Conversational UI

A conversational UI lets people interact by writing or speaking in ordinary language, with the system responding in a running dialogue. Rule-based chatbots and voice assistants used it for years, but large language models made it the default way most people meet generative AI. Jakob Nielsen describes this shift as intent-based outcome specification: "the user tells the computer what outcome they want" rather than issuing step-by-step commands.

Why it matters

Chat is flexible. People can ask for things the designer never built a button for, combine requests, and refine an answer in follow-ups. It also needs no new visual vocabulary to learn.

The costs are real, though. Nielsen points out that prompting is a skill and that many people struggle to put their needs into detailed written prompts. A blank chat box also hides what the system can do. And long, chatty replies slow people down: in NN/g's research on site chatbots, the authors conclude that users want direct answers, not conversation, and quote one participant: "I just want to get the information."

How to apply it

Do:

• Say clearly what the assistant can help with. NN/g's chatbot guidelines say "the chatbot should clearly and concisely indicate what it can do," ideally tailored to the page the user is on.
• Lead with the answer. Cut greetings and filler like "Great question!" and use short sentences, lists and bold for scanning.
• Mix in GUI controls where they are faster than typing. NN/g's GenUI article highlights buttons, checkboxes and form fields inside chat, such as an assistant that asks clarifying questions as multiple-choice options.
• Offer suggested replies as buttons, and use progressive disclosure so detail expands inline instead of filling the thread.
• Be upfront that it is an AI. Google's PAIR guidebook calls "Disclosing the algorithm-powered nature of these kinds of interfaces" a critical onboarding step.

Don't:

• Use chat for tasks a form or a filter does better, like choosing a date or comparing ten products.
• Give the assistant a personality that implies abilities it lacks.
• Autoscroll to the end of a long response, which NN/g notes pushes users past the start of the message.

Common mistakes

• Chat as the only path. Forcing every task through a text box penalizes people who prefer to click, and anyone who doesn't know what to ask.
• Pretending to be human. Human-like phrasing raises expectations the system cannot meet, then trust drops when it fails.
• Losing the thread. No way to save or share useful answers, which NN/g lists among its chatbot guidelines.
• Vague deflections. When the assistant can't help, it should say so plainly and point to a real next step, such as a person.
• Walls of text. Long answers in a narrow chat panel are hard to read. Put the key fact first and offer more on request.