Here is something that surprises a lot of people: Microsoft paid billions of dollars to partner with OpenAI, the company behind ChatGPT, and then built its own competing assistant on top of that same technology. So when someone asks “does Copilot use ChatGPT,” the honest answer sits somewhere between yes and no. Copilot does not run ChatGPT the app, but it absolutely runs on the same family of large language models that make ChatGPT work. Understanding that difference changes how you use both tools.
This matters more than a trivia question. If you pay for Copilot at work, subscribe to ChatGPT Plus at home, or try to decide which one your team should adopt, you need to know what is actually happening under the hood. Are you paying twice for the same brain? Is one smarter than the other? Does your company data end up in OpenAI’s hands? In this guide, you will learn exactly which AI models power each version of Copilot, how Microsoft’s partnership with OpenAI works, how Copilot processes your requests step by step, where the two products genuinely differ, and what Microsoft’s growing in-house model effort means for the future.
The Short Answer: Copilot Runs on the Same Engines That Power ChatGPT
Microsoft Copilot does not use ChatGPT as a product, but it does use OpenAI’s GPT large language models, which are the exact same underlying engines that power ChatGPT. Think of it like two different cars built by two different companies that both use engines from the same manufacturer. The engine is shared. The steering wheel, dashboard, safety features, and driving experience are not.
ChatGPT is OpenAI’s own consumer-facing application. You visit chatgpt.com or open OpenAI’s app, and you talk directly to OpenAI’s servers. Copilot is Microsoft’s application. When you type into Copilot, your request travels to Microsoft’s Azure cloud, where Microsoft hosts licensed copies of OpenAI’s models through a service called Azure OpenAI. Microsoft then wraps those models in its own layers of search, security, data connections, and formatting rules before handing you an answer.
So if a coworker tells you “Copilot is just ChatGPT with a Microsoft logo,” they are partly right and mostly wrong. The core language ability comes from the same source. Nearly everything else, including what data the assistant can see, how it cites sources, and where your conversations get stored, works differently.
Here is a quick way to keep the relationship straight:
- OpenAI builds the models, including GPT-4o, GPT-4.1, and the o-series reasoning models.
- ChatGPT is OpenAI’s own interface for talking to those models.
- Microsoft licenses those models and hosts them on Azure.
- Copilot is Microsoft’s interface, tuned for search, Office documents, code, and enterprise data.
How the Microsoft and OpenAI Partnership Actually Works
Microsoft did not stumble into this arrangement. The company started investing in OpenAI back in 2019 with an initial commitment of around one billion dollars. In 2023, Microsoft expanded that relationship with a multiyear investment widely reported at roughly ten billion dollars, and by 2025 the total commitment climbed past thirteen billion. In exchange, Microsoft became OpenAI’s primary cloud partner and gained the right to build OpenAI’s models into its own products.
That deal created something called the Azure OpenAI Service. Instead of Microsoft sending your Copilot prompt to OpenAI’s servers, Microsoft runs its own copies of OpenAI’s models inside Microsoft data centers. Your prompt never leaves the Microsoft cloud boundary. This distinction is the single biggest reason large companies feel comfortable turning Copilot on for thousands of employees while blocking public ChatGPT.
A Rough Timeline of the Relationship
- 2019 — Microsoft invests one billion dollars and becomes OpenAI’s exclusive cloud provider.
- 2021 — GitHub Copilot launches using OpenAI Codex, a model trained on public code and derived from GPT-3.
- 2023 — Microsoft launches Bing Chat, powered by GPT-4 before most people had access to it anywhere else.
- 2023 — Microsoft 365 Copilot arrives, combining GPT-4 with the Microsoft Graph, which holds your emails, files, and calendar.
- 2023 to 2024 — Bing Chat gets rebranded as Copilot and expands into Windows, Edge, and mobile apps.
- 2024 to 2025 — Copilot gains access to newer OpenAI models, including GPT-4o and reasoning models, while Microsoft begins mixing in its own in-house models.
One important nuance: the partnership has never been exclusive in the direction most people assume. Microsoft can build products with other models, and it has started doing exactly that. GitHub Copilot now lets developers pick models from Anthropic and Google. Microsoft also builds its own smaller models, called the Phi family, and larger in-house models under its MAI research group. OpenAI remains the backbone, but it is no longer the only supplier.
What Happens Behind the Scenes When You Type Into Copilot
People often picture Copilot as a simple pipe: you type, GPT answers. The reality involves several extra steps that shape the final response in meaningful ways. Microsoft calls the coordinating layer an orchestrator. In the early Bing Chat days, the company named it Prometheus, a system that decided when to search the web and how to blend search results with model output.
Walk through a typical request and you can see how much work happens between your keystroke and the answer on screen:
- You submit a prompt. Copilot receives your text, images, or files.
- The orchestrator interprets intent. It decides whether the question needs a live web search, a look at your company files, an image generation call, or nothing but the model’s own knowledge.
- Grounding data gets collected. For consumer Copilot, this usually means Bing search results. For Microsoft 365 Copilot, it means pulling relevant emails, documents, and chats you already have permission to see.
- The prompt gets rebuilt. Microsoft assembles a much longer prompt that includes your question, the retrieved data, safety instructions, and formatting rules. This is often called prompt engineering or retrieval augmented generation.
- The model generates a response. The GPT model hosted on Azure produces the draft answer.
- Post-processing and safety checks run. Microsoft filters the output, adds citations and links, checks for harmful content, and formats the result.
- You see the answer. Often with footnotes, follow-up suggestions, and clickable sources.
That grounding step explains why Copilot and ChatGPT can give you noticeably different answers to the same question even when both use a similar model. Copilot leans hard on Bing search results, which makes it stronger at current events and citations. It also means a bad search result can pull Copilot toward a weaker answer, while ChatGPT might rely more on the model’s internal knowledge.
Here is a practical example. Ask both tools, “What did the Federal Reserve decide at its most recent meeting?” Copilot will almost certainly run a Bing search, pull the top few news articles, and answer with numbered links you can click. ChatGPT will also search the web in most modes, but it uses its own search infrastructure and often summarizes with a different citation style. Same engine family, different research assistants feeding it.
Not All Copilots Are the Same: A Guide to the Product Family
Microsoft attached the Copilot name to a lot of products, which creates real confusion. Somebody asking whether Copilot uses ChatGPT might mean the free chatbot in Windows, the paid Office assistant, the coding tool inside Visual Studio Code, or a security analyst tool. Each one uses AI models differently.
Consumer and Business Chat Assistants
Microsoft Copilot, the free assistant available on the web, in Windows, in Edge, and on mobile, uses OpenAI models combined with Bing grounding. Copilot Pro adds priority access to the newest models during busy hours plus integration into Office apps for individuals. Microsoft 365 Copilot, sold to businesses, adds access to your organization’s data through Microsoft Graph and honors your existing file permissions.
Developer Tools
GitHub Copilot started life on OpenAI Codex, a GPT-3 descendant trained heavily on public code repositories. It later moved to GPT-3.5 and GPT-4 class models. Today GitHub Copilot offers a model picker, so developers can switch between OpenAI models, Anthropic’s Claude models, and Google’s Gemini models depending on the task. That flexibility makes GitHub Copilot the clearest case where the answer to “does Copilot use ChatGPT” becomes “sometimes, and sometimes it uses a competitor instead.”
| Copilot Product | Primary Model Source | Special Data It Can Access | Typical User |
|---|---|---|---|
| Microsoft Copilot (free) | OpenAI GPT models on Azure | Bing web search | General consumers |
| Copilot Pro | OpenAI GPT models, priority access | Bing plus personal Office files | Individuals and freelancers |
| Microsoft 365 Copilot | OpenAI GPT models on Azure | Microsoft Graph: mail, files, chats, calendar | Business employees |
| GitHub Copilot | OpenAI, Anthropic, and Google models | Your open repository and code context | Software developers |
| Copilot Studio | Azure OpenAI, configurable | Custom knowledge sources you connect | Builders and IT teams |
| Security Copilot | OpenAI GPT models plus security-specific models | Threat intelligence and security signals | Security analysts |
Because Microsoft updates these products constantly, the specific model version behind any given Copilot can change without an announcement. Microsoft rarely publishes a permanent “Copilot runs on model X” statement, since the company swaps models as newer, faster, or cheaper options arrive.
Copilot vs ChatGPT: The Differences That Actually Change Your Work
Since both tools share model DNA, the meaningful comparisons come from everything wrapped around the model. Below is a side-by-side look at the areas most people care about.
| Feature | Microsoft Copilot | ChatGPT |
|---|---|---|
| Owner | Microsoft | OpenAI |
| Core models | OpenAI models on Azure, plus Microsoft in-house models | OpenAI models only |
| Web grounding | Bing search, deeply integrated | OpenAI’s own search integration |
| Office integration | Native in Word, Excel, PowerPoint, Outlook, Teams | File uploads and connectors, no native embedding |
| Enterprise data access | Microsoft Graph with permission awareness | Connectors and custom GPTs |
| Free tier | Generous, includes image generation | Generous, includes limited advanced model use |
| Custom assistants | Copilot agents and Copilot Studio | Custom GPTs and the GPT Store |
| Developer API | Through Azure OpenAI Service | Through OpenAI API |
Where Copilot Tends to Win
Copilot shines when your work already lives in Microsoft’s ecosystem. Ask it to summarize the last twelve emails from a client, build a PowerPoint from a Word doc, or find the budget figure buried in a SharePoint spreadsheet, and it can do that because it sees your tenant data. ChatGPT cannot reach into your corporate mailbox without extra setup. Copilot also gives cleaner citations for news-style questions because Bing results feed directly into the answer.
Where ChatGPT Tends to Win
ChatGPT usually gets new OpenAI features first, since OpenAI ships to its own product before partners integrate them. Power users often report longer, more flexible conversations, better handling of very long creative tasks, and a richer ecosystem of custom GPTs. Developers who want direct API control without Azure’s enterprise setup also lean toward OpenAI.
A realistic scenario helps here. Imagine a marketing manager writing a quarterly report. She uses Microsoft 365 Copilot to pull actual sales numbers from her company’s Excel files and draft the first version inside Word, because only Copilot can see those files. Then she pastes a section into ChatGPT to brainstorm five punchier headline options, because she likes its creative range. Neither tool replaces the other. They cover different jobs.
Common Myths People Believe About Copilot and ChatGPT
Misunderstandings about this relationship spread fast, partly because Microsoft’s messaging shifted several times as products got renamed. Let’s clear up the biggest ones.
- Myth: Copilot is a rebranded version of the ChatGPT app. False. Copilot is Microsoft software with its own interface, safety layers, and data connections. Only the language models are shared.
- Myth: Your Copilot prompts go to OpenAI. Microsoft states that Microsoft 365 Copilot prompts and responses stay within the Microsoft service boundary and are not sent to OpenAI or used to train OpenAI’s public models.
- Myth: Copilot always uses the newest OpenAI model. Microsoft mixes model versions based on cost, speed, and task. A simple summary might route to a smaller, faster model, while a complex reasoning task gets a bigger one.
- Myth: Copilot and ChatGPT should always give identical answers. Different grounding data, different system prompts, and different safety filters produce different results regularly.
- Myth: Microsoft owns OpenAI. Microsoft holds a significant investment and commercial rights, but OpenAI operates as an independent company with its own leadership and governance structure.
- Myth: GitHub Copilot only writes code with GPT. Developers can select from several model providers depending on the task and their plan.
One more misconception deserves attention. Some people assume that because Copilot uses OpenAI technology, paying for both Copilot and ChatGPT wastes money. In practice, plenty of professionals subscribe to both, because the value comes from the integration and workflow, not from raw model access. If your job involves heavy Excel and Outlook work, Copilot earns its keep. If you spend your day on open-ended writing, research, and experimentation, ChatGPT might serve you better.
Privacy, Data Handling, and Who Actually Sees Your Prompts
For business users, this section usually matters more than model specs. The core question is simple: when I type something sensitive, where does it go and who can read it?
With Microsoft 365 Copilot, Microsoft applies what it calls enterprise data protection. Your prompts, the data Copilot retrieves, and the generated responses stay inside your Microsoft 365 tenant boundary. Microsoft says this content is not used to train the foundation models, and OpenAI does not receive it. Copilot also respects existing permissions, so it cannot surface a document you were never allowed to open in the first place.
Consumer Copilot works differently. Free personal accounts have looser settings, and Microsoft may use some interaction data to improve services depending on your region and your privacy dashboard settings. Microsoft offers commercial data protection for users signed in with a work or school account, which prevents chat data from being saved or used for training.
ChatGPT has its own tiers. On free and Plus plans, OpenAI may use conversations to improve models unless you turn that setting off. On Team, Enterprise, and API plans, OpenAI states that business data is not used for training by default. So the privacy gap between the two products narrows considerably once you compare paid business tiers rather than free consumer versions.
Here are practical habits worth adopting no matter which tool you pick:
- Sign in with your work account when handling company information.
- Check your organization’s approved AI tools list before pasting internal data anywhere.
- Turn off model training in your account settings if the option exists.
- Avoid entering customer records, passwords, or regulated data unless your legal team has cleared the tool.
- Review generated content before sending it, since both tools can state wrong facts confidently.
Choosing Between Them: Practical Tips for Real Situations
Rather than declaring a winner, match the tool to the task. The decision usually comes down to where your data lives and what kind of output you need.
Pick Copilot When
Your files, emails, and meetings run through Microsoft 365. You need answers grounded in current web results with clickable citations. You want AI directly inside Word, Excel, Outlook, or Teams instead of copying text back and forth. Your IT department requires enterprise controls, audit logs, and compliance coverage under existing Microsoft agreements.
Pick ChatGPT When
You do heavy creative or analytical work that benefits from long, flexible conversations. You want early access to new OpenAI features like advanced voice, deeper research modes, or custom GPTs. You build applications and prefer OpenAI’s API directly. You work outside the Microsoft ecosystem, on Google Workspace or a Mac-heavy team.
Use Both When
Many professionals land here. Use Copilot as your workplace assistant that knows your calendar and documents. Use ChatGPT as your thinking partner for drafting, brainstorming, and exploring ideas. The combined monthly cost often runs less than a couple of hours of billable time, which makes the math easy for knowledge workers.
A few tips that improve results in either tool: give context before asking, since both perform far better when you explain your role, audience, and goal. Ask for a specific format such as a table, a five-bullet summary, or a three-paragraph email. Iterate instead of restarting, because follow-up refinements usually beat rewriting your prompt from scratch. And always verify numbers, dates, names, and quotes against a primary source.
What Is Changing: Microsoft’s In-House Models and the Road Ahead
The tidy answer that “Copilot equals GPT” gets blurrier every year. Microsoft has invested heavily in building its own models, and those models increasingly handle parts of the Copilot experience.
The company’s Phi family of small language models runs efficiently on modest hardware, which makes them useful for on-device features in Windows and for quick tasks that do not need a giant model. Microsoft’s MAI research group, led by executives recruited from other AI labs, works on larger foundation models designed to reduce dependence on any single supplier. Microsoft has publicly described a future where Copilot routes requests across a mix of models, choosing the cheapest one capable of doing the job well.
Model routing is the key concept to watch. Instead of sending every question to one expensive model, Copilot increasingly decides in real time which model fits. A simple grammar fix might go to a small local model. A complex financial analysis might go to a large reasoning model. You will not see this happening, but it explains why response quality and speed can vary between sessions.
Several trends look likely over the next few years:
- More model diversity. Expect Copilot products to support several providers, following the path GitHub Copilot already took.
- More agents, fewer chats. Microsoft is pushing toward agents that complete multi-step tasks on their own rather than just answering questions.
- More on-device processing. Copilot+ PCs with neural processing units handle some AI work locally, improving speed and privacy.
- Deeper business data integration. Connectors into CRM systems, databases, and third-party apps keep expanding.
- Less emphasis on model names. As routing takes over, the question of which model answered you becomes harder to pin down and less relevant to most users.
None of this means OpenAI disappears from the picture. The partnership was extended, and OpenAI models remain central to the most demanding Copilot workloads. But the era of a one-to-one link between Copilot and GPT is fading.
Frequently Asked Questions About Copilot and ChatGPT
Beginners tend to ask the same handful of questions once they learn about the connection. Here are direct answers.
Is Copilot free if ChatGPT is free?
Both offer free tiers. Microsoft Copilot is free on the web, in Windows, and on mobile, including image generation. ChatGPT is free with limits on advanced models. Paid tiers for both typically run around twenty to thirty dollars per user per month, with Microsoft 365 Copilot priced for business licenses.
Which one is more accurate?
Neither wins across the board. Copilot often performs better on questions about recent events because of Bing grounding and visible citations. ChatGPT frequently performs better on long reasoning chains and creative work. Both can produce confident errors, so verification remains your job.
Can Copilot read my company documents?
Microsoft 365 Copilot can, but only files you already have permission to open. The free consumer Copilot cannot access your organization’s internal data at all.
Does GitHub Copilot use ChatGPT?
GitHub Copilot originally used OpenAI Codex and later GPT-class models, so it shares roots with ChatGPT. Today it also supports models from Anthropic and Google, and you can choose which one handles your request.
Will Copilot stop using OpenAI models?
Unlikely in the near term. Microsoft extended its OpenAI partnership and continues to ship OpenAI models through Azure. What is changing is the addition of Microsoft’s own models alongside them rather than a full replacement.
Do I need both?
Only you can answer that. Try each free tier on your actual daily tasks for a week. Most people discover a clear preference based on where their work already lives.
So, does Copilot use ChatGPT? Copilot does not run the ChatGPT app, but it runs on OpenAI’s GPT models hosted inside Microsoft’s Azure cloud, which makes the two products close cousins rather than twins. Microsoft wraps those shared engines in its own orchestration layer, grounds answers in Bing search, connects them to your Office files through Microsoft Graph, and enforces its own privacy and safety rules. That wrapper, not the model, creates most of the difference you feel when using each tool.
Knowing this saves you from two costly mistakes: assuming the tools are interchangeable, and assuming they are completely unrelated. Pick Copilot when your work lives inside Microsoft apps and you need answers grounded in your own data. Pick ChatGPT when you want flexible reasoning, early features, and open-ended creative help. And as Microsoft blends its own Phi and MAI models into the mix, expect the line between the two to keep shifting. The smartest move is to stay curious, test both against your real tasks, and let your results, not the branding, decide which assistant earns a permanent spot in your workflow.
