The Playground side panel contains three testing areas:
On smaller screens these controls are available through the Playground Tools panel.
The Scenario area lets you test the Agent under a selected Site Type and Language context without permanently changing the Agent.
A Site Type override is identified in the Playground header with an (override) indicator.
Use Site Type override when you need to test how the same Agent behaves under another vertical configuration.
This is especially useful before changing the Agent's actual Site Type or when evaluating whether a preset better matches the website.
The Scenario panel also supports a Language override.
When active, the Playground header identifies the override so you can distinguish test behavior from the Agent's normal Language configuration.
Playground provides sample prompts based on the effective vertical.
Select a sample prompt to place it into the Chat input, or enter a custom question that better represents your real visitor traffic.
The Page context tab lets you simulate information about the webpage where a visitor might be chatting.
Page context can include page-level information such as URL, Title, description, headings, visible text, structured metadata, and supported commerce context.
This makes it possible to test page-aware behavior without navigating a real visitor through every production page.
An Agent may produce different answers depending on the page the visitor is currently viewing.
Use Page Context to verify that page-specific information improves the answer rather than introducing irrelevant assumptions.
After a Playground turn, the Diagnostics tab exposes technical information that shaped the response.
The interface identifies diagnostics including:
The Diagnostics panel lists retrieved Sources and can show:
Use this information to determine whether the Agent retrieved the right Knowledge for the test question.
A completed Playground reply can display a Low confidence badge.
When this appears, inspect Diagnostics and the underlying Knowledge before simply lowering the confidence threshold. The better correction may be to improve the Agent's Sources.
Playground can mark replies with:
These labels help confirm which response path handled the visitor message.
Select Compare to test two vertical configurations side-by-side.
The same visitor message is sent through both comparison paths, allowing you to review the resulting Conversations together.
On desktop, the two responses appear in parallel columns. On smaller screens, A and B tabs let you switch between the two comparison results.
The primary side represents the effective main Site Type. Select a different Site Type for the comparison side.
Use this to compare response quality before changing the Agent's permanent vertical.
Use Regenerate to re-run the last visitor message under the currently selected test configuration.
Use Reset when you need a completely clean Conversation and Diagnostics state.