Chatbot
Build a rule-driven WhatsApp chatbot in Lynkist by drawing it — menus, questions, conditions and handovers on a visual canvas, published to your number when you're ready.
A Chatbot is a conversation you draw. You lay out the steps on a canvas — send a message, offer a menu, ask a question, branch on the answer, hand over to a person — and Lynkist walks each customer through it. It answers the predictable part of your inbox without a person: opening hours, order lookups, "which department do you need", booking a callback.
It lives under Chatbot in the sidebar.
Chatbot, AI Agent or Flow?
Lynkist has three automations and they are not interchangeable. Picking the wrong one is the most common reason a build goes badly.
| Use | When |
|---|---|
| Chatbot | You know the questions and the answers in advance, and you want a menu the customer taps through. Rule-driven and predictable. |
| AI Agent | Customers type free text and you want it answered from your knowledge base. |
| Flows | You need a form — several fields submitted at once, inside WhatsApp. |
They work together. A chatbot can hand a conversation to the AI agent, send a Flow, or hand over to a human — all as steps in the same drawing.
Step 1 — Create a chatbot
Go to Chatbot and create one. Give it a name you will recognise later; the customer never sees it. A new chatbot starts as a draft, which means it is not answering anyone yet.
Step 2 — Set the trigger
Every chatbot begins at its Start step, and the Start step decides who gets in. You set the words that open it — the trigger keywords.
A trigger matches whatever the customer types, whichever way they typed it:
- typing a word like
hiormenu - tapping an ice breaker on your WhatsApp profile
- tapping a command
- following a
wa.melink that has text prefilled
All four arrive as plain text, so one set of keywords covers them all.
If two published chatbots could both match the same message, the one with the higher priority wins. Set priorities deliberately when you have more than one bot on a number.
Step 3 — Draw the conversation
Drag steps onto the canvas and join them. The steps come in four families.
Send and carry on
The bot says something and moves straight to the next step.
- Send text — a plain message.
- Send media — an image, video, document or audio file.
- Send template — one of your approved templates.
- Send flow — open a Flow inside the chat.
Send and wait for an answer
The bot says something and stops until the customer replies. These are what make it a menu rather than a broadcast.
- Send buttons — up to a few tappable buttons.
- Send list — a longer tappable menu.
- Ask a question — the customer types a free-text answer and you keep it.
- Request location — the customer shares where they are.
Every waiting step needs two safety exits: one for no answer (the customer walked away) and a default for an answer you did not plan for. Lynkist will not let you publish without them — otherwise a customer can end up in a menu with nowhere to go.
Decide and remember
- Condition — branch on what you know: an answer given earlier, a contact field, a tag.
- Set attribute — save something onto the contact record, so the rest of the conversation and your team can use it.
- Set tags — tag the contact, which also makes them easy to target in a campaign.
- API request — call your own system, for example to look up an order.
- Delay — wait before the next step, so a burst of messages does not arrive all at once.
Finish
These end the bot's turn and carry no further steps.
- Hand over to a human — the thread lands in your shared inbox for an agent.
- Hand over to the AI agent — the AI Agent takes free-text questions from here.
- End — the conversation is complete.
Always give the customer a way to reach a person. "Talk to someone" as a button on your main menu is the single most valuable step in most chatbots.
Step 4 — Publish
A draft answers nobody. Publish puts the chatbot live on your WhatsApp number, and the trigger starts working straight away.
Publishing checks the drawing first and refuses anything that would strand a customer — a waiting step without its exits, a step nothing connects to, a menu option that leads nowhere. Fix what it reports and publish again.
A published chatbot has three states:
- Draft — being built; not answering.
- Published — live and answering.
- Disabled — switched off without being deleted, and without losing the drawing.
Editing a published chatbot does not change what customers get until you publish again. The live version and the version you are editing are deliberately separate, so a half-finished edit never reaches a customer.
Step 5 — Watch it work
Open a conversation in the shared inbox. While a chatbot is mid-conversation on a thread, the inbox says so by name, so an agent can see the bot is asking the questions rather than wondering why nobody has replied.
If a customer is stuck, replying by hand takes the thread back: the bot stands down and the person takes over.
Connecting your own systems
The API request step calls your system during a conversation — an order lookup, a stock check, a booking. Credentials are stored separately from the drawing, so the people who build chatbots do not need to see your secrets, and rotating a key does not mean editing every bot that uses it.
Common patterns
A front-desk menu. Trigger on hi. Send buttons: Sales, Support, Something else. Sales sets a
tag and hands over to a human; Support hands over to the AI agent; Something else asks a question
and then hands over.
Order status. Trigger on order. Ask a question for the order number, call your system with an
API request, branch on the result, and hand over to a human if the lookup fails.
Out of hours. Condition on the time, send text explaining when you open, set a tag so the morning shift can find everyone who wrote overnight.
Next steps
- Shared inbox — where handed-over conversations land.
- AI Agent — free-text answers from your knowledge base.
- Flows — in-chat forms a chatbot can open.