AI Playground
The AI Playground ("Test AI Prompt") runs a prompt against a real conversation and shows the result immediately — without saving anything to the conversation. Use it to validate and iterate on a prompt before enabling an AI task for your whole organization.
Where: Administration › Conversation Analytics › AI Tasks, tab Playground. You can also reach it from a task: the Test button on a task's settings page and Save and Test in the task editor open the Playground with that task's prompt loaded.
Testing a prompt
- Step 1. Select a call — pick the conversation to experiment on. The list is pre-filtered to conversations that have a transcript; adjust the filters to find a representative example, then click Select for an experiment.
- Step 2. Run an Experiment — edit the prompt, run it, and read the AI's response. Iterate: change the wording, run again, and compare.
- When the output is right, save the prompt back to the AI task (Save AI Prompt).
Nothing you run in the Playground is stored on the conversation — results are shown to you only, so you can safely test on production conversations.
Tips
- Test on several conversations, including edge cases: very short conversations, conversations where the answer should be "no evidence", and both positive and negative examples.
- If the task writes JSON (see Custom insights), verify the response is valid JSON with exactly the fields you expect — then add a response JSON schema to the task.
- Prompt changes made in the Playground take effect for future conversations only after you save them to the task and the task is enabled.
Related pages
- AI tasks — enabling tasks and editing prompts.
- Custom insights — prompt-writing guidance and recipes.