Guided chatbot vs AI bot: what's the difference and when to use each

Updated 2026-09-19 · 5 min read · By the Convo360 team

What a guided chatbot actually is

A guided chatbot is a decision tree. The customer sees a menu: buttons, a list, or a typed number, and every branch of the conversation was written by a human in advance. There is no language model involved: the bot matches the customer's tap or keyword to a pre-built path and sends the next pre-written message.

Because nothing is generated, a guided bot cannot go off-script. If a customer asks something outside the menu, the bot either repeats the menu, asks a clarifying question from a fixed list, or hands off to a human agent. That's a limitation and a feature at the same time, depending on what you're building.

Think of it as the WhatsApp equivalent of an IVR phone menu, 'press 1 for order status, press 2 for returns', except the customer taps buttons instead of dialling digits, and the flow can branch, collect information, and trigger actions like saving a lead or sending a payment link along the way.

What an AI bot adds

An AI bot uses a language model, connected via your own OpenAI, Anthropic Claude or Google Gemini key, to read the customer's message and generate a reply, usually grounded in a knowledge base of your documents or URLs (retrieval-augmented generation, or RAG). It can answer questions phrased in a dozen different ways, in Hindi, English or a mix of both, without you scripting each variant.

This matters most when the range of things a customer might ask is genuinely wide: pre-sales questions about a product's specifications, comparisons between variants, or 'does this suit my situation' style queries that no fixed menu can anticipate. An AI agent reads the actual question and searches the knowledge base for the relevant answer, rather than forcing the customer to navigate to the 'right' menu item first.

The trade-off is that a generated answer can occasionally be wrong or off-topic if the knowledge base is thin or the question is ambiguous. See our note on AI hallucination for how to reduce that risk before it reaches a customer.

Cost and risk, side by side

Neither is strictly better. A guided bot for a fixed process like order status or appointment booking is often more reliable than an AI bot, simply because there's less to get wrong: every path has already been walked and checked before it ever meets a real customer.

An AI bot earns its running cost where the alternative is a human repeating the same explanation in slightly different words all day, or where the menu would need dozens of branches to cover realistic questions.

The hybrid pattern most businesses actually need

In practice, the best-performing flows use both. A guided menu greets the customer and handles the majority of requests that are predictable, such as tracking an order, booking a visit or talking to sales, and only the remaining, harder questions go to an AI Reply block with a knowledge base behind it.

This keeps cost down, since AI tokens are only spent when genuinely needed rather than on every single message including 'hi' and 'thanks'. It also keeps the predictable paths bulletproof: a returning customer typing 'order status' should never get a creative, generated answer when a database lookup and a fixed template will do, while still letting you answer the long tail of questions you didn't think to script.

A common pattern for a retail or services business: guided buttons for the top four or five reasons customers write in, and an AI Reply fallback labelled something like 'something else' or triggered automatically when no menu option matches.

Building either on Convo360

Convo360 lets you build both types in the same builder, and switch between them as a flow grows. In the Flow Builder, a purely guided flow uses Buttons, List Menu and Condition blocks with a keyword trigger or new conversation trigger, with no AI agent attached, so it runs as a Guided (no-AI) bot and needs no AI key at all. You can still collect information with Ask a Question and Set Variable blocks, send a WhatsApp Flow native form for structured data, or trigger a Send Template, entirely without a language model.

To add AI, drop an AI Reply block at the point where the menu runs out of options, and connect it to an AI agent with a system prompt and an uploaded knowledge base (documents or URLs). You can mix Guided and AI-agent nodes freely in the same flow, then use Preview / Try-it to test both paths, including what happens when a customer types something completely unexpected, before publishing. See the Guided bots docs and Flow Builder docs for block-by-block details.

Example conversation

Frequently asked questions

Related reading

Available across India — see WhatsApp & AI calling automation by state and city.