The Preview panel is a built-in testing tool inside the AI Agent configuration page. It lets you simulate a real guest conversation and see exactly how your AI Agent will respond, before it goes live or after you make any changes to its setup.
How to open it
Navigate to Inbox Settings > Bots, then select the bot you would like to test. Once you are on the configuration page, look for the split-panel icon in the top-right corner of the configuration area (it looks like two side-by-side rectangles).
Clicking it opens a right-hand panel labelled Preview, with a message input at the bottom that says "Type a prompt to test the bot."
What you'll see
The Preview panel sits alongside the configuration tabs (General, Knowledge, Testing), so you can adjust settings and test at the same time without switching pages.
At the top of the panel you'll notice:
A channel selector, setting which channel the simulated guest is writing from. It offers Web chat, WhatsApp, Email and SMS.
A conversation counter (for example, "1 of 1") with arrow controls, useful if you have multiple test threads running.
A + button to start a new test thread, and a refresh/reset button to clear the current conversation and start fresh.
At the bottom, above the message input, sits Guest profile context — a toggle that lets you run the test as a specific guest. Switching it on reveals a Search for a guest field.
The Testing tab
The Testing tab (next to General and Knowledge) is where you set up a scripted test before running it. It has two fields:
Template. A pre-written guest scenario. Selecting one — for example Room Available — loads a set of realistic guest messages into the Preview panel, which you step through using the conversation counter arrows. This is optional, but it is the quickest way to run the same test consistently every time you change something.
Language. Which language version of that scenario to load. Use it to check that your AI Agent answers sensibly in the languages your guests actually write in.
These are test scenarios, not agent templates. The four starting templates described in Choosing a starting template are something different — they are the property types you pick when creating an agent. The Template field here only loads sample guest messages for testing, and does not change your agent's configuration.
Setting the channel and the guest
Two more controls affect how a test behaves, and both live in the Preview panel rather than on the Testing tab.
Channel. The selector at the top of the panel chooses the channel the simulated guest is writing from: Web chat, WhatsApp, Email or SMS. Together with Guest profile context below, this is what lets you test reservation verification on a trusted channel, a wrong booking reference, or an untrusted channel that redirects. The panel runs the real verification logic in memory: nothing is stored and no real guest data is touched. See Answering reservation questions: how guest verification works.
Expedia is not in the list. The AI Agent answers on Expedia in production, but Expedia conversations cannot be simulated here, so anything channel-specific about them has to be checked on a real conversation.
Guest profile context. Switch this on at the bottom of the panel and search for a guest by name, then select one from the results. The AI Agent then receives that guest's actual background context through the same path it uses live. The summary is printed into the transcript as a grey block above the agent's reply, so you can read exactly what the agent was given — returning visitor, language, audience segment, typical length of stay, and so on. It never includes the guest's personal data. See How the AI Agent recognises returning guests.
You can only pick a guest who already exists in your account. That is the same population the feature covers in production: a guest with no master profile gets no context, in testing or live.
Comparing a returning guest with a first-time guest on the same question is the clearest way to see the effect.
How to use it
Open the Preview panel using the split-panel icon (top-right).
Type a message in the "Type a prompt to test the bot" field at the bottom of the Preview panel (for example, "What time is the check in?") and click the send button (arrow icon).
To test another question, click the input field again, type a new message (for example, "What time is the check out?"), and send.
Click the + button to create a new test thread.
Use the conversation counter arrows (for example, "1 of 2", "2 of 2") to navigate between different test threads.
Review the AI Agent's response. Does the tone match? Is the answer accurate? Does it pull from the right knowledge? If something needs adjusting, switch to the General or Knowledge tab and make the change — the panel stays open beside you — then send the message again.
Optionally: test with a template
Go to the Testing tab and click Select a template.
Select a Template to simulate a specific guest context.
Click the arrow to start the simulation.
Review the AI Agent's response.
Tips
Test the greeting on a fresh thread. If you have set a greeting on the General tab, it appears on the agent's first message of a conversation and not on the ones after it. So to see it, use the + button or the refresh button to start a clean thread — sending a second message in the same thread will not show it.
Test edge cases. Not just the questions your AI Agent should handle perfectly, but the ones it might mishandle: vague messages, off-topic requests, or questions in different languages.
Check the status badge. The top of the page shows whether the AI Agent is Active, Active with draft updates, or inactive. If you have made changes that have not been activated yet, the Preview tests your draft configuration, not the live one. This is useful: it lets you validate changes before publishing them.
Activate only when satisfied. The Activate button in the top-right publishes your current draft. Do not click it until the Preview confirms the AI Agent is behaving as expected.
Need Help?
Please contact us through the 'Talk to Us' option on the left menu in the platform, or through the Bookboost Support email at support@bookboost.io if you have questions or need additional support.