Microsoft’s AI assistant — built into Windows, Word, Excel, Outlook, Teams, and Edge. Powered by OpenAI’s GPT-4o. The AI for people who live and work in Microsoft 365. History, how it works, what it does, 20 prompts, and full technical depth. Three reading levels. Official sources only.
Microsoft Copilot is an AI assistant built by Microsoft. It runs inside the tools millions of people already use every day at work — Word, Excel, Outlook, PowerPoint, Teams, and Windows itself. There is also a free version you can use at copilot.microsoft.com without any Microsoft subscription.
Unlike ChatGPT or Claude, which are separate apps you visit, Copilot is woven into the software you use for work. It is there inside your Word document when you are writing a report. It is inside Outlook when you are reading emails. It is inside Teams when you are in a meeting. You do not have to switch to a different application — the AI comes to you.
If your job involves Microsoft Word, Excel, Outlook, or Teams, Copilot is the AI that sits alongside those tools and helps you use them faster and better. It is the AI for the office.
Copilot was made by Microsoft — the company founded by Bill Gates and Paul Allen in 1975, best known for Windows and the Microsoft Office suite. Microsoft is headquartered in Redmond, Washington, and employs over 200,000 people worldwide.
The AI technology inside Copilot comes from OpenAI — the same company that makes ChatGPT. Microsoft has invested $13 billion in OpenAI and has an exclusive partnership to integrate OpenAI’s models (currently GPT-4o) into Microsoft’s products. In practical terms: when you use Copilot, you are using GPT-4o with Microsoft’s integration layer on top.
Microsoft had been investing in AI for years before ChatGPT arrived. In 2016, they launched Cortana — a voice assistant designed to compete with Apple’s Siri. Cortana was not a success. Users found it less useful than competitors, and Microsoft gradually wound it down.
More importantly, Microsoft was quietly building a relationship with OpenAI. In 2019, they invested $1 billion. When GPT-3 launched in 2020, Microsoft had early API access. They were watching — and waiting for the right moment.
When ChatGPT launched on 30 November 2022 and became a cultural phenomenon, Microsoft CEO Satya Nadella moved with unusual speed. Within weeks, plans for deep AI integration across Microsoft’s products were accelerating. In January 2023, Microsoft announced an additional $10 billion investment in OpenAI — taking their total to $13 billion.
Microsoft launched the “new Bing” in February 2023 — a search engine powered by GPT-4, before GPT-4 had even been publicly released by OpenAI. The integration allowed users to have conversations with a search engine rather than just typing keywords. The launch was dramatic, generating enormous press coverage. It also generated controversy: early users discovered the chatbot could become confrontational, express emotions, and make unsettling statements in long conversations. Microsoft quickly added safeguards.
On 16 March 2023, Microsoft announced Copilot for Microsoft 365 — AI built directly into Word, Excel, PowerPoint, Outlook, and Teams. CEO Satya Nadella called it “a new day in productivity.” The vision was striking: AI that could draft documents, analyse spreadsheets, summarise email chains, and take notes in meetings — all within the tools people already used.
After a limited enterprise preview, Microsoft 365 Copilot became generally available in November 2023 at $30 per user per month (on top of existing Microsoft 365 subscriptions). Consumer versions — Copilot in Windows, Copilot in Edge, and the free Copilot web app — were rolled out simultaneously.
The Windows integration was particularly significant: for the first time, AI was built directly into an operating system used by over a billion people. The Copilot key — a dedicated AI button — began appearing on new Windows keyboards.
In May 2024, Microsoft announced Copilot+ PCs — a new category of Windows computers with dedicated neural processing units (NPUs) capable of running AI models locally. The flagship feature was Recall — a system that continuously screenshots your screen and makes everything you’ve seen searchable. Recall generated significant privacy concerns and was delayed and made opt-in following public backlash.
Microsoft expands Copilot into “agents” — AI that can take actions across Microsoft 365 on your behalf. Copilot Studio allows organisations to build custom agents. Copilot becomes more deeply embedded in Teams, SharePoint, and Dynamics 365. The distinction between AI assistant and active participant in workflows becomes increasingly blurred.
“I had a Teams meeting about the quarterly review. I usually spend an hour afterwards writing up notes and action items. With Copilot, I just asked it to summarise the meeting — it gave me a clear summary with the key decisions, the action items, and who is responsible for each one. It took 30 seconds instead of an hour. That alone justifies the subscription for me.”
Open any Word document. Click the Copilot icon in the toolbar. Type what you need — “write a first draft of a project proposal based on these notes” or “improve this paragraph for clarity.” Copilot writes directly into your document.
Open a spreadsheet. Click the Copilot icon. Ask it to analyse your data, create a chart, write a formula, or identify trends — in plain English.
Select any email. Ask Copilot to summarise it, draft a reply, or prepare talking points for a response. Copilot reads the thread context and generates relevant, informed replies.
Copilot can join your Teams meetings (with permission) and generate real-time summaries, action items, and transcripts. After the meeting, ask it: “What were the key decisions?” or “List all action items with owners.”
GPT-4o powered. Web browsing. Image generation. No Microsoft 365 required.
Priority access, faster responses, Copilot in Word/Excel/PowerPoint/Outlook (personal Microsoft 365).
Full integration across all M365 apps. Copilot in Teams meetings. Enterprise security and compliance.
Source: microsoft.com/copilot — April 2026
Copilot’s greatest advantage is context-awareness within the Microsoft ecosystem. It can reference documents in your SharePoint, emails in your Outlook, and conversations in your Teams — enabling prompts that no external AI tool could match.
You can ask Copilot: “Summarise the emails I received from [person] about [project] this week.” or “Find the latest version of the Q3 report and summarise its key findings.” No external AI can do this — it requires access to your actual Microsoft 365 data.
Microsoft’s Copilot for Microsoft 365 is built on what Microsoft calls the Copilot Stack — a layered architecture consisting of: the foundational models (GPT-4o from OpenAI), Microsoft’s cloud infrastructure (Azure OpenAI Service), the Microsoft Graph (the API layer connecting to a user’s Microsoft 365 data), and the application layer (the Copilot UI within each M365 product).
The Microsoft Graph is the critical differentiating component. It is an API that provides unified access to data across Microsoft 365 services — emails, calendar events, files, Teams messages, SharePoint content, and user information — subject to the user’s permissions. When Copilot accesses your email or documents, it does so through the Graph, meaning it respects your organisation’s existing permission structure. If you cannot see a document, Copilot cannot access it on your behalf.
Microsoft (2023). “How Microsoft 365 Copilot works.” Microsoft Technical Blog. learn.microsoft.com — Copilot Architecture
The underlying language model capability in Copilot is delivered through Azure OpenAI Service — Microsoft’s managed deployment of OpenAI’s models (GPT-4o as of April 2026) on Azure infrastructure. This arrangement provides Microsoft with enterprise-grade SLAs, data residency controls, and compliance certifications that the standard OpenAI API does not offer. Customer data processed through Microsoft 365 Copilot is not used to train OpenAI’s foundation models, per Microsoft’s data protection commitments.
Microsoft (2024). “Data, Privacy, and Security for Microsoft 365 Copilot.” learn.microsoft.com — Copilot Privacy
Microsoft 365 Copilot uses a Semantic Index for Enterprise — a vector index built over an organisation’s Microsoft 365 content that enables semantic search (finding relevant content by meaning rather than keyword). When you ask Copilot a question that requires finding relevant documents, it queries the semantic index rather than performing a raw keyword search, producing more relevant retrieval results.
Copilot Studio (formerly Power Virtual Agents) allows organisations to build custom Copilot agents — AI assistants with specific knowledge bases, personas, and capabilities. These agents can be deployed in Teams, SharePoint, or external websites. They use retrieval-augmented generation (RAG) to ground responses in organisation-specific knowledge bases.
The Microsoft Copilot SDK and Azure AI Foundry provide developer access to build Copilot-powered applications. The Azure OpenAI API is the primary interface for accessing GPT-4o with Microsoft’s enterprise compliance wrapper.
from openai import AzureOpenAI
client = AzureOpenAI(
azure_endpoint="https://YOUR-ENDPOINT.openai.azure.com/",
api_key="your-azure-api-key",
api_version="2024-02-01"
)
response = client.chat.completions.create(
model="gpt-4o", # Your Azure deployment name
messages=[
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Summarise this document: [content]"}
]
)
print(response.choices[0].message.content)
Full Azure OpenAI documentation: learn.microsoft.com/azure/ai-services/openai