Yes and no: Microsoft Copilot can use OpenAI GPT models, but Copilot is not ChatGPT, and the exact answer changes by product surface and license. Copilot Chat (Basic) is generally web-grounded unless licensed work-data access or supplied content changes it.
That distinction matters because two people can open “Copilot” and get different grounding, different permissions, and different model behavior. A consumer Copilot chat, a Microsoft 365 Copilot chat in Word or Teams, and a Frontier feature inside Copilot Chat are not the same experience.
Requirements for Identifying Your Copilot's AI
What you'll need
- Your exact Copilot surface: Copilot app, Copilot Chat, or Microsoft 365 Copilot inside an app.
- Your license state: free/basic access, Microsoft 365 Copilot, or another licensed Copilot plan.
- Your tenant or app settings, especially if Frontier features or model selection are enabled.
- A way to check whether the reply is grounded in web content, work content, or both.
Which model is powering my Copilot?
Start with the surface you are using, then check whether you are licensed for Microsoft 365 Copilot and whether model choice is turned on. The reply may come from a GPT-based model, but the grounding layer and permissions often matter more than the model label alone.
| Product surface | License state | Likely grounding source | Can it use GPT/OpenAI? | Can Claude appear? | One-sentence answer to give users |
|---|---|---|---|---|---|
| Copilot Chat (Basic) | Free or baseline access | Primarily web data | Yes | No by default | “This is a GPT-based Copilot chat, but it is mostly web-grounded unless you add work content or have a licensed work-data experience.” |
| Microsoft 365 Copilot in apps | Licensed Microsoft 365 Copilot | Web data plus work data | Yes | Yes, in some experiences | “This is Microsoft 365 Copilot, which can use GPT models and also ground replies in your work data through Microsoft Graph and Work IQ.” |
| Copilot Chat with Frontier features | Eligible tenant or app setting | Depends on the selected mode | Yes | Yes, if enabled | “This Copilot can switch models or default to Auto, so the reply may not come from one fixed model.” |
| Researcher with Model Council | Eligible tenant or feature access | Multiple models in parallel | Yes | Yes, if available in the experience | “This feature compares several models and combines their outputs, so one screen can reflect more than one model family.” |
What model powers Copilot?
OpenAI GPT is the model family, while ChatGPT is OpenAI’s product. Microsoft says Copilot can use OpenAI’s GPT models, and Microsoft Copilot responses are powered by large language models, but that does not make Copilot the same product as ChatGPT.
OpenAI GPT models versus ChatGPT
ChatGPT is OpenAI’s conversational product. GPT is the model family behind ChatGPT and other systems. In Copilot, the model may be GPT-based while the product experience remains Microsoft Copilot, with Microsoft’s own orchestration, grounding, and controls layered on top.
Where Microsoft Graph and Work IQ fit
Microsoft 365 Copilot is described as model diverse by design, and Microsoft says it combines models with Microsoft Graph and Work IQ. That means the reply can be grounded in your organization’s files, mail, calendar, chats, and other permission-scoped content, not just the model’s general training.
Why the same prompt can return different replies
Microsoft says Copilot responses differ by data grounding, integration depth, and licensing. A prompt in Copilot Chat (Basic) may draw from web data only, while the same prompt in Microsoft 365 Copilot can use both web and work data. Two OpenAI-based replies can still differ because the grounded data is not the same.
Say your finance team asks, “Summarize the Q3 variance.” In Copilot Chat (Basic), the answer will tend to stay general unless the user pastes the report. In Microsoft 365 Copilot, the same prompt can pull from permission-scoped files and mail, so the result can be much more specific.
How Microsoft 365 Copilot uses work data
Microsoft 365 Copilot can ground answers in organizational content through Microsoft Graph and Work IQ, but access stays permission-scoped. That means the assistant can only use the files, messages, and records the signed-in user is allowed to see.
Permission-scoped access
This is the central boundary to understand. If a document is not available to the user through Microsoft 365 permissions, Copilot should not treat it as visible work data. The model may still answer from web content, but work grounding depends on access rights.
Web grounding versus work grounding
Copilot Chat (Basic) is primarily web-grounded. Microsoft 365 Copilot uses both web data and work data. That difference changes whether the assistant can summarize internal files, reference team notes, or answer using company-specific context without the user pasting it in.
Why licensed users see more context
Licensed Microsoft 365 Copilot users can get deeper integration inside Microsoft 365 apps. The practical result is not “a smarter chatbot” in general; it is a system that can combine model output with enterprise data, app context, and the user’s permissions.
Does Microsoft Copilot use Claude?
Sometimes. Microsoft 365 Copilot now lets you choose between Anthropic’s Claude and OpenAI’s ChatGPT in some Frontier features, and the model selector in Copilot Chat can default to Auto. But Claude is not universal, and many users will never see it unless their tenant or feature set enables it.
When Claude may appear
Claude can appear in some Microsoft 365 Copilot experiences as part of Frontier features. Anthropic operates as a Microsoft sub-processor, and Anthropic is off by default in EU tenants. That means availability depends on region, tenant policy, and the specific Copilot surface.
What to tell users who expect one fixed model
Do not tell users that Copilot always uses one model. In Researcher, Auto mode is also called Critique, and Model Council runs prompts across multiple AI models and synthesizes results. Those experiences are designed to compare model outputs, not hide the fact that more than one model may be involved.
Suppose a user sees different behavior in two Microsoft 365 Copilot chats. One is in a standard work app, the other is in a Frontier feature with model choice. The difference may come from the selected model, the grounding source, or both.
Common mistakes
Most confusion comes from mixing product names, model names, and license names. The fix is to identify the exact Copilot surface first, then ask what data it can see, then ask whether the tenant exposes model choice.
- Do not equate ChatGPT with GPT. ChatGPT is the product; GPT is the model family.
- Do not assume every Copilot screen uses the same model. Microsoft Copilot varies by product and license.
- Do not ignore grounding. Web-grounded Copilot Chat and work-grounded Microsoft 365 Copilot can answer the same prompt differently.
- Do not treat one model label as fixed for all Copilot replies. Microsoft Copilot responses can vary by surface, tenant, and feature.
- Do not assume Claude is on for everyone. It may appear in some licensed experiences, but not by default for every user.
What to check if answers look inconsistent
- Open the same prompt in the same Copilot surface again.
- Check whether the session is Copilot Chat (Basic) or Microsoft 365 Copilot.
- Look for a model selector or Auto mode in Frontier features.
- Confirm whether the prompt is using web data only or also work data.
Feature-by-feature comparison
ChatGPT is the standalone OpenAI chat product. Microsoft Copilot is Microsoft’s assistant family, which appears in the Copilot app and Microsoft 365 apps, and can use GPT models while adding Microsoft grounding, enterprise controls, and permission-scoped work data.
| Feature | Microsoft Copilot | ChatGPT |
|---|---|---|
| Main workflow | Inside Microsoft 365 and Microsoft Copilot surfaces | Standalone chat and general-purpose prompting |
| Grounding | Web data, and in licensed contexts, work data through Microsoft Graph and Work IQ | Primarily OpenAI’s own product context and connected tools |
| Work data access | Permission-scoped in Microsoft 365 Copilot | Not tied to Microsoft 365 permissions |
| Model family | Can use OpenAI GPT models, and sometimes Claude in Frontier features | Uses OpenAI models |
| Best fit | Microsoft 365 workflows, files, mail, meetings, and enterprise context | General-purpose chat and standalone brainstorming |
What to answer users in one sentence
If someone asks whether Microsoft Copilot uses ChatGPT, the clean answer is: Microsoft Copilot can use OpenAI GPT models, but it is not ChatGPT, and the exact model, grounding source, and reply behavior depend on the Copilot product, the license, and the tenant settings.
For Microsoft 365 Copilot, add that replies can combine web data with permission-scoped work data through Microsoft Graph and Work IQ. For Copilot Chat (Basic), add that the baseline experience is mostly web-grounded unless licensed work-data access or supplied content changes the result.
Frequently asked questions
Does Microsoft Copilot use ChatGPT or Claude?
It can use either, depending on the experience. Microsoft 365 Copilot now lets some users choose between Anthropic’s Claude and OpenAI’s ChatGPT in Frontier features, while Copilot Chat can default to Auto. Availability depends on tenant settings, region, and product surface.
Does Microsoft 365 Copilot use ChatGPT?
Yes, in the sense that Microsoft says Microsoft 365 Copilot can use OpenAI GPT models. It is still not ChatGPT as a product. Microsoft 365 Copilot also combines those models with Microsoft Graph and Work IQ, so the reply is grounded in work data and permissions.
Does Microsoft Copilot just use ChatGPT?
No. Some surfaces can use GPT models, some can offer Claude, and some can default to Auto, so one label does not cover every Copilot reply.
Does Microsoft use ChatGPT for Copilot?
Microsoft uses OpenAI GPT models in some Copilot experiences, but that is not the same as using the ChatGPT product itself. The product layer, grounding layer, and license controls sit on Microsoft’s side, especially in Microsoft 365 Copilot and Copilot Chat.
Is Microsoft Copilot using ChatGPT 5?
Microsoft’s documentation says Copilot Chat includes GPT-based models, but that does not mean every Copilot experience is fixed to one visible version. The exact model can vary by surface, tenant, and feature, especially when Auto or other Frontier options are enabled.
Does Microsoft 365 Copilot use ChatGPT plus?
No. ChatGPT Plus is an OpenAI subscription product, not a Microsoft 365 Copilot license. Microsoft 365 Copilot uses Microsoft licensing, Microsoft Graph, and Work IQ, so the relevant question is whether your tenant has the Microsoft Copilot feature and license enabled.