Chatbot Buttons vs Free Text and When to Use Each
A friend who runs a two person accountancy practice replaced her old button menu with a shiny new text box last spring, because the sales page told her customers prefer to type in their own words.

A friend who runs a two person accountancy practice replaced her old button menu with a shiny new text box last spring, because the sales page told her customers prefer to type in their own words. Six weeks later her bot was answering fewer questions than before. People opened the widget, looked at the blinking cursor, and closed it again. The ones who did type sent things like "tax" and "help pls".
The choice between chatbot buttons and free text input is not a choice between old and new. It is a choice about how much work you are asking a visitor to do in the first three seconds, and the answer changes depending on what they came for. Both belong in the same bot, in a specific order, and the order is what most people get wrong.
What each input style is good at
Buttons, sometimes called quick replies, do three things well. They tell the visitor what the bot knows, which removes the guesswork. They cost one tap on a phone instead of twenty seconds of thumb typing. And they produce clean data, because every tap maps to one intent with no spelling to interpret.
Free text does the opposite. It makes no promises about what the bot can handle, it costs effort, and it returns messy input. In exchange it handles the question you never anticipated, the one with a date and an order number and a complaint folded together. For a practice like my friend's, where half the questions are genuinely unusual, that matters.
The split I keep seeing in session logs is roughly the same everywhere: four or five intents cover 70 to 80 percent of conversations, and the long tail is everything else. Buttons are for the four or five. Free text is for the tail.
The numbers from three bots I watched
These came from a bike shop, the accountancy practice, and a small language school, all with traffic between 4,000 and 30,000 visits a month. The figures are rounded but the pattern held in all three.
| Question type | Buttons | Free text | Better choice |
|---|---|---|---|
| Opening hours, address, parking | Answered in one tap, near 100 percent resolved | Resolved about 80 percent, rest misread | Buttons |
| Order or booking status | Two taps then a form field | Works if the bot asks follow up questions | Buttons to start, text for the reference |
| Price or quote questions | Only if your prices fit a short list | Handles the "it depends" cases | Free text |
| Complaints and unusual problems | Frustrating, people pick the nearest wrong option | Captures the detail for the human who takes over | Free text |
| First message of a session | About 70 percent engaged | About 35 percent typed anything at all | Buttons |
That last row is the one worth pinning to the wall. Offering buttons on the opening turn roughly doubled the share of visitors who did anything at all. Nothing else in the configuration moved the number that much.
Why just let the AI handle it is a weak default
The common advice right now is that language models made menus obsolete. Anyone can type anything, so why constrain them? I disagree, and not because the models are bad at understanding. They are very good at it. The problem sits earlier than understanding.
An empty text box asks the visitor to invent a question, in writing, on a phone, while standing in a shop doorway. That is a small act of composition and plenty of people decline it. Buttons do the composing for them. They also set expectations honestly: a menu reading Opening hours, Book a fitting, Find my order tells someone in half a second that this bot will not discuss their warranty dispute, and they go to the contact page instead of wasting two minutes.
There is a quieter cost too. Free text sessions are expensive to read. Nobody on a small team has time to skim 900 unstructured transcripts a month, so the data goes unread, which means the bot never improves. Tap counts per button get read, because they are five numbers in a row.
The exception I accept: if your product is genuinely complicated and your traffic is low, say under 1,000 sessions a month, a text box with a strong model behind it will outperform a menu you cannot keep accurate.
The hybrid that works
Here is the arrangement I now set up by default, and the version my friend has been running since June.
Open with a short greeting and three to five buttons, with the text box visible and usable underneath. Do not hide it behind a tap. Visitors who know what they want type; everyone else picks. Our notes on writing a chatbot greeting message that gets visitors talking cover what that first line should say.
After the first turn, drop the menu unless the branch needs one. Mid conversation, buttons start to feel like a phone tree. Keep one persistent option, usually Talk to a person, pinned where it is always reachable.
When the bot fails to understand, do not simply apologise and wait. Show the three most common intents as buttons in the failure message. That single change took the accountancy bot from losing most confused visitors to recovering roughly half of them, and it is the approach we describe in our chatbot fallback message examples.
Writing buttons people actually tap
- Three to five options. Six is the ceiling. Past that the menu becomes a reading task and taps spread thin across the list.
- Start with a verb or a noun people use. Book a fitting beats Appointments. Where is my order beats Order enquiries.
- Keep labels under about 24 characters so they do not wrap on a narrow phone screen.
- Never include an Other button. It collects everything and tells you nothing. The text box is the Other button.
- Order by frequency, not by org chart. Put the question you get forty times a week first, even if it is the boring one.
- Rewrite the menu every quarter. Any option under 5 percent of taps is taking up space that a real question wants.
How to settle it for your own visitors
Run it for two weeks rather than arguing about it. Serve buttons on the opening turn to half your traffic and a bare text box to the other half, then compare three things: the share of opened widgets where the visitor sent anything, the share of conversations resolved without a human, and the number of sessions that ended within ten seconds.
Two weeks at a few thousand sessions is enough to see a difference of ten points or more. Anything smaller than that is noise and not worth rebuilding your flow for. Read the full transcripts of twenty sessions by hand as well, because the aggregate numbers will not show you the person who typed "cancel" four times. Everything else we have written about bots that get used sits in the chatbots section.
Frequently asked questions
Are chatbot buttons outdated?
No. They still win the opening turn by a wide margin because they cost one tap and show what the bot knows. What is outdated is a bot built only from buttons, where every path is a menu and nobody can type a sentence.
How many quick reply buttons should a chatbot show?
Three to five, ordered by how often the question comes up. Six is workable on a desktop and cramped on a phone. If you need more than six, you are describing a navigation menu rather than a conversation.
Should the text box be hidden when buttons are shown?
Leave it visible. Hiding it makes confident visitors hunt for a way to ask their real question, and a surprising share of them give up and leave instead of tapping a near miss.
Do buttons work for a chatbot in another language?
They work better than free text does, since a tap needs no spelling and no grammar. Translate the labels properly rather than running them through a machine, because a confusing button is worse than no button.
More in Chatbots
Chatbots
Chatbot Privacy and What Your Bot Should Not Store
I read fifty transcripts from a small physiotherapy clinic's website bot last spring, as a favour.
Order Tracking Chatbot for a Small Online Shop
If you run a small online shop, a large share of your support inbox is probably one question asked fifty ways: where is my...
Chatbots
Chatbot Analytics and the Metrics Worth Tracking Each Week
Your website chatbot has been live for a month, the dashboard says it handled 1,400 conversations, and you still cannot tell...