Here is how to train team on AI in 30 days. In week 1, set up one shared tool with your brand voice and rules. In week 2, give each role three use cases. In week 3, practice together weekly. In week 4, measure adoption with a short check-in. Plan on about five and a half hours per person.
Key takeaways
- Plan for more than five hours of live, hands-on training per person, spread across the month, instead of one workshop.
- Set up one shared tool with your brand voice, examples and rules in week 1, before anyone starts practicing.
- Give each role three real tasks from their own week, and teach where AI gets things wrong at the same time.
- The owner or team lead should use AI in public and share their own prompts, including the failures.
- Thirty days builds the habit. The bigger payoff comes in the weeks after, so set expectations early.
- Add a five-minute weekly check-in, because built-in analytics on small-team plans won't tell you everything about each person.
Most small teams start AI training the same way. They buy seats, run one lunch-and-learn, and hope it sticks. A month later two people use it every day and everyone else has gone back to old habits. If you have been looking for how to train team on AI without an L&D department, here is the plan I would run. It takes four weeks and has five parts: one shared setup, three use cases per role, weekly practice, and a simple way to see who is using it.
This is the version for marketing agencies and in-house teams of 3 to 25 people. It needs no enterprise budget and no slide decks that nobody opens.
Why one workshop is not enough
The best recent data on this comes from BCG's 2025 AI at Work survey. Three findings shape the whole plan:
- Hours and coaching matter. Regular AI use was much higher among employees who got at least five hours of training and had in-person training and coaching (BCG).
- The boss matters. The share of employees who feel positive about generative AI rises from 15% to 55% when leadership support is strong (BCG).
- People will use something either way. When employees don't have the AI tools they need, more than half say they will find other tools and use them anyway (BCG).
Agencies should worry most about that last point. If you don't give the team an approved tool and clear rules, client data ends up in personal accounts. That makes training partly a security decision. Our AI policy template for agencies is a good place to start the rules.
The plan below follows from these findings. It has more than five hours of structured time per person, most of it live and hands-on, and the owner or team lead goes first.
Set expectations before day one
Microsoft's research on its Copilot users found two things. Saving about 11 minutes a day is what most people need before they feel AI is useful. And gains in productivity and enjoyment build over about 11 weeks (Microsoft WorkLab).
That is a helpful frame. Thirty days builds the habit, and the bigger payoff shows up later. Tell your team that at the start so nobody calls the plan a failure in week three.
Make two more decisions before you start:
- Name one AI lead. This is usually the ops lead or the most curious person on the team. Budget two to three hours a week of their time during the month.
- The owner uses AI in public. Share your own prompts in the team channel, including the ones that went badly.
Here is the time each person commits:
| Activity | When | Time per person |
|---|---|---|
| Kickoff and shared setup | Week 1 | 90 minutes |
| Role session: pick three use cases | Week 2 | 60 minutes |
| Group practice session | Weeks 2, 3 and 4 | 45 minutes each |
| Review and next steps | Week 4 | 30 minutes |
| Weekly check-in form | Every Friday | 5 minutes |
That comes to about five and a half hours of structured time per person. Daily use inside real work comes on top of that.
The plan: how to train team on AI in four weeks
Week 1: one tool, one shared setup, one set of rules
Pick one tool for the whole team, either ChatGPT Business or Claude. Set it up in the company's name, not on personal accounts. When people use a mix of tools, it gets hard to share prompts and impossible to measure anything.
Then build the shared context everyone works from. That means your brand voice, a few strong examples, words to avoid, and client-specific rules. In ChatGPT Business, regular members can use ChatGPT fully and create GPTs (OpenAI Help Center), so a shared brand GPT is one option. I cover the ChatGPT setup and the Claude version in ChatGPT brand voice for teams.
In my experience, this shared setup does more for quality than any prompt training. When everyone starts from the same voice and examples, a junior writer's drafts start to look like a senior writer's drafts.
Finish the week with the 90-minute kickoff. Run it live, and in a room if you can. The owner demos one real task from this week's work. Then everyone does one real task of their own before they leave.
End of week 1: everyone is logged in and has used the shared setup once. Everyone also knows what data can and can't go in.
Week 2: three use cases per role
Generic prompt tips don't stick. Tasks from someone's own week do. In the 60-minute role session, each person picks three recurring tasks to try with AI. They can use the table in the next section as a menu.
Teach the limits at the same time. Harvard Business School and BCG ran a "jagged frontier" field experiment with BCG consultants. On tasks AI handled well, people using it finished 12.2% more tasks, worked 25.1% faster and produced work rated more than 40% higher in quality. On a task outside what AI could do, they were 19 percentage points less likely to get the right answer (Harvard Crimson). So AI helps a lot on some tasks and hurts on others, and the line between them is not obvious. Every role needs to know where its own line sits.
Week 3: practice on real work, together
The weekly 45-minute session is the core of the plan. Keep the format simple:
- Each person shows one output from this week and the prompt behind it.
- One person shows something that failed, and the group talks about why.
- Add the best prompts to a shared prompt library. A doc works, or you can use the shared setup itself.
- Pair a confident user with a hesitant one for the next week.
This is the coaching the BCG data points to. It costs nothing but time, and it is the part most teams skip.
Week 4: measure, then decide
Look at the numbers and the check-in answers. Both are covered below. Then make three calls:
- Which use cases are working and should become standard practice.
- Which ones to drop, with no guilt.
- Which repeated tasks are steady enough to turn into automations later. Our post on what to automate first helps with that sorting.
Starter use cases by role
Copy this table into your week 2 session. Each person picks three tasks from their row or swaps in tasks from their own week.
| Role | Good first tasks | Where to stay careful |
|---|---|---|
| Account manager | Turn meeting notes into a recap email; draft a weekly client update; prep an agenda from the last call's notes | Promises, dates and numbers sent to clients |
| Copywriter or content lead | Headline and hook variants; repurpose a blog post into social posts; edit a draft to brand voice | Factual claims; the final voice check |
| Paid media | Ad copy variants within character limits; summarize a search terms export; draft a test plan | Budget and bid decisions; any math on exported data |
| Social media | Caption drafts; monthly theme ideas; draft replies to common comments | Sensitive replies; anything posted without review |
| Analyst or reporting | Plain-English commentary on a chart; spreadsheet formulas; first draft of report notes | Every number it calculates, checked against the source |
| Ops or project manager | SOP drafts from a recorded walkthrough; brief templates; meeting summaries | Client data rules; deadlines |
| Owner or team lead | Proposal first drafts; turning notes into a plan; hiring scorecards | Pricing and people decisions |
The right-hand column matters as much as the middle one. It is where the jagged frontier shows up in marketing work.
How to measure adoption on a small team
Logins don't tell you much. You want to know two things: whether people use AI on their chosen tasks, and whether the output holds up.
What the tools show you:
- ChatGPT Business has workspace analytics. An Analytics Viewer role can see them without admin rights (OpenAI Help Center). Give that role to your AI lead.
- Claude Team lets Owners and Primary Owners see usage analytics for team activity, feature adoption and spend from the admin dashboard. Per-person analytics are only for Enterprise organizations on usage-based billing, so the toggle doesn't appear on Team plans (Claude Help Center).
On either tool, a small team can fill the gaps with a weekly check-in. It takes five minutes every Friday:
- Which of your three tasks did you use AI for this week?
- About how many minutes did it save you on a typical day?
- Which output needed heavy fixing, and why?
- Which prompt is worth adding to the library?
- Was there anything you weren't sure was allowed?
Track three things across the month:
- How many people used AI on at least two of their three tasks.
- Average minutes saved per day, compared with the 11-minute mark.
- How often outputs needed heavy rework.
Here is an illustrative example with made-up numbers. On a 10-person team, four people use AI on two of their three tasks in week 2, and eight do by week 4. Self-reported savings rise from 5 to 12 minutes a day. That team is on track. If the count stalls at four, look at who is stuck and pair them up before you buy another tool.
Mistakes that stall training
- Letting everyone pick their own tool. You lose shared prompts, shared context and any way to measure.
- Training on demos instead of the team's own work. People forget generic examples by Monday.
- The owner sits it out. The BCG leadership finding is hard to ignore.
- Skipping the limits. If nobody teaches where AI gets things wrong, someone will find out in front of a client.
- Automating too early. Build the habit first. Automations built on tasks nobody has tested by hand tend to break in quiet ways.
Doing it yourself vs getting help
You can run this plan yourself. You need three things: one person with two to three hours a week for a month, an owner willing to go first, and the discipline to keep the Friday check-in going.
Getting help makes sense when nobody has that time, or when you want the shared setup done right the first time. At Keeplight, I do this work for small marketing teams. A Brand Brain Setup builds your shared voice, examples and rules into the team's Claude or ChatGPT Business workspace, so everyone starts from the same place. If you are not sure which workflows are worth training on, the AI Workflow Audit (from $1,500) reviews every recurring workflow and names the top three to build first. You can read a sample audit report or book a free 20-minute intro call to talk it through. Either way, the plan above works if you run it.
Frequently asked questions
How long does it take to train a team on AI?
A month is enough to build the habit if you give people structured practice every week. The payoff takes longer. Microsoft's research found that gains in productivity and enjoyment build over about 11 weeks, so plan to keep the weekly check-in going after day 30.
How many hours of AI training does each person need?
Plan for more than five hours per person, spread across the month and mostly live and hands-on. BCG's 2025 survey found regular use was much higher among people with at least five hours of training plus in-person coaching. One workshop won't get you there.
Should my team use ChatGPT or Claude for AI training?
Either works. What matters is that the whole team uses the same one, set up in the company's name with shared brand context. A mix of tools makes it hard to share prompts and hard to see who is using AI.
How do I know if my team is actually using AI?
Start with the built-in analytics. ChatGPT Business has workspace analytics, and Claude Team shows team-level usage to Owners. Claude Team does not show per-person usage, so add a five-minute Friday check-in that asks which tasks each person used AI for and how much time it saved.
What if some people on my team don't want to use AI?
Pair them with a confident user and let them pick tasks they dislike doing by hand. Make it safe to show failures in the weekly session. Visible use by the owner also helps, because people follow what leaders do more than what they say.
Do we need an AI policy before we start training?
Yes, at least a short one. People need to know what client data can go into AI and what must be reviewed before it goes out. Write it in week 1 so training and rules arrive together.
Bryce Walls, Founder, Keeplight
Bryce runs marketing for a global consumer brand and builds the AI systems he uses every day. He has run paid media, analysis and marketing ops in-house for more than seven years, and Keeplight sets up the same kind of AI and automation for small marketing teams.