Test the chatbot before customers do
Chat with your chatbot yourself, in a private room that no customer ever sees. Every reply tells you what produced it, so a wrong answer points straight at the thing you need to change.
Do this before you switch the chatbot to Live, and again after every change you make to a stage.
Before you start — you need a connected WhatsApp number. The Test button is greyed out while nothing is connected, and while the settings are still loading.
Nothing here reaches a real person. The test room is between you and the chatbot. No WhatsApp message is sent, and no contact is created.
Steps
Press Test
The Test button is in the toolbar at the top right of the AI Chatbot page, next to the AI chatbot enabled switch. It carries a flask icon, and on a narrow screen the word disappears and only the flask is left.
It is on every tab, so you never have to go back for it.
See which room you land in
The button opens whichever surface suits your chatbot, and it will be one of two:
| Your chatbot | What opens |
|---|---|
| Has a playbook | The Playbook tab switches to Test mode, beside the stages |
| Has no playbook, or is v1 · legacy | The Test Chatbot dialog |
The pill beside the title is the connection, not the chatbot
The pill beside Test Chatbot reads Connected in green, Connecting… in amber while it is still opening, or Disconnected in red if the connection dropped. Close at the bottom ends the session; on a phone the whole thing is a sheet you can swipe away.
Send something a customer would send
The box at the bottom says Message the agent… — or Connecting… while the connection is still coming up, which is about the connection and not about the chatbot thinking.
Type a real first message. The awkward ones are the ones worth testing: how much ah, you open tomorrow?, can I get a discount.
While it works, the transcript shows your chatbot’s name followed by is thinking….
If the chatbot replies with buttons or a menu, they are real: tapping one sends back exactly what a customer tapping it on WhatsApp would send. A row of buttons is labelled Quick replies.
Read the chip under each reply
Under each reply — once per turn, not once per bubble — sits a small chip naming what produced it. This is the whole point of testing here.
| The chip says | What it means |
|---|---|
| A stage’s name | That stage wrote the reply. It also tells you when a guard rerouted the turn |
| An always-on rule | An always-on rule fired, and the chip shows the rule’s first line |
| Your written instructions | There is no playbook, so your whole brief is the instruction set |
| Nothing you wrote covered this turn | Nothing of yours applied and the chatbot answered on its own |
An amber “Nothing you wrote covered this turn” is a finding, not an error. No stage was active and no rule fired, so the model improvised. It is exactly the turn a new stage or an always-on rule should be given an address for.
Jump from a bad reply to the thing that caused it
In Test mode the chip carries a button, and it takes you to the editor rather than making you find it: Edit stage → for a stage, Edit rules → for an always-on rule, and Edit instructions → for your brief, which goes to the Brief tab.
Edit stage takes you to Build with that stage already open, on the right section
In the Test Chatbot dialog the chip still tells you what produced the reply, but it does not carry a button — there is no editor behind a dialog to land in.
Fix the stage, then start a new conversation
Fixing a stage does not restart the conversation you are in. The conversation keeps the stages it began on, so a change you just made may not show up in the next reply.
New conversation is the real reset. It starts a fresh room, so the chatbot opens from the beginning with your change in place.
The ring on the path board is the stage the conversation is in right now
If the playbook changed under a conversation that is still running, the pane tells you: The playbook changed since this conversation began — start a new one to test the change from the opening.
Working in Test mode
Test is the third segment of the Playbook tab’s own mode strip, so you can go back to it without the toolbar button.
One line of orientation sits at the top: Each reply says what produced it — open that stage to change it.
Switching to Build and back does not lose your conversation. The room stays open behind the editor, so the loop of send a message, read the chip, fix the stage, send another is one continuous conversation.
On a wide pane the whole path board is drawn beside the transcript with a ring on the current stage. As the pane narrows it collapses to just the path the conversation is on, and on a phone to a single line.
Sending files and voice notes
The composer takes attachments and recordings as well as text, which is how you test a chatbot that is meant to collect documents.
- Attach a photo, audio or file — the limit is 5 MB per file for a test message, which is lower than the knowledge base’s.
- Record a voice note records up to 60 seconds. While it runs, the row shows the elapsed time and Recording — tap stop to attach the clip., with Cancel beside it.
A staged attachment tells you what the chatbot will do with it before you send:
| What you attach | What it says |
|---|---|
| An image | Read for its printed text, then sent to the agent’s vision input |
| Audio | Transcribed, then answered |
| A PDF | Read at ingest, and sent to the agent as a file |
| Another readable document | Its text is extracted and read — the file itself can’t be opened |
| A video | Carried as a reference — no model can watch a video |
| Anything else | Carried as a reference — nothing here reads this file type |
If something goes wrong
| What you see | Why | What to do |
|---|---|---|
| Test is greyed out | No WhatsApp number is connected, or the settings are still loading | Connect a number; a test room needs a real one |
| Test opened a dialog and there is no mode strip | This chatbot has no playbook, so there is no editor to sit beside | Expected. Build a playbook if you want the side-by-side view |
| Connecting… in the message box, and it stays there | The connection to the test room has not come up | Wait a moment, then close and press Test again |
| Disconnected in red beside the title | The connection dropped mid-session | Close the dialog and press Test again |
| Couldn’t load your user identity — can’t open a test room without it. | We could not confirm who you are | Reload the page and try again |
| You fixed a stage and the next reply is unchanged | The conversation is still on the stages it started with | Press New conversation |
| No New conversation button | You are in the Test Chatbot dialog, which has no reset | Close it and use Test mode on the Playbook tab |
| That file is over the 5 MB limit for a test message. | Test attachments cap lower than knowledge base uploads | Send a smaller copy |
| Couldn’t read that file — try attaching it again. | The file could not be read from your device | Attach it again, or try a different file |
| A chip’s Edit stage → does nothing | That stage has since been deleted | Press New conversation and test the flow as it stands now |
| Replies are good in testing and wrong to customers | The chatbot is still Paused, or you are testing a different number | Check the number chip and the AI chatbot enabled switch — see AI Chatbot |