How do I get my team to use AI well instead of ignoring it or misusing it?
Short answer: Train people on their own real tasks, not on AI in general. In four weeks and a few short sessions you set one data rule, give each person one workflow to try with an approved AI tool such as ChatGPT (OpenAI's chat assistant) or Claude (Anthropic's), and teach a review habit so mistakes get caught before they reach a customer. A generic course teaches features. This teaches judgment on work your team already does.
You want your staff using AI on real work, safely, without paying for a course nobody finishes.
I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.
If you want to know which of your team's tasks are worth training on first, the AI Strategy Sprint picks them with you in 3 days.
The plan below needs no special software and about three hours of each person's time over four weeks. It works because people learn a tool by using it on something they already know how to do. They can tell when the answer is wrong.
First, a fact worth knowing. Microsoft and LinkedIn's 2024 Work Trend Index reported that 75% of knowledge workers already use AI at work, and that 78% of AI users bring their own tools to work. At small and medium companies it was 80%. That is a survey of knowledge workers by Microsoft, which sells AI products, so read it as a signal, not a precise count. But it fits what I see: your team is probably using AI already. The question is whether they do it well and with care for your data. If you suspect they are using it without telling you, read my guide to what to do when staff use ChatGPT without asking first.
What good AI training actually teaches
Three skills, in this order:
- What goes in. Which information may be pasted into an AI tool and which may not.
- How to ask. Giving the tool the context, the format and an example, so the answer is usable.
- How to check. Treating every output as a draft from a fast, confident junior who sometimes makes things up.
In my experience a course tends to spend its time on the second one and skip the first and third. The first is where your risk sits. The third is where your quality sits.
Before you start: three decisions
1. Pick one approved tool
Choose one or two paid business plans, not a free personal account. OpenAI states that by default it does not train its models on business data from ChatGPT Business, ChatGPT Enterprise or its API, unless you opt in. That is a different default from personal accounts, so the plan matters. My guide to AI tools that do not train on your data shows what to check for any tool, and the ChatGPT for business guide covers the plans.
2. Write the data rule on one page
Keep it short enough that people remember it. A simple version:
- Never paste: passwords, card or bank numbers, health details, anything under a client confidentiality clause, and personal data about customers or staff.
- Paste with care: your own drafts, internal notes and anonymised examples, in the approved tool only.
- Fine to use: public information and your own marketing copy.
- For every account: turn on multifactor authentication, a second check such as a code on a phone, as the US Cybersecurity and Infrastructure Security Agency (CISA) advises, especially for email, social media and financial accounts.
The US Federal Trade Commission's guide for business says to collect only the personal data you need, keep it safe and dispose of it securely. The same logic applies here: the safest customer data in an AI tool is none. Ask a lawyer if you handle regulated data.
3. Choose one task per person
Pick a task that is frequent, low risk and annoying. Examples: summarising a long email thread, drafting a first reply to a common customer question, turning meeting notes into action items, cleaning up a rough job description. Avoid anything where a wrong answer costs real money. My guide to business tasks you should not automate with AI explains where to draw that line.
The four-week plan
| Week | Session | Time | Output |
|---|---|---|---|
| 1 | Rules and a live demo on a real task | 45 min, whole team | Everyone has an approved account and has read the one-page data rule |
| 2 | Each person runs their one task three times | 20 min a day, 3 days | A saved prompt that works and a note of what went wrong |
| 3 | The error-spotting session | 45 min, whole team | A shared list of mistakes the AI made and how they were caught |
| 4 | Share and decide | 30 min, whole team | Keep, change or drop each workflow; pick next tasks |
Week 1: rules and a live demo
Show one real task from your own business, start to finish, on screen. Use a task someone dreads. Then walk through the data rule with two examples of what would and would not be allowed. End by making sure every person can sign in to the approved tool. Do not teach ten features.
Week 2: one task, three times
Each person runs their chosen task through the tool three times on real work, then saves the prompt that gave the best result. A useful prompt has four parts: who the AI should act as, the context, the exact format you want, and one good example. Ask them to write down what they had to fix each time. That note is the real training material.
Week 3: the error-spotting session
This is the session most courses skip, and it is the most useful. Before the meeting, take three real AI outputs from your team's work and plant one error in each, such as a wrong date, an invented figure or a policy that does not exist. Hand them out and give people ten minutes to find the errors. Then discuss what gave each one away.
The point is to build a habit: check names, numbers, dates and quotes against the source, every time. AI tools can state false things in a confident tone. A person must own the final answer. My guide to a human-in-the-loop AI workflow shows how to build that check into a process.
Week 4: share and decide
Each person shows their task in two minutes: the prompt, the time saved, the errors found. For each one, decide: keep, change or drop. Pick the next task for each person. Keep the saved prompts in one shared document so the next hire starts with them.
A worked example, with made-up figures
These numbers are invented to show the method.
A 10-person landscaping company has an office manager who writes quote follow-up emails. She spends about 5 minutes on each and sends 12 a week, so about an hour a week. After week 2 she has a saved prompt that drafts the email from the quote details. Drafting plus checking takes 2 minutes. That saves about 36 minutes a week, roughly 30 hours a year. She also caught that the tool once invented a discount, which is why the check step stays.
Notice what is small here: one task, one person, one saved prompt. That is the training working. The rest of the team copies the pattern for their own tasks.
How to check it worked
- Can each person name the data they must never paste?
- Does each person have a saved prompt for one real task?
- Can each person describe one mistake the AI made on their work?
- Has anyone used a personal account for work in the last month?
- Is there a named person who answers "can I put this into the AI tool?"
If the answers are yes, yes, yes, no and yes, the training did its job. Also look at your software spending: auditing your AI tool subscriptions shows who is paying for what and removes duplicates.
When to stop training and build
If three people are doing the same task with the same prompt, stop training them and build the workflow once. Copying a prompt into a chat window forty times a week is a sign that the task should run on its own, with a person approving the result. The AI workflow automation guide and the AI implementation guide cover how.
Training also has limits. It cannot fix a messy process, and it cannot make a risky task safe. If a task involves contracts, money moves or hiring decisions, the answer is usually a stricter review step, not more prompt skill.
Common mistakes
- Training everyone on everything. Pick one task per person.
- Skipping the data rule. This is the one that can cost you a client.
- A one-off workshop. Practice on real work sticks better than one long session. The short weekly structure matters more than the length of any session.
- Banning it. People use it anyway and stop telling you.
- No owner. Name one person who keeps the shared prompt document current.
The US National Institute of Standards and Technology (NIST) publishes a voluntary AI Risk Management Framework. You do not need it for a small team, but the idea of naming an owner and checking results is the same.
How I'd approach this
I would run the four weeks with the team's real work and keep it short. The first session sets the rule, the third session teaches the check habit, and everything else is practice. I would not buy a course.
I would also watch for the moment a task repeats across people. That is where a built workflow beats more training, with a person approving the result. If you want help picking the first tasks, the AI Strategy Sprint is a paid 3-day first step where we find where AI could create real value in your business and write down the exact steps, and it begins with a free 15-minute AI brainstorm call.
Questions people ask
How long does it take to train employees to use AI?
About four weeks of short sessions, roughly three hours of each person's time in total, if you focus on one real task each. Basic comfort comes quickly. Good judgment about checking the output comes from repeated use on real work.
Do I need an outside trainer or a course?
Not for a small team. Someone who has used the tool for a few weeks can lead the sessions. Consider outside help when your team handles regulated data or you want a workflow built, not just taught. The employee training plan template helps you keep records either way.
What if some staff refuse to use AI?
Start with the task they dislike most and show a result in ten minutes. Do not force it on tasks they do well. Some roles gain little, and that is fine. The rule everyone must follow is the data rule, whether they use AI or not.
Is it safe to let staff paste customer information into ChatGPT?
Not by default. Use a business plan, set a written data rule, and keep personal customer data out unless you have checked the plan's terms. OpenAI says it does not train on business data from its business plans by default. Read the data guide and ask a lawyer about regulated data.
How do I measure whether the training worked?
Count three things: tasks with a saved prompt, minutes saved per week on each, and mistakes caught before they reached a customer. Review them in the week 4 session. If time saved is small and mistakes are many, change the task, not the person.
Should I write an AI policy?
Yes, but keep it to one page: approved tools, the data rule, who to ask, and the rule that a person checks anything sent to a customer. Longer policies go unread.
More in this topic: Problem solving: AI strategy and adoption (24 guides) →