AI training is easy to buy and hard to judge. Two proposals can both promise "hands-on AI training for your team" and deliver completely different days: one a tour of features on a projector, the other three hours of people doing their own work with a tool and learning when not to trust it. This checklist is meant for the person who has to choose: an owner, an HR lead, an operations manager. Use it to compare proposals, including ours.
We wrote it from the curriculum we deliver, so it reflects one firm's view of what works. Where an item comes from outside evidence, we say so.
What is AI training for employees?
AI training for employees is instruction that teaches staff to use generative AI tools such as ChatGPT, Microsoft 365 Copilot, Gemini or Claude on their real work, safely and with judgment. It differs from technical AI education, which teaches people to build models, and from vendor product demos, which show features rather than build habits.
The need is visible in most offices. Microsoft's 2024 Work Trend Index, a survey of 31,000 knowledge workers, found that 75% were using AI at work but only 39% of those users had received AI training from their company, and 78% were bringing their own AI tools. In other words, many staff are already using these tools, often on personal accounts, without anyone having shown them the rules.
What should every AI training program cover?
Every AI training program should cover seven foundations and then apply them to the team's own work. The checklist below is what we look for in any proposal. The "red flag" column is what tends to appear when a provider is stretching a generic deck.
| Topic | What good looks like | Red flag |
|---|---|---|
| How generative AI works | Tokens, context windows and why the same question gets different answers, explained in a few minutes with examples | An hour of AI history, or no explanation at all |
| Prompting | A repeatable structure (role, task, context, format, examples) practiced on real tasks | A list of "magic prompts" to copy |
| Checking output | Concrete checks against the source; practice catching invented facts | "Always double-check" said once |
| Data privacy rules | What may and may not go into which tool, tied to your accounts and policy | Generic warnings with no link to your plans |
| Working from your own material | Uploading or connecting your documents, notes and spreadsheets, with practice data where needed | Exercises only on made-up trivia |
| AI as a thinking partner | Using AI to challenge a plan, compare options or prepare for a difficult conversation | Treated only as a writing tool |
| Reusable workflows | Turning a prompt that worked into saved instructions the team can reuse | Every task starts from a blank chat |
| Labs on your processes | Guided exercises built around your roles and recurring work, using practice or fictional data | The same labs every client gets |
| Multi-step and scheduled tasks | Chaining steps, and setting recurring work to run on a schedule where your tool supports it | Single questions only |
| Your own accounts | Training runs on the tool and plan the company pays for | A trainer-owned demo account |
| A written policy | An acceptable-use policy, or at least the rules in it, covered in the session | Rules left for "later" |
| Measurement and follow-up | A before-and-after measure and a check-in weeks later | Only a smile sheet |
| Certificate of completion | Issued to every attendee, naming the session, date and hours | Not offered |
Why does checking AI output need its own module?
Because generative AI states wrong things with confidence, and the people most likely to miss it are the people who have only seen it work. Checking takes practice, so it needs time in the agenda.
The U.S. National Institute of Standards and Technology, in its Generative AI Profile (NIST AI 600-1), names this risk "confabulation": the production of confidently stated but erroneous or false content, which is what people usually call hallucinations. Good training gives people specific checks: do the names, numbers and dates match the source; does every quoted figure exist; did the tool add anything that was not in the material; would you sign this. Our prompt engineering basics for employees include a checking routine you can use on its own.
Should AI training use our own AI tools and accounts?
Yes. Training should run on the tools and plans your company already pays for, signed in with work accounts. People learn faster on the screen they will see on Monday, and it puts work on accounts where the business terms apply.
This also settles a privacy question. Business plans from OpenAI, Microsoft, Google and Anthropic state that customer data is not used to train their models by default, while some personal plans share data unless the user opts out. If you have not chosen a tool yet, our ChatGPT vs Copilot vs Gemini vs Claude comparison sets out pricing and data controls for a 20-person company.
How much of AI training should be hands-on?
At least half, and ideally about two thirds. People should spend most of the session typing into their own accounts on tasks they brought, with the trainer moving around the room.
A useful test for any proposal: ask for the agenda in minutes and add up the blocks where attendees are working rather than watching. In our 2–3 hour workshops, the explanation blocks are short and each one is followed by a "your turn" exercise on the person's own material. For larger rooms, ask how many instructors or helpers will be present; a single presenter cannot help 25 people who are stuck at the same moment.
Should AI training be tailored by role and industry?
The foundations should be the same for everyone; the applications should be tailored. A paralegal, a dispatcher and a controller need the same understanding of prompting and privacy, but they need to practice on different work.
We split this into a standard Day 1 for everyone and a Day 2 built around the client's roles and industry, with three labs on their own processes using practice or fictional data. Some clients take our standard program with examples matched to their industry; others want fully custom labs built after interviewing their team leads. How much tailoring you want is the main thing that changes the price, so it is worth deciding consciously. Industry-specific programs, such as AI training for law firms, show what tailoring looks like in practice.
What questions should you ask an AI training provider?
Ask questions that force specifics. A strong provider will answer all of these without hesitation:
- Which tools and plans will the training run on, and does every attendee use their own account?
- Can I see the agenda in minutes, and how much of it is hands-on?
- How do you collect our real tasks before the session, and what data do you use in exercises?
- What will attendees be told about what never goes into AI, and how does that match our policy?
- How do you teach people to check output, and do they practice catching errors?
- What do people leave with: saved instructions, a prompt library, a policy draft, a certificate?
- How do you measure whether it worked, and when do you follow up?
- Who exactly will deliver it, and how many helpers will be in the room?
- What is the scope and price in writing, and what changes it?
- Can you supply the paperwork for Illinois ETIP reimbursement, including certificates of completion?
Be wary of answers that promise outcomes ("your team will save ten hours a week"). Nobody can promise that before seeing your work, and how to measure results honestly is covered in our guide to measuring AI training ROI.
What should AI training leave out?
A good program is defined as much by what it skips as by what it covers. Leave out:
- Long tool tours. Ten minutes of features nobody will use is ten minutes not spent practicing.
- Real customer or patient data in exercises. Use practice data, fictional records or redacted samples.
- Hype and fear. Neither "AI will replace you" nor "AI does everything" helps someone write a better client email.
- Promised time savings. They cannot be known in advance.
- Building custom software in a general workshop. That is a separate project with a separate scope.
Does the law require AI training for employees?
Illinois does not require AI training as such, but Illinois law does set rules that training should cover. The European Union does require it for companies within the scope of its AI Act.
Since January 1, 2026, the Illinois Human Rights Act, as amended by HB 3773 (Public Act 103-0804), makes it a civil rights violation for an employer to use AI in recruitment, hiring, promotion, discipline and similar decisions in a way that discriminates against protected classes, or to fail to notify employees when AI is used for those purposes. Anyone who might use AI on résumés or performance reviews should hear that in training. Our Illinois AI employment notice kit and the AI acceptable use policy template cover the paperwork.
Chicago companies with operations in the EU should know that Article 4 of the EU AI Act, in force since February 2, 2025, requires providers and deployers of AI systems to take measures to ensure, as far as possible, a sufficient level of AI literacy among their staff. This is general information, not legal advice; confirm with counsel.
How long should AI training be, and what does it cost in Chicago?
For a first session, 2–3 hours per group of up to about 15 people covers the essentials. A full Day 1 builds the foundations properly, and a second, tailored day is where people build workflows on their own processes.
At 312 IT Consulting, a 2–3 hour workshop for up to about 15 people typically starts from about $2,500. One-day and two-day programs are quoted in writing after a free first call, with a written scope and quote before work starts. Our guide to AI training costs in Chicago compares formats, and Illinois employers may be able to recover up to 50% of eligible costs through ETIP with pre-approval; see funding AI training in Illinois. We train on-site across Chicago and the suburbs and live online for teams anywhere, and every attendee receives a certificate of completion.