What a chatbot is and how it works
A chatbot is software that exchanges messages with people automatically, on WhatsApp, website chat or another channel. It receives the customer’s message, works out what was asked and replies with an answer, a menu or an action, such as opening a ticket or booking a time slot.
What changes from one chatbot to another is how it decides the reply. Some follow a hand-drawn script. Others use artificial intelligence to understand what the customer wrote. The newest ones, AI agents, use generative AI to run the whole conversation toward a goal.
The 3 types of chatbots: three generations of customer service
Not everything sold as “AI” works the same way. Before choosing, separate the three generations:
| Generation | How it works | Where it gets stuck |
|---|---|---|
| Flow chatbot (rule-based) | A decision tree of menus, buttons and keywords, designed path by path | When the customer writes in their own words, sends a voice note or asks something off-script |
| AI chatbot (hybrid) | The same flow, with AI to detect intent or answer inside one block | It is still tied to the tree: every change means a new flow |
| Generative AI agent | Understands natural language, checks the company’s knowledge base and systems, and decides the next step | Only where you did not allow it: the agent follows the stages and criteria you set |
Many platforms use AI only to detect intent and stay locked into the flow they drew. The difference shows when the customer leaves the menu: the chatbot says “I didn’t understand”, and the agent answers from your knowledge base and gets back on track. Our chatbot vs AI agent page shows both conversations side by side.
1. Flow chatbot (rule-based)
This is the classic chatbot. Someone draws a decision tree: “Press 1 for Plans, 2 for Support, 3 to talk to an agent”. Each answer leads to a new menu, and every possible path has to be planned in advance.
- Understands: buttons, menu numbers and exact keywords.
- Replies with: the fixed text written in each block.
- Gets stuck when: the customer types a free sentence, sends a voice note, photo or PDF, or asks two things in one message.
- To change it: someone redesigns the flow and tests it path by path.
2. AI chatbot (hybrid)
This is the middle generation, and the one most often confused with an AI agent. The chatbot is still a flow, but it gets an AI layer to detect intent (“the customer wants a copy of the invoice”) or to write the reply for one specific block.
- Improves: understands sentences written in different ways, not just the keyword.
- Does not fix: once intent is detected, the conversation goes back on the flow’s tracks. If the customer asks something nobody designed, it still falls into “I didn’t understand”.
- To change it: you still edit the flow, now with intents and training examples.
3. Generative AI agent
An AI agent does not follow a tree. You explain in plain language what it does at each stage and when to move on, and it drives the conversation to the goal: answer the question, qualify, book the meeting or hand off to the team.
- Understands: natural language, including voice notes, photos, PDFs, stickers and links.
- Replies with: the company’s knowledge base (documents, pages and text) and data from its systems, such as the CRM and calendar.
- When the customer goes off-script: it answers and gets back to the stage goal.
- To change it: ask for the change in writing and approve it before it goes live.
Chatbot vs ChatGPT vs AI agent: what is the difference
All three talk by text, but they do different things:
| Flow chatbot | ChatGPT | AI agent | |
|---|---|---|---|
| What it is | A support bot with a fixed script | A general-purpose generative AI assistant | Generative AI trained on one company’s process |
| Knows your company? | Only what was written in the blocks | No, unless you explain it in every chat | Yes: the company’s knowledge base, rules and stages |
| Acts in your systems? | A little, through fixed integrations | Not by default | Yes: calendar, CRM, contracts and other systems |
| Serves the company’s customers? | Yes, on the configured channels | No: it is used by whoever opens the app | Yes, on WhatsApp and the company’s other channels |
In practice, an AI agent uses the same technology as ChatGPT, large language models, but with a company’s rules, knowledge and integrations. Our post on what an AI agent is explains it in detail.
What only an AI agent can do
The gap between the middle generation and an AI agent is not just the wording of the reply. It is what happens during the conversation:
- Uses the calendar and CRM mid-conversation: checks open slots, books the meeting on the rep’s Google Calendar and updates stage, fields and owner in HubSpot, Salesforce or RD Station.
- Switches topics without losing the customer: if a support chat turns into a sale, it moves to sales with history and summary, and the customer repeats nothing.
- Is the same on every channel: the same agent works on WhatsApp, LinkedIn, calls, SMS, RCS, email and website chat, with the same knowledge and rules.
- Follows up with people who went quiet: at the time you set, with a message tied to the context.
- Connects to the AI you already use: through MCP, the protocol that links assistants like Claude and ChatGPT to other platforms, you can build and improve the agent by chatting with your AI.
In Nuvia’s Smart Funnel, every stage has a plain-language instruction, and the agent moves the conversation forward once the stage criterion is met.
Gets better with every rated conversation
Mark conversations as good or bad. Agent Builder finds the pattern and proposes instruction changes that you approve before they apply.
Learns from your team
It compares the conversations your team handled well with the AI's and proposes bringing over the questions, the order and the tone.
Simulate before the customer
Up to 10 messages with sample data. The agent replies as it would for real, without talking to any customer.
Google Calendar
HubSpot
MCPIntegrations you ask for in writing
Calendar with availability checks, CRM, webhooks and calls to your system, set up in plain language and reviewed before publishing.
- Always a draftEverything published via MCP is a draft. You activate it in the platform.
- Lock on live agentsChanges to a live agent require your confirmation.
- Quality checkBlockers and warnings flagged before activation.
- MCP audit logEvery access and action is recorded, with user and result.
Nothing goes live without you
Any user can read and simulate. Creating and editing agents is limited to admins.
5 signs your “AI chatbot” is still a flow chatbot
Signs of a flow chatbot
- To change a reply, someone has to edit boxes and arrows in a builder.
- When the customer asks something off the menu, the reply is “I didn’t understand” or “press 1, 2 or 3”.
- Voice notes, photos and PDFs are not understood.
- Handoffs reach the team without a summary, and the customer has to explain everything again.
- Each channel has a different bot, with different rules and replies.
If your operation ticked two or more, the problem is not the wording of the replies: it is the generation of the technology.
When a flow chatbot still makes sense
A flow chatbot still works for short, fixed menus where the customer only picks between a few options, like confirming attendance or choosing a location. When the customer writes in their own words, sends voice notes, needs to be qualified or wants to buy, an AI agent does better. An AI agent also follows stages: the difference is that you do not have to draw every possible path in advance.
How to replace a flow chatbot with an AI agent
What your team already knows becomes the agent’s starting point. Your current chatbot script works too:
Gather what you have
Manuals, playbooks, FAQs and the script your team follows today.
Let Agent Builder set it up
It proposes funnels, stages and knowledge base from your material for you to review.
Connect calendar and CRM
Ask in writing: Google Calendar, HubSpot, Salesforce or RD Station.
Simulate and go live
Test conversations with fake data. Everything is a draft, and you decide when to activate.
With Agent Builder, you explain the process in Claude, ChatGPT, Gemini or Cursor, or upload your playbook, and get the agent set up in minutes to review and simulate.
At Chocolat du Jour, the AI agent handled 91% of conversations and served 25,000 customers in 8 days during the Easter peak, when WhatsApp revenue doubled.
Customer service and reactivation of a historical customer base. At the Easter peak, WhatsApp demand grew fivefold and the channel doubled its revenue, with quality in sales and after-sales.
Read the storyTo see how AI agents for WhatsApp support and sell, read the guide. If you are choosing a tool, our roundup of the best AI customer service software shows what to evaluate.
Frequently asked questions
What are the types of chatbots?
There are three: the flow chatbot, rule-based and built on a decision tree; the AI chatbot, a hybrid that uses AI to detect intent but follows the flow; and the generative AI agent, which understands the conversation and decides the next step.
What is a chatbot?
A chatbot is software that talks with people by message and replies automatically, on WhatsApp, website chat or another channel. It can follow a fixed script, use AI to understand intent or, in the case of an AI agent, run the whole conversation with generative AI.
What is the difference between ChatGPT and a chatbot?
ChatGPT is a general-purpose generative AI assistant that anyone uses in the app. A chatbot is a company’s support bot. When that bot uses the same technology as ChatGPT with the company’s rules and knowledge, it becomes an AI agent.
Is an AI chatbot the same as an AI agent?
No. Many chatbots use AI only to detect intent and stay locked into the flow they drew. An AI agent understands the conversation, answers from the company’s knowledge base and decides the next step without a decision tree.
What is a rule-based chatbot?
It is a chatbot that follows fixed rules: menus, buttons and keywords designed in advance. It works for short, predictable menus but gets stuck when the customer writes freely or asks something off-script.
What is a WhatsApp chatbot?
It is a chatbot that serves customers on the company’s WhatsApp, through Meta’s official API. Today the most effective option for support and sales is an AI agent on WhatsApp, which understands voice notes, qualifies leads and books meetings without a menu.
