to a support case you opened, and the access log on request.
08If there is a breach, how fast do we hear about it, and from whom?
A good answer: A number of hours and a named channel.
09Can we have the current sub-processor list, and how do we learn about a new one?
A good answer: Name, function and location of each company, the model provider included, notice of changes some weeks ahead, and a stated consequence if you object.
10For anything processed outside the EU, which transfer mechanism applies?
A good answer: The mechanism for each sub-processor outside the EU, with standard contractual clauses in place as a fallback.
11Which actions can the agent take without a person approving them, and where is that limit enforced?
A good answer: A list of the agent's actions, write permissions narrower than read permissions, and anything external going past a person at first.
12When we leave, what do we get back, in which format, and when is the rest deleted?
A good answer: An export in an open format, a deletion date that includes backups, and a written confirmation.
Where the questions come from
Each question and its good-answer line come from the note AI vendor data privacy: 12 questions, which also explains why each question matters and what a weak answer sounds like.
This sheet is a working aid for your conversation with the vendor, not legal advice. Your data protection officer or lawyer decides what your contract needs.
Prompt Consulting builds AI agents too, so we are one of the vendors these questions are for. To put them to us, book a thirty-minute scoping call.