Module 2Academy

Working with ChatGPT, Copilot and Claude

Prompts that work, your own assistants built on your own documents, and safe use in the business version.

Working with ChatGPT, Copilot and Claude is a training day for teams that already use the tools but still ask only loose questions. The team learns a fixed prompt structure, builds its own assistant on its own documents, and sets up use so that data stays safe. Each participant leaves with a usable work instruction.

After this module the team can
  • 01Each participant writes prompts with role, context, task, format and check, and tests them on their own work
  • 02The team builds one assistant of its own, as a GPT, a project or an agent, with its own documents and instructions
  • 03The team sets up the business account so it knows what is and is not kept or used
  • 04The team has a one-page work instruction that everyone can use
Agenda
  1. 01

    The prompt structure

    Role, context, task, format and check. We practise the five parts on real tasks and compare the results side by side.

  2. 02

    Your own assistants

    GPTs in ChatGPT, projects in Claude and agents in Copilot. What they do, what they do not do, and which documents may go in.

  3. 03

    Safe use

    The difference between a personal and a business account. What the vendor says about training on your data, and which agreements you must record.

  4. 04

    The work instruction

    The team writes one page together: what we do with AI, what we do not, and how we check. That page stays useful after the day.

From loose questions to a way of working

At most teams that use ChatGPT, Copilot or Claude, the tool sits in the small things: rewriting an email, making a summary, explaining a term. That delivers something, but the gap between a handy question and a way of working is large. A way of working decides what the tool does, with which material, who checks and what happens next.

This training makes the step from loose questions to a way of working. We start with the prompt, because it is the smallest piece everyone uses daily. Then we build the assistant and record the agreements that decide whether it stays safe.

The prompt structure, and why it works

A good prompt has five parts. The role tells the model from which point of view to write. The context supplies the material and background. The task says exactly what must happen. The format sets the shape of the answer: a table, three points, an email of at most 120 words. The check asks the model to mark uncertain points and to say which information is missing.

In the exercise we compare the same task with a short and a full prompt. The difference in result is usually large, and the team sees that it can steer the tool instead of waiting for a better model. The check is the part that is most often skipped, and so it needs the most practice.

Your own assistants with your own documents

Each tool offers a way to build an assistant that follows the same instruction every time and uses your own documents. In ChatGPT they are called GPTs, in Claude projects, and in Copilot agents. The details differ per vendor and per subscription, and change regularly. The training shows where the controls sit on the day, and what the assistant can and cannot do.

For a team the question is not which tool is best, but which documents may go in. An assistant that knows the quote templates and the price list is useful. An assistant with access to the full client file needs a different judgement. We build one assistant with a limited set of documents and show the team how to keep that boundary.

Safe use in the business version

A personal account and a business account are not the same. The business account usually comes with an agreement on whether the vendor uses the data to improve the model, and a data processing agreement. Those agreements differ per vendor. Anthropic states that Claude Team does not train on your content by default; Google states that for Workspace with Gemini, business data is not used outside your organisation to train models. For other vendors you must read the terms yourself, and we do that together in the training.

The team then sets the controls to match the agreements, records who the administrator is and who may create new assistants, and decides which data may never go into a prompt. That is the least spectacular part of the day and the part most often missing.

The work instruction as the result

The day ends with one page the team uses: which tasks we do with the tool, which prompts we use as standard, which documents the assistant uses, and which check we run after each answer. That page is the only thing that remains after the training, and that is deliberate. A team that still works from the page a week later has taken the training; a team that never opens it has attended a course.

Frequently asked questions

  • Which tool does the training teach?

    The tool your team uses or chooses. We work with ChatGPT, Copilot and Claude, and the principles hold for all three. We do not pick one for you; the comparison on /compare/ is meant for that.

  • Do we get a subscription included?

    No. You arrange the business subscription yourself, with the vendor or a reseller. We do help with the settings and the agreements that come with it.

  • Is this the same as a prompt course?

    Prompting is one of five parts. The training also covers the assistant, the settings and the work instruction, because that is what the team still uses after the day.

Which team needs this module?

For up to 12 participants €3,950 excl. VAT. In the conversation we say whether the module fits, or whether a build day or the adoption programme is better.