Questions to ask any AI vendor

The nine

1. Which model provider sits behind this feature, and is that in writing?
2. Is our content used to train or improve any model, theirs or a third party's?
3. What's retained, for how long, and what does deletion actually delete?
4. Where is it processed and stored, and which jurisdictions can compel access?
5. Who can review our content, under what circumstances, and is it logged?
6. Can we turn the feature off per user, and does that stop processing or just hide it?
7. What happens to our data if we leave, or if you're acquired?
8. What does the contract say about breach notification, and on what timeline?
9. Has this been assessed against the Australian Privacy Principles, and can we see it?

How to use them

Send them before the demo. A vendor whose answers are specific, written and quick has done the work. A vendor whose answers arrive as reassurance rather than detail has told you something useful too.

Get the list

There is a printable version in resources, free, no email required.

Beyond the two we teach

Every month brings another AI product aimed at your industry, usually a wrapper around one of the big models with your sector's vocabulary on the front. Some are genuinely useful. The questions that separate them:

• Whose model is underneath, and does the answer change? If a vendor won't say, your data answer is unknowable.
• What happens to our content — the same five questions as any tier, applied to the wrapper and to the model underneath.
• Where does it sit when you stop paying? Setups, templates and history — exportable, or stranded?
• What does it do that a general tool plus a good setup doesn't? Sometimes a real answer. Often the answer is a nicer interface for something you could build in an afternoon.
• Who is accountable for a wrong output in a regulated context? Read that clause before the pricing.