What is Microsoft Copilot?

Copilot is a brand name, not a single product

The word Copilot is an umbrella brand. Microsoft applies it to very different AI features that are partly free, partly paid, and that access different data depending on context. Saying 'Copilot' therefore requires specifying which product you mean. The common denominator is a generative AI assistant that produces text, code or summaries and is embedded inside a Microsoft surface.

• Microsoft Copilot (free): the public web and chat assistant, formerly Bing Chat, also in Windows and Edge.
• Microsoft 365 Copilot: the paid assistant in Word, Excel, PowerPoint, Outlook and Teams with access to organisational data.
• Copilot Studio: a low-code environment for building custom assistants and agents.
• GitHub Copilot: a separate assistant for software development, integrated into coding environments.

How Microsoft 365 Copilot works technically

Microsoft 365 Copilot combines three building blocks: a large language model, the Microsoft Graph and the Microsoft 365 apps. When you enter a prompt, Copilot first searches the Microsoft Graph for relevant content you are permitted to access. That context is attached to the request and passed to the language model. This technique, called grounding or retrieval augmented generation, is meant to tie answers to your real data rather than to the model's general knowledge alone.

The permission model is decisive: Copilot only surfaces content the user's account is already allowed to see. The quality and safety of answers therefore depend directly on clean access management in SharePoint, OneDrive and Teams. If files are shared too broadly, Copilot can surface that information too.

Microsoft 365 integration: what Copilot does in each app

• Word: draft, rewrite, shorten or generate documents based on existing files.
• Excel: analyse data, suggest formulas, explain trends and structure tables.
• PowerPoint: generate presentations from a document or prompt and restyle slides.
• Outlook: draft emails, summarise long threads and suggest replies in a chosen tone.
• Teams: summarise meetings live or afterwards and extract open points and action items.
• Copilot Chat: a cross-app chat interface that draws on organisational data and, optionally, the web.

Typical business use cases

Copilot's value lies mainly in recurring, text-heavy office work. Whether the licence cost pays off depends on how many employees work in Microsoft 365 daily and how well the underlying data is maintained. Realistic areas of use include:

• Automatically generating meeting notes and task lists from Teams meetings.
• Reducing inbox load by summarising long threads and drafting replies.
• Producing first drafts of reports, proposals or internal documentation faster.
• Searching large document sets and extracting key points for decision-making.
• Building internal support or HR assistants on company content with Copilot Studio.

Data protection and the Swiss perspective

For business Copilot offerings, Microsoft commits that prompts and organisational data from the Microsoft 365 tenant are not used to train the underlying language models and stay within the contractual and compliance boundary of Microsoft 365. The free public Copilot, used without a signed-in business account, is governed by different terms and should not be fed confidential company data.

In Switzerland the revised Data Protection Act (revDSG/nFADP) applies, supervised by the FDPIC (EDOEB). The term GDPR concerns the EU, not Swiss law, even though the two are similar. Microsoft operates datacentre regions in Zurich and Geneva and offers data-residency options for Microsoft 365 core services; which Copilot processing is covered should be verified against Microsoft's current commitments and your own data-processing agreement. Because Copilot builds on a general-purpose AI model, the transparency duties of the EU AI Act also apply and can reach Swiss providers with an EU nexus extraterritorially.

Limits and challenges

• Errors and fabrication: like all generative models, Copilot can produce plausible-sounding but false statements. Output must be checked.
• Data hygiene: over-broad sharing in SharePoint can lead Copilot to surface sensitive content. Permissions should be cleaned up before rollout.
• Ecosystem dependence: the full value of Microsoft 365 Copilot only appears when your data and workflows already live in Microsoft 365.
• Cost and licensing: Microsoft 365 Copilot is a paid per-user add-on; see the provider's current pricing for figures.
• Change management: without training, clear use cases and governance, adoption often stays low and the promised productivity gains fail to appear.