Tithi Mondal

Article7 min readUpdated

Chatbot UI Design: Building a Helpful AI Assistant Screen

Chatbot UI design explained for founders: what the first screen shows, how to handle wrong answers and when to hand over to a real person.

By Tithi Mondal, UI/UX Designer

Read inहिन्दीবাংলাमराठीதமிழ்తెలుగుગુજરાતી

Chatbot UI Design: Building a Helpful AI Assistant Screen: cover illustration

Imagine a tiffin service in Durgapur that adds a chat assistant to its ordering app. On day one a customer types, "My dabba did not come and I paid by UPI." The assistant replies with a long paragraph about delivery policy. The customer is still hungry and still angry. This is the problem that chatbot UI design has to solve.

Chatbot UI design is the design of the screen where a person types or speaks to an automated assistant and reads its replies. It covers the first screen, the message layout, the input box, the error cases and the exit to a human. Many founders think the work is a bubble layout. It is not. The bubbles take an afternoon. The decisions around them take weeks.

This article is about the interface decisions. For the wider question of what AI tools can do in design, see what AI design tools can and cannot do.

What should the first screen of an AI assistant show?

The first screen should say in one sentence what the assistant can help with, and give 3 or 4 tappable starting points. An empty box with a blinking cursor makes people freeze, because they do not know what is allowed.

For the tiffin service, the first screen could read: "I can help with your order, payment and delivery." Below it sit four buttons: "Track my order", "Payment problem", "Change my address" and "Talk to a person". Each button sends a ready-made message.

Starting points do three jobs:

  • They show the limits of the assistant, so users do not ask for things it cannot do.
  • They help users with low typing confidence, which matters in tier-2 and tier-3 cities where many people type slowly on mid-range Android phones.
  • They give you clean data about what people actually ask.

Keep the buttons short. Use the words your customers use, not your internal words. If your customers say "refund", do not write "reimbursement request".

How do you design AI features in an app without making everything a chat?

Use a chat window only when the user's need is open-ended. When the need is known, a normal screen with buttons and fields is faster and safer than a conversation.

Think about "Track my order". A customer does not want to chat about it. They want a map and a time. A chat reply can link to that tracking screen, but the tracking itself should be a designed screen. In the same way, an AI feature that fills a form should show the filled form for the user to check, not describe it in a message.

A simple test helps. Ask: could this task be done in two taps on a normal screen? If yes, build the screen, and let the assistant open it. Use chat for things like "I ordered the wrong plan, what are my options?" where the answer depends on the person.

Other AI features that are not chat at all:

  • A suggestion line under a text box, which the user accepts or ignores.
  • A summary at the top of a long document.
  • A smart search that understands spelling mistakes and Hinglish.

I cover how search results and filters should look in search and filter UX design.

How do you show that the assistant might be wrong?

Show the basis of an answer when you can, and write plain wording when the assistant is not sure. An assistant that sounds confident about a wrong refund rule creates a real problem for the business and a worse one for the customer.

Some patterns that work:

  • Name the basis. "Based on your order from 14 March" is better than a bare answer. The user can see which data the assistant used.
  • Allow a quick check. Put a small "Is this correct?" or a thumbs up and down under each answer. Keep it light, but read the results.
  • Use honest wording. "I am not sure about this. Do you want me to ask a person?" is better than a guess.
  • Never let the assistant promise money. A reply such as "Your refund is approved" should only appear if a real system approved it. Otherwise the assistant can create a promise your team did not make.

Remember that a language model produces text that sounds right. It does not know your refund rules unless you give them to it, and it can still get them wrong. This is a product risk, so the interface must be built for it.

When should a chatbot hand over to a person?

Hand over when the user asks for a person, when the assistant fails twice in a row, or when the topic involves money, safety or strong emotion. The handover button must be visible from the first screen, not hidden after five failed attempts.

A good handover has four parts:

  1. A clear button. Label it "Talk to a person". Do not hide it in a menu.
  2. A summary passed on. The person who takes over should see the earlier chat. Customers hate repeating their story.
  3. A time promise that is true. "A person will reply within 2 hours" only works if your team can keep that. If not, say "We reply from 10 am to 7 pm."
  4. A channel the user already uses. Many Indian customers prefer WhatsApp or a phone call. Offer to continue there, and tell the user before the switch happens.

During the handover, show a state such as "Waiting for a person" with the chat history above it. Do not leave a blank screen. I wrote about the screens around this in in-app chat and support screen design.

What makes a chat interface for customer support easy to use?

A support chat is easy to use when messages are short, long answers are broken into steps, and the user can always see what to do next. Support users are often stressed. They want the problem solved, not a conversation.

Here is a table of common elements and what to decide for each.

ElementDecision to make
Message lengthKeep replies to 2 or 3 short lines; link to a screen for more
Buttons in repliesUse them for choices like "Yes" or "No" and for common next steps
Input boxAllow typing, and add a photo option for damaged goods or screenshots
Waiting stateShow that the assistant is working, and set a time limit
LanguageLet the user switch language at the top of the chat
HistoryKeep past chats so the user can find an old answer

Language needs extra care. If your users write in Bengali, Hindi or a mix with English, test the assistant with real messages in those scripts. Check that the font renders properly and that replies are in the language the user wrote in. Read more in regional language app design in India.

What does a designer do that the AI model does not?

A designer decides what the assistant should refuse, what it should hand over and how it should sound. The model produces text. It does not know your business limits unless someone writes them down and tests them.

In my work on mobile apps, the screens that carry the most risk are the ones with money and trust, such as payments and verification. I would treat the assistant the same way. I would list the five or six most common user problems, design the exact path for each, and mark where a person must step in. Then I would write the opening message, the button labels and the error lines, and test them with a few real users before launch. For the wording side, see UX writing and microcopy for apps.

Here is a short checklist to take to your team:

  • Can a user tell what the assistant can do within 5 seconds?
  • Is the "Talk to a person" button visible on the first screen?
  • Does every money or account answer come from a real system, not from guessed text?
  • Is there a clear state for slow replies and for errors?
  • Has someone tested it in the languages your customers write in?

If you are planning an assistant inside an app and want the flows and screens designed before development starts, you can look at my mobile app design service or get in touch.

Frequently asked questions

How long should a chatbot reply be?

Keep replies to 2 or 3 short lines. If the answer needs more, link to a normal screen, such as an order tracking page, instead of writing a long paragraph in the chat.

Do I need an AI chatbot for my small business app?

Not always. If most questions are about order status or timings, clear screens and a WhatsApp link may work better. Add an assistant when users ask many different questions that your screens cannot answer.

What should a chatbot say when it does not know the answer?

It should say it is not sure and offer to pass the user to a person. A guess that sounds confident is worse than an honest admission.

Working on something like this?

Tell Tithi what you are planning. You get a clear scope, a fixed price and a timeline before any work begins. I reply to every enquiry within 24 hours.

Your details go only to Tithi and are used only to reply to you. No newsletters, no sharing.

Have a product in mind? Let us design it properly.

Tell me what you are building. I will reply with questions, a rough timeline and a quote.