Guides
How should AI assistant training be planned for teams working with ChatGPT, Claude, and Copilot?
Short answer
For teams using multiple AI assistants, training should be planned based on the task at hand rather than the specific tool. A common foundation covering prompt writing, output verification, and data rules—applicable across all tools—is provided first, followed by short modules specific to the environment each tool operates in. A separate program focused on coding assistants is planned for software teams.
Key concepts
Microsoft 365 Copilot: An assistant that works within Word, Excel, PowerPoint, Outlook, and Teams, using corporate data that the user has access to.
GitHub Copilot: A coding assistant that offers code suggestions, conversational support about code, and multi-step code changes within the development environment (IDE). It is not the same product as Microsoft 365 Copilot.
Should training be planned by tool or by task?
Training should be planned by task, because writing effective prompts, verifying output, and protecting sensitive data are the same skills regardless of the tool. Tool interfaces change frequently; training that only teaches menus becomes outdated within a few months. An efficient structure consists of a common core that applies across all tools, combined with a short tool-specific module tailored to the tool the team actually uses.
What is the training difference between these assistants?
The most important difference lies in where the tools operate and what data they can access; the focus of training changes accordingly.
| Criteria | ChatGPT / Claude | Microsoft 365 Copilot | GitHub Copilot and coding assistants |
|---|---|---|---|
| Where it operates | Separate chat interface and applications | Within office applications | Development environment and terminal |
| Typical use | Drafting, summarizing, working with uploaded documents | Email, meeting notes, document and spreadsheet tasks | Writing code, testing, refactoring |
| Training focus | Prompt structure, providing context | Retrieving accurate content from corporate documents | Code review, security of generated code |
| Point of caution | Which data can be uploaded | Correct configuration of access permissions | Protection of secret keys |
Since Microsoft 365 Copilot uses documents the user has access to, files that were mistakenly shared may become visible through the assistant. File sharing permissions should be reviewed before training.
How should training be structured for teams using multiple assistants?
Training should be built around a common prompt template and a short corporate guide that answers the question "which task belongs to which tool." This guide, for example, directs office document tasks to Microsoft 365 Copilot and long document analysis to the organization's approved chat assistant, and shows in a single table which data classification can be entered into which tool.
Academy Club's Generative AI Tools and Effective Usage Training targets this common core by covering tools such as ChatGPT, Claude, and Gemini together. For managers at organizations within the Microsoft ecosystem, Microsoft AI Strategies for Creating Business Value is a relevant topic.
How does assistant training differ for software teams?
Training for software teams focuses on integrating the coding assistant into testing, code review, and documentation steps, and on maintaining the quality of generated code. For example, the training Academy Club delivered for Softtech engineers covered running large language models from the command line and integrating them into CI/CD processes (news). For teams using Claude, AI-Assisted Software Development with Claude is a topic geared toward this focus.
How is AI assistant training planned step by step?
The plan starts with a tool inventory and ends with measurement.
- Create a tool inventory. Which team uses which assistant, and under which license?
- Prepare policies and permissions. Determine which data can be entered into which tool, assign licenses, and review sharing permissions.
- Deliver the common core. Provide all users with prompt structure, verification, and data rules.
- Add role and tool modules. Hold separate hands-on sessions for business units, managers, and software teams.
- Measure usage. Compare usage data from before and after training using the tools' usage dashboards.
Observations from the field: Where do organizations struggle with assistant training?
Organizations most often struggle because they design training as a walkthrough of a tool's menus. The observations below are based on recurring situations Academy Club has encountered in corporate technical training sessions.
Employees use the tool they're used to, not the approved one. Even when an organization has purchased a license, employees may continue using a different tool through their personal accounts. Clearly explaining in training why the approved tool was chosen and the data risks of personal accounts helps change this habit.
In-office assistants expose permission issues. When an embedded assistant goes live, folders that have been broadly shared for years surface in unexpected answers. If permission adjustments aren't made before training, trust issues arise.
[EDITOR: A 1-2 sentence comment from an Academy Club instructor on designing training for teams using multiple AI assistants should be added to this section.]
Conclusion
For teams using tools like ChatGPT, Claude, and Copilot together, training should be planned around a common core, short tool- and role-specific modules, and a predefined usage policy. If you'd like to design a customized program for your organization based on the tools your teams use, you can fill out Academy Club's training request form.
Frequently Asked Questions
Should an organization standardize on a single assistant?
Standardizing on a single tool simplifies management and training, but it may not suit every task. A common approach is to use one primary assistant for general purposes and a separate coding assistant for software teams.
Can employees use these tools through their personal accounts?
This is not recommended if corporate data will be involved. Personal accounts are not covered by the organization's data agreement or oversight.
These tools update frequently—how often should training be repeated?
Refreshing the common core once a year is sufficient. Tool-specific modules can be updated through short sessions whenever major feature changes occur.
Who is responsible for incorrect information provided by the assistant?
The employee who uses the output and the manager who approves it are responsible; the assistant is merely a draft generator. This rule should be one of the first topics covered in the common core.
Should an organization build its own assistant or use ready-made tools?
Ready-made tools are usually sufficient for general correspondence and document tasks. If the organization needs an assistant connected to its own internal knowledge sources, a development project and technical team training will be required.