Use fixed flows when every conversation must go the same way: qualifying a lead, taking a booking, giving a legal notice. Use AI answers when customers ask many different questions and your help content already has the answers. Most teams end up with both: a flow that handles the known paths and hands the rest to an AI step or a person.
The short version
- Rule-based bots say exactly what you wrote. They are easy to check and often cheaper to run.
- AI bots cover the long tail of questions, but their answers vary and can be wrong.
- Several vendors price AI separately from scripted chats, so the choice shows up on the bill.
- Regulated or legally sensitive wording belongs in a fixed flow.
- Every builder we review supports some mix of the two. Check which plan unlocks the AI part.
How each one works
A rule-based chatbot is a flowchart. You draw the messages, buttons and conditions, and the bot moves along the path the visitor picks. An AI chatbot sends the visitor's question to a large language model (LLM), the kind of model behind ChatGPT, along with passages from your knowledge base, and the model writes the reply. The first is predictable by design. The second is flexible by design.
| Factor | Rule-based flow | AI answers |
|---|---|---|
| Predictability | Same input, same output | Wording and sometimes substance vary |
| Coverage | Only what you scripted | Anything your content covers, and sometimes things it does not |
| Upkeep | Edit the flow for every new path | Keep the knowledge base current |
| Checking before launch | Click through every branch | Test many sample questions; you cannot try them all |
| Running cost | Often unmetered or cheaper per chat | Often metered separately, per AI chat, outcome or credit |
When fixed flows win
Predictability. If a step must happen in order, such as collecting consent before an email address or asking three qualifying questions before booking, a flow guarantees it. You can read the whole bot and know what it will say.
Cost. Scripted messages are usually cheaper to run, and some vendors do not meter them at all. HubSpot's rule-based chatflows bot is not billed per conversation, while its AI agent costs US$0.50 per resolved conversation (HubSpot pricing). Chatfuel's wallet pays for AI replies, but keyword replies and other automation without AI are not metered (Chatfuel pricing). Landbot charges €0.05 for an extra chat and €0.10 for an extra AI chat (Landbot pricing).
Compliance. When the exact words matter, such as a disclosure that the visitor is talking to a bot, a consent notice or a regulated product statement, write them into a fixed step. An AI model may paraphrase or leave them out. Our guide to chatbots and the law covers the rules that apply.
When AI answers win
Many different questions. A support inbox gets hundreds of phrasings of dozens of questions. A flow needs a branch for each. An AI step can answer from the help centre you already maintain.
Content that changes. When prices, policies or stock change weekly, updating one knowledge base is easier than editing every branch. Some builders refresh sources on a schedule: Botpress and Landbot both document scheduled re-crawls of web pages.
Free-text input. Visitors rarely click the button you expected. An AI step can understand a typed question and still route it. What you must add is a clear exit: a handoff to a person when the AI is unsure. Our guide to writing a bot knowledge base covers how to give the model content it can answer from.
The hybrid most teams end up with
The builders we review mostly combine the two, and the documentation shows how:
- Landbot puts AI inside the flow as a block. The AI Agent block collects data into fields and exits back into the rule-based flow.
- Manychat's documentation says AI only answers when no other automation matched the message, so your scripted keywords run first.
- HubSpot lets a rule-based chatflow pass the chat to its AI agent with the Send to customer agent action.
- Intercom Fin pairs AI answers with the Workflows builder for routing and scripted paths, on Advanced and higher.
- Typebot is rule-based by default and adds AI as optional blocks that use your own provider account, so the AI bill comes from the provider, not from Typebot.
- Chatfuel is the exception: its documentation says new accounts get no button menus or decision trees, so it suits teams that want AI-first.
A common split: scripted paths for the few jobs that make money or carry risk (lead capture, bookings, anything legal), AI for open questions, and a person for everything the AI cannot answer with confidence.
A quick test for your own bot
- List the ten questions or jobs your bot will see most often.
- Mark each one: does the answer have to be word for word the same every time? Script those.
- Mark the ones that need live data, such as order status. Those need an action, which is the step from AI chatbot to AI agent.
- Give the rest to an AI step, and decide in writing when it must hand over to a person.
- Price both parts. Check whether the vendor meters AI chats separately in what counts as a conversation.
Frequently asked questions
Are rule-based chatbots outdated?
No. Rule-based flows are still the best tool for jobs that must run the same way every time, such as lead qualification, bookings and legal notices. The AI-focused products we review, including Intercom Fin and Tidio, still ship scripted flow builders next to their AI.
Is an AI chatbot more expensive than a rule-based one?
Often, because vendors meter AI separately. Landbot charges more for an extra AI chat than for an extra regular chat, HubSpot's rule-based chatflows are not metered while its AI agent is billed per resolution, and Chatfuel's wallet pays for AI replies but not for keyword replies. Check how your vendor counts AI use before you compare.
Can one chatbot use both fixed flows and AI?
Yes. Landbot and Typebot add AI as a block inside a flow, HubSpot can pass a chatflow to its AI agent, and Manychat's AI only answers when no scripted automation matched. A hybrid lets you keep exact wording where it matters and use AI for open questions.
Keep reading
- What is the difference between a chatbot and an AI agent?guide
- How is chatbot software priced?guide
- What counts as a conversation in chatbot pricing?guide
- How do you write a knowledge base for a chatbot?guide
- Best chatbot buildersbest list
Researched and drafted with AI assistance from the sources listed on this page. We have not built or tested bots on these platforms yet. Method: How we review