AI training for employees: a 30-day plan by role, with the hours

AI training for employees works best as four one-hour sessions over 30 days, built around tasks people already do, with about 4 hours per person for most staff, 6 for managers and 12 for a few internal champions. Week one covers the rules and what data never goes into a tool. Weeks two to four are hands-on, and you measure time saved on real work, not course completions.

This guide gives you the plan, the hours by role, a worked cost example, a copyable training record and what the EU AI Act now says about staff AI literacy.

On this page

• What AI training for employees should cover
• The 30-day AI training program for employees
• How many hours of AI training does each role need?
• Worked example: the time cost for 120 people
• Generative AI training for employees: the safe-use session
• Does the law require AI training?
• How to measure whether it worked
• Training record template
• FAQ

What AI training for employees should cover

Most AI training programs for employees cover the same core: what AI tools can and cannot do, responsible use, data privacy, and how to apply the tools at work. The difference between programs that change behavior and programs that get forgotten is the ratio of practice to theory. Aim for at least two thirds of the time with people working on their own tasks.

Cover these five topics, in this order:

1. Which tools are approved. Name them. If you have a company account for ChatGPT, Copilot, Gemini or Claude, say which one and why. People default to personal accounts when nobody tells them.
2. What data never goes in. Customer personal data, health data, passwords, unreleased financials, anything under NDA. Use your own AI acceptable use policy template as the source, so training and policy say the same thing.
3. How to ask well. Give the role, the task, the context, an example of good output and the format you want. This is the most practical of all AI skills and takes one session to learn.
4. How to check the output. Facts, figures, names and anything legal get checked against a source before they leave the building. The person who sends it owns it.
5. How to make it repeatable. Save prompts that worked in a shared library so the next person starts from a tested version.

The 30-day AI training program for employees

Run one 60-minute session a week for four weeks. Keep groups to one team or function where you can, so the examples are shared.

Week 1: rules and data (60 minutes, everyone)

Walk through the approved tools, the data rules and how to check output. Show two real examples from your company of a good use and a risky one. End by asking each person to pick one task they do every week that takes at least 30 minutes.

Week 2: the first real task (60 minutes, everyone)

Each person does their chosen task with AI, live, with a facilitator in the room. Before they start, they write down how long it usually takes. Afterward they note how long it took and what they had to fix.

Week 3: save and share (60 minutes, plus 2 hours for managers)

People turn the prompt that worked into a saved template and add it to a team library. Managers get a separate 2-hour session on reviewing AI-assisted work: what to spot-check, when to ask how something was produced, and how to set team-level rules for client-facing output.

Week 4: show and log (60 minutes, plus 8 hours for champions)

Each team gives a 5-minute demo of one task that now takes less time. Everyone logs the before and after times. Your champions, one per team or roughly one per 25 people, spend 8 extra hours over the month learning to build shared assistants such as ChatGPT Projects, custom GPTs or Copilot agents for their team.

If your company has no written policy yet, write that first. Our guide on how to roll out a company AI policy people follow covers the order of steps.

How many hours of AI training does each role need?

The hours below are a planning default for a company rolling out general-purpose AI assistants. They are our recommendation, not a legal minimum. Adjust them for teams that handle sensitive data or use AI in decisions about people.

Group	Hours per person	What they learn
All staff	4	Approved tools, data rules, asking well, checking output, saving prompts
People managers	6 (4 + 2)	All of the above, plus reviewing AI-assisted work and setting team rules
AI champions (about 1 in 25)	12 (4 + 8)	All of the above, plus building and maintaining shared assistants and the prompt library
HR, legal, finance reviewers	6 (4 + 2)	All of the above, plus the extra data and decision rules for their records

For teams using AI in hiring, credit or other decisions about people, plan separate training with your legal team. Those uses carry extra rules in many places.

Worked example: the time cost for 120 people

This is an example with assumed numbers. Swap in your own headcount and hourly cost.

• 100 staff at 4 hours each = 400 hours
• 15 managers at 6 hours each = 90 hours
• 5 champions at 12 hours each = 60 hours
• Total = 550 hours, or about 4.6 hours per person on average (550 ÷ 120)

At an assumed fully loaded cost of $50 an hour, that is 550 × $50 = $27,500 of staff time. Add facilitator time: 4 sessions for each of 6 groups is 24 hours of delivery, plus preparation.

Now the break-even. If training helps each of the 120 people save 30 minutes a week on one task, that is 60 hours a week across the company. 550 hours ÷ 60 hours a week = about 9 weeks to win back the time spent. This is why the week 2 task matters: it gives you a real before-and-after number instead of a guess.

Generative AI training for employees: the safe-use session

The week 1 session is where most of your risk reduction happens. It should leave every employee able to answer four questions without looking anything up:

• Which AI tools am I allowed to use for work, and which account do I log in with?
• What must I never paste in? (Give a short list with examples from your business.)
• What do I check before AI-assisted work goes to a client or a colleague?
• Who do I tell if I think I made a mistake?

Many programs skip the last question. People hide mistakes when they think they will be punished, and hidden use is how shadow AI grows. Say clearly that reporting a mistake early is the expected behavior.

For a structure to hang this on, the NIST AI Risk Management Framework is a free, voluntary US framework that many companies use to organize AI risk work. You do not need to teach it to staff, but it helps the people who own the program.

Does the law require AI training?

In the US, there is no general federal law that requires companies to give staff AI training. Specific sectors and states may have their own rules, so check with counsel.

In the EU, the AI Act includes an AI literacy article. According to the European Commission's AI literacy Q&A, Article 4 has applied since 2 February 2025, and supervision by national market surveillance authorities starts on 2 August 2026. The same page says no certificate is needed and that organizations can keep an internal record of training.

The article was amended in 2026 by the Digital Omnibus on AI. The current text on the Commission's AI Act Service Desk says providers and deployers "shall take measures to support the development of AI literacy of their staff". The Commission's Q&A says no specific level is now mandated, but staff who oversee high-risk AI systems must still be trained for that oversight. If you operate in the EU, the 30-day plan plus the record below is a sensible base. Your lawyer should confirm what applies to you.

Many guides written in 2025 still describe the original wording. If you read elsewhere that you must reach a "sufficient level" of AI literacy, check the date.

How to measure whether it worked

Attendance tells you nothing about use. Track these at day 0, day 30 and day 90:

• Share of staff who used an approved AI tool in the last 7 days (from your admin console)
• Minutes for the week 2 task, before and after, averaged by team
• Number of saved prompts or assistants in the shared library, and how many were used this month
• Data incidents or near misses reported (a rise early on can be a good sign: people are reporting)
• Requests for new tools or use cases, which shows people are thinking about where AI fits

Report the results to whoever owns your AI governance framework so the next round of training targets the gaps.

Training record template

Keep one row per person per session. Copy this into a spreadsheet.

Name | Team | Role group (all staff / manager / champion / reviewer)
Session | Date | Facilitator | Duration (minutes)
Policy version covered | Approved tools covered (Y/N)
Data rules covered (Y/N) | Week 2 task chosen
Task time before (min) | Task time after (min)
Questions or issues raised | Follow-up owner

This record does three jobs: it shows who has been trained, it gives you the before-and-after numbers for the measurement above, and it is the kind of internal record the European Commission's Q&A describes.

AI training for employees: FAQ

What is AI training for employees?

It is a short program that teaches staff which AI tools the company allows, what data must never go into them, and how to use them on their own work. Good programs are hands-on: people bring a real weekly task and finish it with AI during the session. A slide deck about how large language models work is not enough on its own.

How long should AI training for employees take?

Plan for about 4 hours per person over four weeks for most staff: four 60-minute sessions, one a week. Managers need around 2 extra hours on reviewing AI-assisted work. The few people who build shared assistants for their teams need around 8 more. Spreading the hours out works better than one full-day workshop because people practice between sessions.

Is AI training mandatory for employees?

In the US there is no general federal rule that requires it. In the EU, Article 4 of the AI Act has applied since 2 February 2025 and asks providers and deployers of AI systems to support the AI literacy of their staff. The European Commission says no certificate is needed and an internal record of training is enough. Check with your own counsel for your situation.

What AI skills should employees learn first?

Start with three: knowing what data is off limits, writing a clear request with context and an example, and checking the output before it leaves their hands. Those three cover most of the risk and most of the early time savings. Building custom assistants, automations and agents can come later, and only for the people who will maintain them.

How do you measure whether AI training worked?

Measure use and outcomes, not attendance. Before training, ask each person to log the time one weekly task takes. Repeat the log after week 4. Also track how many people used an approved tool in the last 7 days, how many saved prompts are in the team library, and how many data incidents were reported. Compare the numbers at 30 and 90 days.

Should we buy an AI course or build our own training?

Buy the general part and build the part about your company. Off-the-shelf courses cover how AI works and general risks well. They cannot cover your approved tools, your data rules or the tasks your teams do every week. Those sessions should use your own policy and real examples from each team, which is where most of the value comes from.