Synonyms: conversational agent, virtual assistant, conversational UI, AI assistant, bot
Do not index
Definition
A chatbot is a text-based conversational interface that helps users complete tasks by typing in natural language instead of navigating a UI. It ranges from rule-based bots that follow scripted flows to LLM-powered assistants that handle open-ended input.
In UX the work is less about the tech and more about conversational design and prompt design: setting scope, guiding the dialogue, and failing gracefully.
Use cases
When a bot hides what it can do and can't recover from a wrong turn, users bail on the first miss and carry a grudge into every bot they meet after.
The mystery box: A bot opens with "How can I help?" and no scope. Users guess, ask something out of range, hit a dead end, and leave.
The script trap: A support bot forces free text where a button would do, or refuses to accept "thanksgiving" as a date. Small rigidities read as the bot not listening.
The no-exit loop: A user needs a human, but the bot keeps looping clarifying questions with no visible handoff. Frustration spikes and trust craters.
How it's used in practice
State the scope up front: Open with what the bot does and 3 to 6 quick-start options. Users decide in seconds whether it's worth their time.
Offer hybrid input: Buttons for common paths, free text for everything else. Don't make people type what a tap could handle.
Design the failure states: Every bot misunderstands. Plan clarifying questions, a "start over," and an always-visible route to a human that carries the transcript.
Give it a consistent persona: A tone that matches your brand and holds steady. Say it's a bot; don't fake being human.
Test with real transcripts: Read actual logs. Users leave your happy path immediately, and the logs show where.
🪄
Pro-tip: The moment of truth is the miss, not the hit. Any bot looks fine on the happy path. What separates a trusted bot from an abandoned one is what happens the first time it doesn't understand. Design that moment first.
Challenges & limitations
The bad-bot bias: Years of useless support bots mean many users arrive already annoyed. You're designing against a grudge.
Discoverability: Chatbot entry points are inconsistent across the web, so people don't know where to look or that a bot is even there.
LLM unpredictability: Generative bots hallucinate, drift off-scope, and answer confidently when wrong. Guardrails and citations become part of the UX, not an afterthought.
