Every public agency in the country is being pitched AI training right now, and a large share of what is being sold is a recorded webinar with a certificate attached. Buying badly here is expensive twice — once for the money, and again for the false confidence that staff are now trained.
Decide what you are buying before you take a call
There are three different products under one name, and mixing them up is the root of most bad purchases:
| Product | What it delivers | Who needs it |
|---|---|---|
| Awareness training | A shared baseline: what the tools do, what may not go in, how to verify output, who to ask | All staff who use or may use AI |
| Role training | Applied skill for a specific function — analysts, communications, HR, procurement | Teams whose work changes materially |
| Governance training | Policy, acceptable use, records, procurement review, incident handling | The small group accountable for how the agency uses AI |
Most agencies need all three, in that order, and most vendors will happily sell one as though it were all of them.
The questions that separate real programs from decks
- Who wrote the curriculum, and who is in the room delivering it? Not the company — the person. Ask for the instructor's qualifications, not the founder's.
- What will staff be able to do afterward that they cannot do now? If the answer is a list of topics rather than behaviors, the program has not been designed for outcomes.
- May we see one complete module? Not a sales deck. An actual module. This single request eliminates a large share of providers.
- How is completion evidenced? Attendance record, assessment, or a verifiable credential — they are not the same thing, and you should know which one you are getting before you report it upward.
- What do we keep? Materials, job aids, and a policy template you own after the engagement, or nothing?
- How is it updated? AI content decays quickly. A program with no revision schedule is a snapshot of last year.
- What happens in the thirty days after? Training with no follow-through is a day out of the office. Ask what the provider expects supervisors to do.
Procurement details that matter here specifically
- Records and public disclosure. Training content, attendance, and any prompts staff produce during exercises may be subject to your retention and disclosure rules. Decide before, not after a request arrives.
- Data handling in the training itself. If exercises use live agency data, that is a data-handling decision requiring the same review as any other vendor access.
- Accessibility. Public agencies have obligations here. Ask for captioning, accessible materials, and a screen-reader-compatible platform in writing.
- Who may buy from whom. Cooperative contracts, interlocal agreements, and state vehicles change what your options are and how fast you can move. Confirm the mechanism before you shortlist.
- Verifiability of any credential. If the deliverable includes a credential, confirm it carries a number and can be verified publicly. A certificate that cannot be checked is a PDF.
Sequence the money
The pattern that works in agencies with limited budget: awareness training for everyone first, because it prevents the incidents; governance training for the small accountable group second, because it produces the policy; role-specific training last, once you know which functions actually changed. Reversing that order is common and it produces deep training for a handful of people inside an organization with no rules.
How to evaluate fairly
Send every provider the same written questions and require written answers. Score curriculum, instructor qualifications, evidence of completion, and post-training support separately from price. Ask for one reference at an organization of your size and call it — not the flagship client, one that looks like you.
Then ask the finalist to deliver a single module to a small group before you commit to the full rollout. Providers who are confident say yes.
Questions people ask
What should AI training for government staff actually cover?
At minimum: what the technology does and does not do, acceptable use of public data, records and disclosure implications, verification of output, and what to do when something goes wrong. Tool demonstrations are the least durable part of any curriculum.
How long should the training be?
Long enough to change behavior and short enough that staff complete it. For general awareness, a focused hour that people finish beats a half-day that they skip.
Should we train everyone or only certain staff?
Everyone who touches the tools needs awareness-level training. Deeper training belongs with the people who will govern use, procure systems, or make decisions affected by AI output.
How do we evaluate providers fairly?
Ask every provider the same written questions, require a sample module rather than a sales deck, and score curriculum, instructor qualifications, and post-training support separately from price.