The difference between a chatbot and chatbot conversational AI decides whether your website chat books meetings or frustrates visitors. A classic chatbot follows a decision tree of buttons you drew; conversational AI reads what the visitor typed and answers from your knowledge. This guide answers a specific question: if your firm already runs a button-based bot, or is about to build one, how do you move to conversational AI on the levelpartner.ai platform without losing what the old flow did well?
Chatbots and conversational AI compared
A decision-tree chatbot is predictable. It asks "Are you a new or existing client?" and branches. It never says anything you did not write. That is its strength, and also why visitors abandon it: the moment their question is not on a button, the bot has nothing to say.
Conversational AI understands free text. A visitor can type "do you handle commercial property and how soon can someone call me" and get both parts answered. The cost is that you control the meaning, not the exact words. You manage that through instructions and a knowledge base rather than a flowchart.
What to keep from your old bot
Old flows contain useful decisions, even if the format is dated. Before switching, list:
- The qualifying questions the flow asked, and in what order.
- The branches that led to a booking or a hand-off.
- The answers to common questions written into the buttons.
- Where visitors dropped off, if your chat reports show it.
Those become the agent's prompt, its training content and its handover settings. Branches that ended with a person become Human Handover scenarios, and the old booking path becomes Appointment Booking on the same calendar.
Before you start
- An export or screenshots of your current bot's flow.
- Your website chat widget installed from Sites, or the code ready to add.
- A conversation agent in AI Agents, or a plan to create one.
- The calendar the old bot booked into, if any.
- A cut-over date and a person who reads every conversation in the first week.
How to set it up
- Go to AI Agents > Conversation AI and click + Create Bot.
- Select Prompt Based Bot.
- Choose Start from Scratch.
- Continue into the bot editor.
- Configure Bot Settings.
- Configure Bot Training with the old bot's canned answers.
- Configure Bot Goals using the old flow's goals and questions.
- Select the Deploy tab and turn on Chat Widget.
- Open the Chat Widget channel's edit icon and turn on Assign channel.
- Select the widget and add tag conditions if needed.
- Click Update, save the bot and test the widget on your site.
Run the agent in Suggestive mode while you replay the old bot's most common paths as test conversations, and switch the old flow off only when the agent reaches the same outcomes. Then move it to Auto-Pilot and compare bookings and abandonment for the four weeks either side of the switch.
When a decision tree is still better
Sometimes the old model wins. If every visitor must answer the same structured questions in a fixed order, such as an eligibility check, a form or a short survey from Sites is clearer than chat. If your legal team needs exact wording in every reply, a fixed flow is easier to approve. Use conversational AI where visitors arrive with varied questions; use structure where answers must be standard.
Say the old bot has nine branches and loses about two thirds of visitors before the final button. Suppose the replacement agent keeps the same four qualifying questions but accepts free text. If around half of chat visitors then complete qualification and a third of those book, that is roughly one booking for every six chats, where the old flow got only one visitor in three to the end at all.
Weekly time to run
Migration takes a day, most of it rewriting branches as instructions. In the first month, read every conversation for a week, then a sample of twenty a week. Old bots needed editing when your process changed; agents need the same, through the knowledge base.
Teams copy every branch of the old flow into the instructions as a script. The agent then behaves like the old bot with extra words. Keep the goals and questions, drop the script, and let the agent phrase things itself.
When neither is the answer
If your site gets very little traffic, a clear contact form and fast human reply may outperform any chat. If visitors mainly want a phone number, show it. For a fuller explanation of the underlying technology, read conversational AI. For widget installation and placement on your site, see chat bot for website, and for automated website chat set-ups, automated chatbot for website.
Troubleshooting
The agent skips qualifying questions. They are phrased as suggestions. State them as required before any booking.
Visitors still ask for buttons. Some prefer quick choices. Write a greeting that names two or three things visitors can ask, while still accepting free text.
Answers differ from the old bot. The knowledge base was rewritten loosely. Compare the old canned answers with the new entries and correct any changed facts.
Chat volume drops after the switch. The widget's placement or greeting changed. Keep the same position and a short opening line.
Frequently asked questions
Is chatbot conversational AI more expensive to run?
AI replies are billed as used, so each conversation has a small running cost that a button bot does not. The licence covers the feature. Most firms find the extra bookings outweigh it.
Can we keep something like buttons?
Partly. A greeting that names common questions helps visitors start, and the agent still understands anything typed. It is not a choice between buttons and free text.
Will the agent go off-script?
It stays within your instructions and knowledge base. It can phrase answers differently from what you wrote. Strict rules at the top of the instructions keep it inside the lines.
How long does migration take?
A typical flow moves over in a day, with a week of close review afterwards. Complex flows with many branches take longer to map, though the agent usually needs fewer rules than the old bot had branches. The chat bot for website guide mentioned above also walks through a first launch on a site that has never had chat.