Chatbot Design Blueprint: Sketch Your Bot’s Menu Tree on Paper and Catch Problems Before You Build

Chatbot design goes wrong long before anyone opens a builder. It goes wrong when the menu is planned in someone’s head, grows one button at a time, and ends up six taps deep with branches that lead nowhere. This guide shows a paper-first method: list the reasons people contact you, draw the menu as a tree, test it against five rules, and only then build it. A free checker reads your tree and flags paths that are too deep, screens with too many buttons, dead ends and missing routes to a person.

๐Ÿ‘ค What the visitor seesVisitor
๐Ÿค– AI AgentHi! What can we help with?
Book a table
Menu and prices
Delivery
Talk to us

Illustrative: a top level with four clear choices.

โšก In short

  • Plan the menu on paper first. Building first means you design by accident.
  • Start from the real reasons people contact you, ranked by how often they come up.
  • Five rules: three taps deep at most, five buttons per screen at most, no dead ends, short labels, a person always one tap away.
  • Paste your tree into the free checker below and fix what it flags before you build anything.
Chatbot design in LiveChatBot: chat widget appearance settings with colour and welcome message
How it looks in LiveChatBot: chat widget appearance settings with colour and welcome message.

Why paper first

Builders make it easy to add, and hard to see the whole

Every bot builder lets you add a button, then another, then a sub-menu under it. Each step feels reasonable. The trouble is that you only ever see one screen at a time, so you never notice that “Delivery” now has three levels under it while “Returns” has none, or that a visitor who wants to change an order has to guess which of four buttons hides it.

A tree on paper (or in a plain text file) shows the whole shape at once. You can see which branches are bushy, which are thin, and where paths stop without an answer. You can show it to a colleague and ask “where would you tap to cancel an order?” in thirty seconds. You can rearrange it with an eraser instead of rebuilding screens.

This is the core of good chatbot design for a small business: the visible structure of choices matters more than clever wording or animations. A visitor who finds their answer in two taps will forgive a plain bot. A visitor lost in a maze won’t come back, however friendly the greeting.

The method below takes about an hour for a typical shop, clinic, institute or service business. You need your last month of messages (WhatsApp, email, phone notes or chat history), a pen and the checker further down this page.

Step 1

List the reasons people contact you, and count them

Go through the last 50 to 100 messages and write down the reason behind each one in two or three words: “delivery time”, “price of X”, “book appointment”, “change booking”, “refund”, “job enquiry”. Don’t sort yet; just tally.

Most businesses find that five to eight reasons cover four out of five messages. Those become your top-level buttons. The rest either sit under a “More” button or are left for free typing, where the keyword bot or AI answers from your saved knowledge.

Pay attention to how people phrase the reason, not how you’d phrase it internally. Customers say “where is my order”, not “order status enquiry”. They say “can I come today”, not “appointment availability”. Write the button labels in their words.

Also note which reasons need a person every time. A complaint about a damaged item, a custom quote or a medical question shouldn’t sit three levels deep behind menus. Those get a direct route to your team, and the bot’s job is only to collect the details first.

If you don’t have message history, ask whoever answers the phone to keep a tally for one week. Guessing the top reasons is the most common way a menu ends up built around what the owner wants to promote rather than what visitors want to know.

Step 2

Draw the tree, one line per button

Write each button on its own line and indent the choices that appear after tapping it. Finish every last step with an arrow and what happens.

  • Book a table โ†’ booking form
  • Menu and prices โ†’ link to menu page
  • Delivery
  • Delivery areas โ†’ reply with pin codes
  • Delivery charges โ†’ reply with charges
  • Opening hours โ†’ reply with hours
  • Talk to us โ†’ person
  • Book an appointment โ†’ booking link
  • Timings and doctors โ†’ reply with timings
  • Reports
  • Collect a report โ†’ reply with steps
  • Report not received โ†’ person
  • Fees โ†’ reply with fees
  • Talk to the front desk โ†’ person
  • Courses and fees โ†’ link to courses page
  • Batch timings โ†’ reply with timings
  • Free demo class โ†’ form
  • Already a student
  • Change batch โ†’ person
  • Fee receipt โ†’ reply with steps
  • Talk to a counsellor โ†’ person

Step 3

Test the tree against five rules

Once the tree is on paper, check it against five rules. These chatbot design rules come from how people use chat on a phone: one thumb, a small screen, limited patience.

1. Three taps deep at most. Every extra level loses people. If a path needs four taps, either merge two levels or replace the deepest branch with a direct answer. “Weekend โ†’ Dinner โ†’ Party size โ†’ Time” is a form, not a menu; ask those details in one message instead.

2. Five buttons per screen at most. More than five and people read slowly, scroll, or pick the wrong one. If the top level has eight reasons, keep the five most frequent and put the rest under “Something else”.

3. No dead ends. Every final button must lead somewhere: an answer, a link, a form or a person. A button that just shows the next menu again, or nothing, is the single most annoying thing a bot can do.

4. Short labels. Two or three words, under about 28 characters, so they don’t wrap onto two lines on a phone. “Party orders and catering for large groups” becomes “Party orders”.

5. A person one tap away. Somewhere on every screen, or at the end of every answer, offer “Talk to us”. Visitors who can always escape trust the menu more, not less.

The checker below applies all five rules to the tree you type, so you don’t have to count by hand.

Free tool

Check your menu tree before you build it

Type your tree with two spaces per level and an arrow at each end point. The checker counts depth, buttons per screen and dead ends, and lists what to fix.

Type your bot’s menu as an indented list, two spaces per level. End every last step with -> and what happens (an answer, a link, a person, a form).

Runs only in your browser; nothing is sent or saved.

Reading the results

What to do with each warning

Too deep. Look at the deepest path and ask whether each level is a real choice or just a question the bot could ask in words. Often the bottom two levels collapse into one message: “Which day and time, and how many people?” Then let the visitor type.

Too many buttons. Sort the buttons on that screen by how often the reason came up in Step 1. Keep the top four, add “Something else”, and move the rest beneath it. Don’t hide anything that came up in more than one message in ten.

Dead ends. Usually these are buttons someone added “for later”. Either write the answer now or delete the button. An honest “Talk to us” is better than a placeholder.

Long labels. Cut adjectives and filler. “Do you deliver to my area” becomes “Delivery areas”. If you can’t shorten a label without losing meaning, it may be two different buttons.

Duplicate labels. Two “Prices” buttons in different branches confuse people who tap back and forth. Rename one (“Party prices”) or merge them.

No route to a person. Add it at the top level at minimum. In LiveChatBot, a visitor can always type a message, and your team sees every chat in the shared inbox, so the button just makes that obvious.

After fixing, run the checker again. Most trees need two or three rounds before they come back clean.

Step 4

Write the answers at the end of every branch

A tree is only the skeleton. Each end point needs an answer that does three jobs: gives the fact, mentions any condition, and offers the next step. “We deliver within 5 km of Koregaon Park for โ‚น40, free above โ‚น500. Want to order now or see the menu?”

Keep answers short enough to read on a phone without scrolling: two to four sentences. Link to the full page for detail rather than pasting it in.

Write the answers as saved replies so your team can reuse the same wording when they answer by hand. LiveChatBot’s free keyword bot also answers from saved replies when someone types instead of tapping, so one set of answers covers both. The saved replies guide explains how to set them up, and you can use {name} to greet returning visitors by name.

If you plan to use AI replies, the same answers become the knowledge the AI Agent works from. Clear, specific end-point answers are what make AI replies accurate; see how to train your AI chatbot.

Finally, read every answer aloud. If it sounds like a policy document, rewrite it the way you’d say it at the counter.

Before and after

Three real-world shapes the five rules fix

The maze
๐Ÿšฉ BeforeBook โ†’ Location โ†’ Branch โ†’ Date โ†’ Time slot โ†’ Confirm (six taps before anything happens)
โœ… AfterBook โ†’ booking form that asks branch, date and time on one screen (two taps)
The wall of buttons
๐Ÿšฉ BeforeNine top-level buttons including Careers, Blog, About us and Franchise
โœ… AfterFour buttons from the top reasons plus Something else, with the rest under it
The loop
๐Ÿšฉ BeforeHelp โ†’ Other questions โ†’ Help (the same menu again)
โœ… AfterHelp โ†’ Other questions โ†’ a short answer plus Talk to us

Step 5

Build it, then watch real visitors for two weeks

Now open the builder. In LiveChatBot, menus and proactive messages are built with Flows, and the step-by-step flow guide shows each block. Because the tree is already settled, building takes minutes rather than an afternoon of trial and error.

Decide where the menu appears. A greeting with buttons suits the home and contact pages. On a product or pricing page, a single targeted question often works better than a full menu. Flows can show different messages by page, traffic source, country, device, browser language, number of visits, lead score or whether you already have the visitor’s contact details.

Then watch. For two weeks, read every chat and note three things: where people stopped tapping and typed instead (the button they wanted didn’t exist), which buttons nobody tapped (candidates for removal), and which answers led to “that didn’t help” or a request for a person (answers to rewrite).

Good chatbot design is never finished on day one. The paper tree gets you a sensible first version; real conversations tell you what to change. Redraw the tree, run it through the checker again, and update the flow. A monthly fifteen-minute review keeps the menu matched to what people actually ask.

Quick reference

The five rules on one card

๐Ÿชœ

Depth

Three taps or fewer from greeting to answer.

โ†”๏ธ

Width

Five buttons or fewer on any one screen.

๐Ÿ

Endings

Every path ends in an answer, link, form or person.

๐Ÿท๏ธ

Labels

Two or three words in the customer’s language.

๐Ÿ™‹

Escape

A route to a person on every screen.

๐Ÿ”

Review

Redraw the tree monthly from real chats.

Frequently asked questions

What is the first step in chatbot design for a small business?

List the reasons people contacted you in the last month and count them. The five to eight most common reasons become your top-level buttons, written in your customers’ own words.

How many buttons should a chatbot menu have?

Five or fewer per screen. If you have more reasons, keep the most frequent four, add a Something else button, and put the rest under it.

How deep can a bot menu go?

Three taps from the greeting to an answer is a good limit. Deeper paths usually mean you’re asking form questions through buttons; ask them in one message instead.

Should a menu bot let people type?

Yes. Some visitors will always type. In LiveChatBot the free keyword bot answers typed questions from your saved replies, and your team sees every chat in the inbox.

Do I need the AI Agent if I have a good menu?

Not necessarily. A clear menu with good answers handles most routine questions. AI replies help with free-form questions your menu doesn’t cover, using the knowledge you give it.

Draw it, check it, then build it

Sketch your menu tree, fix what the checker flags, and build it in minutes with Flows. Start free on livechatbot.in.