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.
When you prepare generative answers by using public data, it can be challenging to target specific content in the context of expected question flows. In these cases, you can integrate Bing Custom Search to optimize generative answers and prompting opportunities.
Example scenario
Suppose you create an agent for an automobile company website designed to help customers find the right car for their next purchase. The website might mention many vehicles across different pages and perspectives. When a customer asks about a specific model, the agent uses any content related to that model found on the website, as configured in the settings or the Generative Answers node. This broad context approach can lead to suboptimal answers and less relevant answers for the customer.
Bing Custom Search supports Copilot Studio generative answers as it helps to target and/or exclude public-facing content. It also gives the agent maker a way to test, optimize, and refine common queries.
Targeting content: Use the Active tab of Bing Custom Search to specify which URLs are active and available for search queries.
Excluding content: Use the Exclude tab of Bing Custom Search to specify which URLs are to be excluded (and therefore won't be used in query results).
Optimizing queries: Pin queries that users ask frequently, as described in Pin slices to the top of search results.
Use Bing Custom Search when it helps the agent provide the most accurate answers in the context of user intent. Consider how might use this capability in the Contoso Premium Vehicles agent scenario.
Responding to the example scenario
Contoso Premium Vehicles uses an agent to help customers find the right car for their needs. The company's goal is to increase conversion rates and expedite purchases. To start the customer journey, you first need to determine the preferred purchase option: purchase (buy) or lease. Then, use Bing Custom Search to send the agent to the appropriate pages for that type of purchase, while excluding any expired offers.
Note
Contoso Premium Vehicles is a fictitious website. Therefore, you won't be able to test these steps with its website links. Just note the logic behind these steps for your own purposes, replacing the links with your own website specifics.
Create an account for Bing Custom Search at Bing Custom Search.
Create two new instances for the Contoso Premium Vehicle's agent, one for Purchase and one for Lease (as shown in the following screenshot). Two instances are needed because there are different content targets depending on the purchase option, and you plan to surface each instance in a different plugin or topic.
Start with Lease. Open the instance "Contoso Premium Vehicles – Lease" and add a page for <www.ContosoPremiumVehicles/LeaseInventory> on the Active tab.
Exclude the page with last year's expired lease inventory by selecting the Blocked tab and adding this URL: <www.ContosoPremiumVehicles/2024Specials>.
Select Publish to save this change. The first publish also gives a configuration ID for use in Copilot Studio.
Next, open the instance "Contoso Premium Vehicles – Purchase" and add a page for <www.ContosoPremiumVehicles/PurchaseInventory> on the Active tab.
Select Publish to save.
Now you're ready to apply the results of these configurations to topics with a Create generative answers node, and your agent will only generate answers based on the content you specified.
