Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
With a generative answers node, your agent can respond to your users based on knowledge sources at the topic level. Knowledge sources allow you to give your users authoritative responses, based on information that you provide your agent.
When you first create your agent, you can enter a URL your agent uses to generate responses. This URL is used in all features of the agent. However, you can boost your agent's conversations by using multiple internal and external knowledge sources within individual topics.
Generative answers as a fallback
When your agent can't find a matching intent (defined in a topic) for the user's query, it uses generative answers to try to answer the question. This behavior is known as generative answers as a fallback. If the user's intent isn't matched to topics or generative answers, the Fallback system topic is used. System topics can escalate a query for the agent.
Generative answers knowledge sources
Within an agent's topic, you can add a generative answers node. This node allows you to specify more sources that the node searches based on your inputs. Sources defined in the generative answers node override knowledge sources you specified at the agent level, which functions as a fallback. For the best results, configure your generative answers nodes with specific knowledge sources.
These sources include:
External resources:
- AI general knowledge
- Bing Web Search (doesn't require external configuration)
- Bing Custom Search (requires external configuration)
Internal resources:
- Azure OpenAI on your data
- Documents uploaded to Dataverse
- SharePoint (only specific file formats are supported)
- Custom data (internal or external): supply your own source, such as a Power Automate Flow or from Skill.
Supported knowledge sources
Different knowledge sources offer different capabilities. These capabilities include the number of inputs, the type or format of the data, and how (or if) the source needs authentication to access it.
| Name | Source | Description | Number of inputs supported in generative answers | Authentication |
|---|---|---|---|---|
| Public website | External | Searches the query input on Bing, only returns results from provided websites | Generative mode: 25 websites Classic mode: Four public URLs (for example, microsoft.com) |
None |
| Documents | Internal | Searches documents uploaded to Dataverse, returns results from the document contents | Generative mode: All documents Classic mode: Limited by the Dataverse file storage allocation |
None |
| SharePoint | Internal | Connects to a SharePoint URL, uses GraphSearch to return results | Generative mode: 25 URLs Classic mode: Four URLs per generative answers topic node |
Agent user's Microsoft Entra ID authentication |
| Dataverse | Internal | Connects to the configured Dataverse environment and uses a retrieval-augmented generative technique in Dataverse to return results | Generative mode: Unlimited Classic mode: Two Dataverse knowledge sources (and up to 15 tables per knowledge source) |
Agent user's Microsoft Entra ID authentication |
| Enterprise data using connectors | Internal | Connects to connectors where your organization data is indexed by Microsoft Search | Generative mode: Unlimited Classic mode: Two per custom agent |
Agent user's Microsoft Entra ID authentication |
Note
Agent user authentication for knowledge sources means that when a specific user asks a question of the agent, the agent only surfaces content that the specific user can access.
Knowledge sources in generative answers nodes currently don't support Bing Custom Search, Azure OpenAI, or Custom Data. Instead, from the generative answers node properties, use the Classic data option for Bing Custom Search, Azure OpenAI, or Custom Data sources.
For websites, you need to confirm which website(s) your organization owns that Bing will search through Copilot Studio.
You can perform language-agnostic querying across all supported file types and languages.
If you're using unstructured data, such as individual SharePoint files and folders, OneDrive files and folders, or connectors, there are different limits and limations. For more information, go to Limits and limitations.
Currently, citations returned from a knowledge source can't be used as inputs to other tools or actions.
Add a generative answers node
Go to the Topics page and open the topic you want.
Select the Add node icon
under the node after which you want to use generative answers, point to Advanced, and select Generative answers. A new node called Create generative answers appears.Select the More icon (…) of the node, then select Properties. The Create generative answers properties pane appears.
You can specify and configure your new data sources:
Customize the agent's response
This section shows you how to store the agent response in a variable instead of immediately returning it, so that you can include it on an Adaptive Card.
On the Create generative answers node, for Input, select the Activity.Text system variable.
In the Create generative answers properties panel:
Expand the Advanced section.
Create a global variable. Enter a meaningful name for the variable. This variable stores the answer generated in response to a query from the user.
Clear the Send a message box.
Note
Clearing this option prevents your agent from immediately returning the generated answer, which allows you to customize the answer. Limitation: If you publish this agent to Teams, you must explicitly include rendering of citations for the customized answer. In Teams, Copilot Studio agents automatically return citation links only for answers without customizations.
Add either a Question node or a Message node, and add an Adaptive Card to this node.
In the Adaptive Card properties panel, switch to Formula, and replace the default content with a Power Fx formula that uses your global variable and the customizations you want.
Authentication
Some sources require authentication because the agent makes calls on behalf of the user in the chat window. The calls use the account credentials from copilotstudio.microsoft.com.
The authentication settings you configure in the agent require manual authentication with the Service Provider type of Microsoft Entra ID.
To configure authentication and create a Microsoft Entra ID, see:
- Configure user authentication in Copilot Studio.
- Configure user authentication with a Microsoft Entra ID.
Include delegated permissions for Microsoft Graph:
Files.Read.AllSites.Read.All
Supported content
Generative answers get content stored in these formats:
- SharePoint pages (modern pages)
- Word documents (docx)
- PowerPoint documents (pptx)
- PDF documents (pdf)
Note
For a list of limits and supported SharePoint functionality, see Copilot Studio web app SharePoint limits.
Content moderation
Set the content moderation settings for a generative answers node from the Properties pane. The content moderation settings allow your agent to provide more answers. However, the increase in answers might affect the allowance of harmful content from the agent.
Select the More icon (…) of the node, and select Properties.
Select the desired moderation level.
The moderation levels range from Lowest to Highest. The lowest level generates the most answers, but they might contain harmful content. The highest level of content moderation generates fewer answers, and applies a stricter filter to restrict harmful content. The default moderation level is High.
Select Save at the top of the page.
Note
If you set your generative answers node to moderate content, it might not return answers. For more information on how to diagnose answers not being returned, see Content blocked by content moderation.