Conversational UI
An interface where the primary interaction is dialogue — and a pattern applied far more often than it is warranted.
Design DisciplinesConversational UI suits open-ended tasks where the user's intent is hard to express in a form: asking something with many possible phrasings, refining a result across turns, working through an ambiguous problem. It is a poor fit for anything with a known, bounded set of options — a date, a plan, a status — where a form or a picker is faster, less error-prone, and does not require the user to guess what phrasing works.
The recurring mistake is replacing a working interface with a chat because chat is available. A booking form with four fields becomes six conversational turns, each with a chance of misunderstanding. The stronger pattern is hybrid: conversation for intent and disambiguation, conventional controls for constrained input, with the conversation summarising what has been captured so far.
The blank input is the first problem
A chat box with a blinking cursor gives no signal about what the system can do. Suggested starting points, visible capability boundaries and worked examples are what make a first message a good one — see prompt design.
In practice
An insurer replaced its four-field quote form with a chat flow. Completion fell from 68% to 41%: people typed dates in formats the parser rejected, and clarification turns doubled the length. A hybrid — chat to identify the product, then the original form — recovered completion and kept the discovery benefit.
Where teams get it wrong
- Replacing a working form with a conversation because chat is available.
- An empty input with no indication of what the system can do.
- Free-text entry for constrained values such as dates and plan names.
- No way to correct an earlier turn without starting over.
- Hiding captured state in the transcript instead of summarising it.
Learn more
You may ask
Frequently Asked Questions
When should you use a conversational UI?
For open-ended or ambiguous tasks where intent is hard to express in a form, and for refinement across turns. For bounded choices — dates, plans, statuses — conventional controls are faster and less error-prone.
Why do chatbots fail?
Usually because they replace an interface that worked. A four-field form becomes six turns with misunderstanding possible at each, and users are left guessing what phrasing the system accepts.
Related terms
All terms- Prompt DesignDesigning what a product asks a model, and what it asks the user — the interface layer around a prompt, not just the wording of one.
- AI Product DesignDesigning products whose behaviour is probabilistic — where the system can be confidently wrong and the interface has to account for it.
- Generative UIInterfaces assembled at runtime by a model, rather than laid out in advance by a designer.
- Form DesignDesigning the fields, sequence, validation and error handling of a form so people can finish it without stopping to think.
Defined by Mara Last reviewed .
Let's talk about your product.
Happy to look at what you're building and say where design would move the needle.
Contact Us