To get employees to actually use AI, start from the work they already repeat. Give them approved tools with your own documents behind them, and agree that whoever asks AI for a draft checks it before passing it on. Then train on their real work, in short sessions, and keep the examples that worked.
What the surveys show
Pew Research Center surveyed 5,273 employed US adults in October 2024. 55% said they rarely or never use AI chatbots at work. Among workers who had used chatbots for work, 40% found them extremely or very helpful for doing things more quickly, and 29% said the same about improving the quality of their work.
Microsoft and LinkedIn surveyed 31,000 knowledge workers in 31 markets in early 2024. Among the people who use AI at work, 78% bring their own AI tools, 80% at small and medium-sized companies. 52% are reluctant to admit using it for their most important tasks, and 39% have received AI training from their company.
Both surveys date from 2024, so I read them as a direction rather than a picture of your team. People use AI on their own accounts, often quietly, and mostly without training. Speed shows up before quality.
Start from the work people repeat
I would ask each person which piece of work comes back every week. Summarising a long email thread, drafting a first version of a standard letter, finding the right clause in old contracts. The person knows what a good result looks like, so checking is quick.
A goal like "use AI more" gives nobody anything to try on Monday morning. A named piece of work does. If you want help choosing, I wrote about which tasks to automate with AI first.
Approve the tools and say what data goes in
When people are unsure what is allowed, they hold back or use their own accounts. Name the tools they may use, the data that may go in and the data that stays out. My AI acceptable use policy for a small business has nine rules you can copy and adapt.
Once the approved tool is the easy choice, people can use it openly and share what works.
Agree who checks the result
One of the related searches Google shows for this question is "My co worker uses AI and I'm sick of redoing her work". It names a cost that is easy to miss: an unchecked draft passes the checking work to the next person.
I would make one rule clear: whoever asks AI for a draft owns the result. They check it against the source and sign it off before it goes to a colleague or a customer. Then count that checking time as part of the work. I use cost per accepted result for this: every attempt and every minute of review, divided by the results good enough to use.
Train on their own work
I would skip generic courses and run short sessions where each person brings a real piece of their work. Show three things on it: how to give the assistant the right document, how to ask where the answer came from, and how to check that answer against the source.
Keep the requests and examples that worked on a shared page, with the name of the person who found them. A colleague's example from your own business teaches more than any tutorial.
An assistant that answers from your company documents and shows the page it used is also easier to learn. People can see why it said what it said, and check it in seconds.
What I build for this
I build AI assistants that answer from your company documents, with the source shown next to every answer, and automations that do the work and hand the result to a person for approval. If your team already uses AI on its own and you want that on approved tools with your documents, tell me how your teams work today in a free 45-minute call.
Frequently asked questions
How do I train employees to use AI?
Short sessions on their own work. Each person brings something they repeat and leaves with a request that works on it. The shared page of working examples then becomes the starting material for new starters.
Why do employees stop using AI tools?
The reasons I would look for first: they are unsure what is allowed, nobody showed them how to use it on their own work, or the result needs checking that nobody planned for. Each one has a practical fix.
Which AI tool should staff use?
The one you approve for the data involved. For public information a mainstream chat assistant can be enough. For client files and internal documents, choose a setup where you know where the data goes, which can be an open-weight model running on your own server.