How to Train Employees on AI: Claude, ChatGPT, Copilot

How to train employees on AI comes down to four steps: pick the business plan your company will pay for, write a one-page usage policy, train people on their own recurring tasks and documents, then check a few weeks later whether they still use what they built. This guide for Singapore SMEs (small and medium-sized enterprises) covers what AI training for employees should include, how to set up Claude, ChatGPT, Microsoft Copilot and Gemini for a team, which plans keep your data out of model training, and how to tell whether the training worked.

Cortex Lab AI runs in-house AI training for Singapore teams, and has trained teams at multinational companies and SMEs, the largest of them more than 40 people.

Take an example: a 15-person freight forwarding company that pays for Claude Team. Its coordinators answer quote requests all day, and most replies need the same few things: the rate sheet, the surcharges and the documents each shipment type needs. Each coordinator writes those replies from scratch, so the answer a customer gets depends on who picked up the email. Training turns that into one shared set-up the whole team uses, checked against the rate sheet before anything goes out.

What should AI training for staff cover?

It should cover the tool your company pays for, used on the tasks your team does every week. One good format is a half-day hands-on workshop, then a follow-up clinic once people have used the tool at work. Agree beforehand which recurring tasks to cover and which documents your team is comfortable using in the session, so the examples on screen come from your company.

A workshop should cover:

• what your tool does well, and where it guesses
• writing instructions for a recurring task once and saving them, so anyone on the team can reuse them
• giving the tool your own documents to work from, and checking its answer against the source
• what can and cannot go into the tool under your plan and your company's rules
• which parts of a task still need a person to decide

For the freight forwarder, the first task picks itself. Quote replies go out dozens of times a week, they draw on documents the company already has, and a wrong surcharge is easy to spot against the rate sheet.

Write the AI usage policy first

Write it before the workshop, so the training teaches the rules your staff will follow afterwards. For most SMEs one page is enough if it answers five questions:

1. Which tools are approved? Name the plan and the account, for example "the company Claude Team account".
2. What never goes in? Customer and staff personal details, unless your plan and your data protection policy allow it.
3. What must a person check? Prices, names, figures and anything sent to a customer, checked against the source document.
4. What does AI never decide alone? Pricing exceptions, hiring and legal wording are common choices.
5. Who do staff ask when unsure? One named person.

Then boil it down to a line people remember, such as "customer names, numbers and messages only go into the company account".

The policy also does a legal job. The PDPA, Singapore's privacy law, expects a business to have the internal policies it needs to meet its obligations under the Act (section 12). In its summary decision on iClick Media, the PDPC found the company in breach of section 12 because it "had not developed any internal policies and practices that are necessary for it to meet its obligations under the PDPA". The PDPC directed it to put written policies in place and to train all its employees on their PDPA obligations (PDPC summary decision, case DP-1901-B3254). An AI usage policy is where those rules meet the tools your staff now use every day.

How do you train a team on Claude?

Claude training for teams works best built around Projects, on a Team or Enterprise plan so the projects can be shared. Anthropic, the company that makes Claude, says "Projects allow you to create self-contained workspaces with their own chat histories and knowledge bases". You upload the documents a task depends on and write the project instructions once, and every chat in that project starts from them. Anthropic adds that "For users on Team and Enterprise plans, projects can be shared with other members of your organization".

In the freight example, the coordinator who knows the rate sheet best builds a "quote replies" project. She uploads the rate sheet and surcharge list, and writes the instructions once: which details to ask the customer for, which documents each shipment type needs, and the tone to use. The rest of the team then tests it on five past quote requests where everyone already knows the right answer.

That test matters more than the set-up. It shows each person where Claude is reliable and where it fills a gap with a confident guess. We recommend every person builds one project for a real task and tests it the same way. We use Claude in our own work, including the chat assistant on this website.

Claude Team and Enterprise plans sit under Anthropic's Commercial Terms, its contract for business customers. For those products, Anthropic says "By default, we will not use your inputs or outputs from our commercial products" to train its models. Free, Pro and Max are personal plans, where each user chooses whether their chats are used for training. If your staff use personal Claude accounts for work, the data section below applies to you.

How does training differ for ChatGPT, Copilot and Gemini?

The method stays the same across tools: one recurring task, the right documents, and an answer checked against the source. The features you spend time on change. Train on whichever tool your company already pays for; switching tools for the sake of a workshop rarely pays off.

Microsoft says "Microsoft 365 Copilot is now named Microsoft Copilot", and that some licences and screens "might continue to reference Microsoft 365 Copilot" during the change. Microsoft also says Copilot "only surfaces organizational data to which individual users have at least view permissions". A person can only pull files they could already open into an answer, so check your folder permissions before a Copilot session.

How do you keep company data safe during training?

Run every exercise on the company's business plan, using only the documents agreed beforehand. Whether an AI company trains on what your staff type depends on the plan:

Our guide to the PDPA and ChatGPT covers the sources for each plan and what the PDPA expects of you.

How do you know the training worked?

Check a month later whether people still use what they built, on real work. Three checks are enough for most teams:

• Time one task. Time a recurring task before the workshop and again a month after. For the freight forwarder, that is one quote reply.
• Count reuse. Ask how many people use the shared project or saved instructions each week, and who has built a new one.
• Spot-check answers. Pick a few recent outputs and ask whether someone checked them against the source before they went out.

After the workshop, people on your team should be able to:

• set up a project or saved instructions for a recurring task, and share it with colleagues
• give the tool the right documents and check its answer against them before using it
• tell which tasks the tool handles well and which it gets wrong
• explain what can go into the company account and what cannot

Whether they keep using it depends on whether their set-ups save time on real work. The follow-up clinic is where you fix the ones that do not.

Can AI training be claimed under SkillsFuture or SFEC?

Yes, when an SSG-registered training provider runs a course that SSG funds. In-house training from a provider that is not registered cannot be claimed, and that includes ours. Cortex Lab AI is not an SSG-registered training provider, so its training cannot be claimed under SkillsFuture or the SFEC (SkillsFuture Enterprise Credit). Registered providers in Singapore run AI courses that SFEC can fund, and for a team that wants a funded, certified course they are the cheaper route.

In-house training suits a different need. It runs on your own documents and the plan you already pay for, the room holds only your staff, and a follow-up clinic comes once people have tried the tool at work.

If you plan to use SFEC with a registered provider, check the dates. SSG says "the current SFEC support will expire on 30 November 2026". The course run must finish on or before that date, subject to SSG's approval. A redesigned SFEC launches on 1 December 2026, and SSG says the support level and the list of eligible programmes "will be announced subsequently" (SSG Information Memorandum EED/2026/1).

What are the limits of AI training?

A half-day workshop teaches people to use a tool well. It does not make anyone a developer, and it does not connect your systems to each other. If a task needs AI to read your inbox, update your customer database or pull from your accounts, that is a build. Our guide to what AI integration involves covers it.

The tools change every few months, so some set-ups will need adjusting. Teach the method behind each set-up, and your team can adjust it themselves. When the tool does not know something, it often guesses instead of saying so, which is why every answer still needs a person to check it before it goes to a customer. In the freight example, a project that has never seen a new surcharge will quote the old one until someone uploads the new rate sheet. We design our training to take repetition out of the work a team already does, and none of it is planned around reducing headcount.

Is AI training the right first step for your team?

Often, but not always. If nobody in the company has a business AI plan yet, choose the tool first. Training on personal accounts builds habits you will have to undo. If your staff want a recognised certificate or a government-funded course, start with a registered provider (see the funding section). If one manual process takes up most of the hours, automating it may return more than training everyone, and our guide on how to choose an AI consultant helps you work out which. For a wider view, read AI for Singapore SMEs: where to start.

Frequently asked questions

Do staff need a technical background for AI training?

No. Good sessions are in plain English, and the exercises are the emails, proposals and reports your team already writes.

Which AI tool should we train staff on?

The one your company already pays for on a business plan, whether that is Claude, ChatGPT, Microsoft Copilot or Gemini in Google Workspace. If nobody has a business plan yet, choose one before you train.

Can staff use real company documents in AI training?

Yes, on your business plan and after agreeing which documents to use. Keep customer and staff personal details out unless your plan and your usage policy allow them.

Is AI training SFEC claimable?

Only when an SSG-registered provider runs an eligible course. Current SFEC support expires on 30 November 2026, and courses must finish by then. Cortex Lab AI is not an SSG-registered training provider, so its in-house training cannot be claimed under SkillsFuture or the SFEC.

To plan in-house training around your team's own tasks, see AI training for teams.

Sources checked on 18 October 2026: SSG Information Memorandum EED/2026/1; PDPC summary decision, iClick Media; Anthropic, What are projects?; Anthropic Privacy Center, Is my data used for model training?; Anthropic, Updates to our consumer terms; Microsoft Learn, Data, Privacy, and Security for Microsoft Copilot; OpenAI Help Centre, keeping history on but disabling model training; OpenAI Help Centre, ChatGPT agent; Google Workspace, Generative AI in Google Workspace Privacy Hub.