Testing Environments
There are two ways to test your chatbot within the dashboard: the Design Preview for visuals and the Dedicated Test Page for behavior.1. Design Preview
The Design page features a real-time preview of your widget’s appearance.- Navigate to your website dashboard.
- Click Design in the sidebar.
- The right side of the screen shows a fully interactive preview.
2. Dedicated Test Page
For evaluating the chatbot’s intelligence and responses, use the dedicated Test page.- Click Test in the sidebar menu.
- The page has two tabs: Chat (the default) and Answer tests.
- The Chat tab is a full-screen chat interface with a Debug & Improve side panel.
- Full-screen chat: Focus entirely on the conversation without distraction.
- Debug & Improve panel: A side panel (desktop only) that shows confidence scores, sources, and improvement tips for each bot response.
- Confidence monitoring: Every bot response is analyzed and given a confidence score — helping you quickly spot weak answers.
- Quick Q&A creation: Add Q&A entries directly from the test page when the bot struggles with a question.
- Reset Chat: Start a new conversation by clicking “New Chat”.
Debug & Improve Panel
The Debug & Improve side panel (visible on desktop) gives you real-time insight into how the bot answered each question.Confidence Scores
After each bot response, the panel displays a confidence indicator:Sources
The panel lists the sources the bot used to generate its answer. You can click any source to open it and verify the content is accurate and up to date.Low-Confidence Alerts
When the bot’s confidence drops below 70%, a low-confidence alert appears below the response. On mobile devices, this alert appears inline; on desktop, the side panel highlights the issue.Quick Actions
For medium and low-confidence responses, a “Add Q&A for this question” button appears in the panel. Clicking it opens a dialog pre-filled with the question you asked, so you can quickly provide the correct answer.Answer test suite
Answer tests are saved test cases that re-check your chatbot’s answers after every change. They are available on every plan, including Free. Open Test in the sidebar, then the Answer tests tab. Owners and editors can create, edit, delete, and run tests.Create tests
- Suggest tests: BubblaV builds tests from your top real visitor questions (last 90 days) and your Content Gaps. Review the suggestions, edit them, and add the ones you want.
- Save as test (Chat tab): click Save as test in the Chat tab header. BubblaV uses your latest question and the bot’s answer. You write what the chatbot should have said.
- Save as test: open the ⋯ (More actions) menu in an Inbox conversation header and click Save as test, or use Save as test in the Reports thumbs-down conversations table. The visitor messages are prefilled, and you write what the chatbot should have said.
- New test: enter the test by hand.
- 1–5 visitor messages, sent in order as one conversation.
- Expected behavior, in plain language. For example: “Mentions the 30-day return window and does not invent a discount code.”
- An optional expected tool: a tool name, or
flow:<name>to require that a flow is detected. The test fails if the chatbot does not use it. - Optional tool mocks: fixed replies for a tool, so the test does not depend on live data.
Run tests
- Run all runs every enabled test. Use Run on a row to run one test. Disabled tests are skipped, so enable a test before you run it.
- Enabled tests also run automatically about 5–10 minutes after you change custom instructions, Q&A, files, flows, or tool settings. Disabled tests are not rechecked. Automatic runs only happen when you have at least one enabled test.
Read results
Each test shows Passed, Failing, or Error. An AI grader decides pass or fail and gives a short reason. Expand a row to see the full conversation, the chatbot’s answer, the sources it used, and the tools it called. A test shows “Failing since” with the date it first failed. When you change a test’s visitor messages, expected behavior, expected tool, or tool mocks, its last result is cleared and the test shows as not run until it runs again. Renaming a test or turning it on or off keeps the result. Click Try in chat on a test row to replay the test in the Chat tab. BubblaV starts a new chat and sends the visitor messages of the test in order.- Tools run live. Mocks are not used.
- Flows can start.
- Each replayed message uses one AI message.
- BubblaV asks you to confirm first when the test has tool mocks, an expected tool, or its last run called any tool.
- Replayed chats are not counted in Reports or Suggest tests.
Tool safety
Test runs never change your live systems.- Write actions (cancel an order, create a booking, send a message) never run. The chatbot is told the action succeeded, and the result shows Write blocked.
- Read tools (order lookups, searches) run live unless you add a mock.
- Flows are only detected. The result shows Flow would start, and no flow step runs.
Limits
Every plan includes answer tests; plans differ only in the monthly number of test runs. One run is one test case that is executed and graded. Test runs do not use AI messages.
When you reach the limit, automatic runs are skipped and manual runs are refused until the monthly reset.
Weekly email
The weekly report email lists tests that started to fail during the week and have not passed since.What to Test
Knowledge Accuracy
Ask questions about your business to ensure the AI has learned your content correctly.- Product details: “What is the price of X?”
- Policies: “What is your return policy?”
- Support: “How do I contact support?”
Behavior
Check that the bot adheres to your settings:- Does it use the correct tone?
- Does it escalate to human support when needed?
- Does it capture lead information?
Debugging
If the bot says “I don’t know” or gives wrong information:- Check the confidence score and sources in the Debug panel to understand why.
- Add missing information to the Knowledge Base.
- Use the “Add Q&A” button in the panel to provide a direct answer for that specific question.
- Use the Q&A section or resolve Content Gaps for broader improvements.
- Click Save as test in the Chat tab header, so the fix stays fixed. BubblaV re-checks the question after every change.