Copilot knowledge bases are not updated directly from individual user chats. To make information like this available in Copilot answers, it must be added to a supported knowledge source that Copilot can index, or implemented as part of a custom knowledge experience.
Based on the available information, there are two main patterns for adding or influencing Copilot’s knowledge:
- Add content to a knowledge source Copilot can index
In products that support knowledge sources (for example, Dynamics 365 or Copilot Studio agents), knowledge is pulled from configured sources such as:- Uploaded files (PDF, Word, RTF, etc.)
- Websites
- External systems via connectors (for example, ServiceNow Knowledge)
- Open Copilot Studio and select the appropriate Dataverse environment.
- Open the relevant copilot/agent (for example, Copilot for finance and operations apps).
- On the Knowledge tab, select Add knowledge.
- Upload a file or configure another supported knowledge source that contains the statement or article you want Copilot to use.
- Wait for the knowledge source status to become Ready, test with sample questions, and then Publish so users can benefit from it.
- Build or extend a custom agent that uses your own knowledge
If the goal is to have a copilot reliably answer questions about a specific product behavior (such as the Workflowy markdown behavior), a custom agent can be created and wired to a custom knowledge source:- In Copilot Studio, create or open an agent.
- Add knowledge sources (files, websites, or connectors) that describe the issue and its resolution.
- For advanced scenarios, build a custom knowledge source using a topic with the
OnKnowledgeRequestedtrigger that calls a search API and returns results in the format Copilot Studio expects. This allows full control over what content is returned as “knowledge” when users ask related questions.
In all these cases, Copilot does not learn permanently from a single conversation. The information must be stored in a supported knowledge source (file, site, connector, or custom API) and then indexed or exposed through a Copilot Studio agent so that future users can benefit from it.
References: