The Test Knowledge feature can help surface any gaps in your knowledge base, which can then inform your approach to uploading updated information - all to ensure the audience interacting with your agent receives the best and most up-to-date information available.
You can use this feature to test the assets stored in your Knowledge section with a status of "Ready" in the status column.
Note: Only "Ready" assets can be tested.
How testing knowledge differs from testing your agent
Testing Knowledge will tell you whether a question is answerable based on content in the knowledge base, or whether it is unanswerable based solely on that content.
Note: Results when testing knowledge focus on factual information and are not simulations of an actual response that your AI agent might supply.
In other words, testing Knowledge helps highlight content gaps in your Knowledge assets, but does not reflect the tone, context, or style in which an agent might actually respond to inbound questions.
To test how your agent might respond to your constituents, you will want to use the test feature within a workflow.
Steps to test knowledge
Navigate to your Knowledge section.
Click Test Knowledge at the top of the screen.
Enter a question in the test window.
Review the agent's response and examine which assets were used to generate it.
The test view highlights the documents, webpages, interviews, or emails the agent drew from. These sources allow you to confirm where the agent looked to confirm the requested information and confirms your asset library supports accurate, helpful responses.
Evaluate your results. You can continue to add assets and test. When you are satisfied with your testing results, you are ready to publish your assets!
Testing in staging vs. live
The Knowledge Tester includes a toggle that lets you switch between Staging and Live environments so you can compare how results differ based on your vault's publish status.
Staging includes all vault content regardless of whether it has been published.
Live includes only content that has been published.
The toggle defaults to Live so that test results reflect your agent's currently active knowledge — the same content your constituents are interacting with. Each answer returned in the test view includes a small indicator showing which environment it was generated from, making it easy to compare results across both settings.
Note: To ensure your Live results are up to date, remember to publish your knowledge base changes regularly. Unpublished assets will only appear when testing in Staging.
Testing best practices
Keep the following in mind as you begin testing your agent
Confirm your agent has information related to the most common questions. Consider the top questions that your users will ask your agent, and then use the testing feature to ensure the agent is adequately informed on those topics.
If you are not satisfied with the test responses, ask yourself- what additional training assets does my Agent need to be able to answer this question accurately? Then, add those assets, wait a few moments for them to be "Ready," and retest!
Update as you go. If you have additional assets handy, go ahead and make updates once you have the results of your testing. You can see the test results update once the updated information has been fully ingested!
Reminder: Knowledge testing only looks at assets that are flagged as "Ready" in the status column, so allot a few minutes for your assets to process before re-testing.
Optimizing your knowledge base
After reviewing the test results, you can:
Add missing assets if the agent lacks the information it needs
Remove outdated or irrelevant content
Edit or replace assets to improve clarity or accuracy
Boundaries and expectations
Testing shows how an agent would behave, not how every real interaction will unfold. Ongoing monitoring and iteration remain important after launch.
Your CollegeVine team can assist with test planning and interpretation if needed.
Next steps
Once testing is complete and behavior meets expectations, workflows can be launched to the intended constituent segment.