Once the name, tone, assignment rule, and knowledge are set, you activate the AI Agent to take it live.
Activate does two jobs. The first time you use it, it takes the agent live. After that, it publishes whatever changes you have made since — knowledge edits, tone changes, a new greeting. Until you click it, the header shows Active with draft updates and guests continue to see the previous version.
Before you activate
The name is what you want guests to see. Once live, this is visible immediately.
The assignment rule is correct. An AI Agent with no rule, or a too-broad rule, can start replying to conversations you did not intend to automate.
You know what this agent's channel can do. Reservation questions only work on WhatsApp, SMS, and email. If this AI Agent is assigned to Expedia, web messenger, or a social channel, it will redirect those questions instead of answering them. See Answering reservation questions: how guest verification works.
The knowledge section has at least one entry. An AI Agent with no knowledge will hand off almost every conversation to a human. It can still answer reservation questions on a verified channel, but nothing about your property.
If the agent is empty, the quickest way to fill it is Import — from your hotel's website, from a policy document, or by copying the knowledge out of an agent you have already built. See Importing knowledge into your AI Agent.
Every knowledge item has content in both fields. If you leave the content of an item blank, you will not be able to activate the agent and you will get an error. Either fill in the missing content or delete that item.
You have tested it in the Preview panel. Run through the questions guests actually ask. Use the channel selector and Guest profile context in the Preview panel to check how the AI Agent behaves for a returning guest, and for a guest asking about their own booking. See Testing Your AI Agent with the Preview Panel.
A test message from your own phone will not exercise reservation answers — a staff number is not stored on any reservation, so verification will correctly fail.
Decide whether to start in draft mode. If you would rather an operator checked the agent's replies before guests see them, switch on Enable draft mode on the General tab before you activate. Its answers then wait in the operator's composer to be sent, edited or rejected. See Draft mode: letting an operator check the agent before it sends.
After going live
Watch the inbox closely for the first 24 to 48 hours. Look at how the AI Agent is answering, where it is handing off, and whether there are gaps in the knowledge section. Conversations the agent could not handle show as Bot Left in the thread.
The Verified tag tells you when reservation data is in play. On threads where a guest has verified, a Verified tag appears next to their name in the sidebar. Guest background never appears in the inbox — it shapes the reply without being shown.
Iterate. Quality compounds with knowledge updates. Most teams see meaningful improvement after the first two or three rounds of refinement.
When you go looking for a gap, filter the Knowledge tab by tag rather than scrolling the whole list — it is much faster to read everything the agent holds on one topic at once. See Organising knowledge with tags.
Remember to Activate again. Every change you make after going live sits as a draft until you publish it. If a fix does not seem to have taken effect, check whether the header still says Active with draft updates.
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.