Most government offices are already using artificial intelligence without ever deciding to. Over the past year or two, the everyday tools your team relies on — the scheduling software, the system that sorts job applications, the chatbot on your website — quietly added AI to the way they work. No one held a meeting about it. It simply arrived inside products you already owned.

That’s worth pausing on, because the moment a tool starts making or shaping decisions about real people, someone becomes responsible for how it behaves. If an auditor, a reporter, or a resident ever asks how your agency oversees the AI it’s using, “we didn’t realize it was in there” is not an answer that will protect you.

The good news is that you don’t need a technology background, or a data scientist on staff, to handle this well. You already have the tool that does the job: the contract. Every time your office buys something, you ask questions and set expectations before you sign — and governing AI is really just knowing which questions to add. After twenty-five years on the buying side of government, here are the five I’d ask any vendor before saying yes.

1. Where is the AI actually being used? Ask the vendor to point to the exact place their product uses AI and the specific decisions it touches. “It’s AI-powered” is marketing language; what you need to know is whether it’s recommending, sorting, scoring, or deciding — and whether a person still reviews the result before it affects someone.

2. Where did its information come from? An AI tool is only as fair as the information it learned from. Ask what that was, and whether the vendor can actually show it to you. If they can’t explain where the tool’s knowledge comes from, they can’t promise you it’s treating people fairly.

3. Can we see how it makes its choices? In government, decisions get questioned — so when one is, you’ll need to explain how it was reached. Ask whether the tool keeps records you can review, and whether the vendor can walk you through why it landed where it did. If the answer is essentially “it just works,” that’s a problem.

4. Who is responsible when it gets something wrong? Every tool makes mistakes eventually, and what matters is who’s on the hook when it does. Make sure the contract itself — not a friendly sales conversation — spells out the vendor’s duty to fix errors and to cover the cost if a mistake causes harm.

5. Does it meet the new rules everyone’s being measured against? A handful of national standards and state laws now set expectations for how government uses AI — the best known are the federal government’s AI risk framework and its 2025 guidance for agencies, plus a new industry standard written specifically for buying AI. Ask the vendor plainly whether their product meets them, and get the answer in writing.

None of these are trick questions. They’re the same basic check your office already runs on every vendor — are you reliable, can you deliver, and will working with you create a problem we’ll regret — simply pointed at a newer kind of product. The offices that come out ahead won’t be the ones buying the least AI. They’ll be the ones that ask these questions early and write the answers into the contract, so the record exists long before there’s ever an incident to explain.

Schedule a 45-minute consultation →  (ask about the AI Procurement Readiness Audit) 

FAQ 

Do I need technical training to vet an AI vendor?

No. These five questions are plain-English versions of the reliability, capacity, and risk checks your office already performs on every vendor.

What if a vendor won’t answer these questions?

That’s useful information before you spend a dollar. A vendor who can’t explain where AI is used or how it decides is telling you the tool isn’t ready to be trusted with public decisions.