Copilot Grounding.

Copilot grounding is the use of relevant context to guide an AI model’s response to a specific request. Instead of relying only on patterns learned during model training, a Copilot experience may also draw on information available in the current interaction or retrieved from permitted sources. The exact sources depend on the product experience, account, configuration, and request.

Grounding is not the same as giving Copilot unrestricted access to an organization’s information. Access controls, user permissions, enabled capabilities, and the information supplied or retrieved for a request all affect what may be used. Grounding can make an answer more relevant to a task, but users still need to check important claims against the underlying information.

A request provides the starting point: its wording, any details included by the user, and relevant prior turns in the conversation can help establish what the user wants. If the experience supports information retrieval, Copilot may use additional context, such as web results or work content available to that user, to formulate a response.

The model uses the available context to generate an answer. This process does not necessarily mean that every possible source is searched, or that every relevant detail is found. A response may omit information, misunderstand context, or combine retrieved material with the model’s general learned patterns. The sources and amount of context can differ between experiences and requests.

Depending on the Copilot experience and how it is configured, a response may be influenced by:

• The current prompt, including text, questions, and instructions provided by the user.
• Earlier messages in the conversation that remain part of its context.
• Files, text, or other content the user provides or makes available in the interaction.
• Public web information, when web access is available and used for the request.
• Work content the user is authorized to access, where the Copilot experience supports relevant Microsoft 365 data grounding.
• Other connected data sources, if the product, permissions, and organizational configuration support them.

These are possibilities, not a guarantee that every response uses every source. The available context can vary by account type, subscription, application, tenant configuration, region, and service updates.

Copilot Chat is generally oriented toward conversational assistance, which may include working with a user’s prompt, supplied content, and web information where available. Microsoft 365 Copilot is designed to work within Microsoft 365 experiences and may ground responses in organizational work data the user is permitted to access. The two experiences can therefore differ in work-data grounding and in how they connect with Microsoft 365 applications.

Neither label alone tells a user exactly what data a particular answer used. Available capabilities and data access can depend on the account, subscription, application, tenant configuration, permissions, region, and service updates. Users should distinguish between content they explicitly provide, information retrieved from the web, and work information surfaced through an eligible and configured experience.

Suppose a project manager asks Copilot to draft a status update. If the manager supplies meeting notes and schedule details, those materials can provide context for the draft. In a Microsoft 365 experience configured to access relevant work data, Copilot may also be able to use permitted project information, depending on the available integration and the manager’s access rights.

The draft should still be checked against the original notes and project records. If a milestone is missing from the supplied material or is not accessible to the user, Copilot may not include it. A confident-sounding summary is not proof that all relevant project data was available.

Grounding can improve relevance, but it introduces practical limits. Retrieved information may be incomplete, outdated, ambiguous, or unrelated to the question. User permissions and system configuration also affect which work content can be considered. A result should not be treated as a complete inventory of organizational information.

For sensitive or consequential work, users should pay particular attention to:

• Whether the response identifies or links to the information it relied on.
• Whether the cited or referenced material supports the answer as written.
• Whether the user has permission to access the underlying data.
• Whether the request includes confidential information that should not be entered into that Copilot experience.
• Whether an important decision needs confirmation from an authoritative business record or subject-matter expert.

Copilot grounding connects a request with context that may include conversation details, user-provided material, web information, or permitted work data. This can make responses more useful for business tasks, while the actual sources depend on the Copilot experience and its configuration.

Grounding does not guarantee that an answer is accurate, complete, or based on every relevant record. Users should verify consequential claims, understand the permissions and data boundaries that apply, and avoid assuming that different Copilot experiences draw on the same information.