Conversational interfaces keep getting simpler to build and harder to build well. A chat UI in 2026 is a weekend project; a chat UI that feels natural and actually helps users is a months-long design effort. The models are good enough. The UX around them isn't. This post is the catalog of conversational UX patterns we've battle-tested across chat, voice, and hybrid surfaces.
The core tension
Chat feels like freedom — type anything, get a response. But pure chat is also the worst form factor for tasks with a lot of structure. Users end up typing long paragraphs to express what a form with three fields would capture instantly. Good conversational UX mixes chat with structured elements: quick-reply buttons, embedded forms, cards with actions, inline rich content. The UI speaks chat but offers structure for structured tasks.
Patterns that work
Quick-reply chips
After any question with a small set of likely answers, offer buttons. 'Which plan?' followed by 'Free / Pro / Enterprise' chips. User can still type, but tapping is faster and unambiguous. Dramatically reduces failed parse rates on common questions.
Inline forms for structured input
When the AI needs structured data (date, address, number), render a form inline in the chat rather than asking in free text. 'What date?' rendered as a date picker. Users fill it in seconds instead of typing and hoping the AI parses correctly.
Progress indicators for long operations
For actions that take >2 seconds, show a progress indicator with substantive status. Not 'thinking...' — 'searching knowledge base,' 'calling order API,' 'reviewing options.' Users tolerate long waits when they can see progress; they abandon silent ones.
Suggested follow-ups
After each AI response, offer 2-3 likely follow-up questions as chips. 'Tell me more,' 'Show an example,' 'Try a different approach.' This reduces the 'blank canvas' problem where users don't know what to ask next.
Cards with embedded actions
When the AI retrieves entities (orders, products, contacts), render them as cards with action buttons inline in the chat. 'Cancel order' button on the order card. Users can act without navigating away.
Clear edit and retry
Users typo, change their minds, want to rephrase. Every message should be editable; every AI response should have a regenerate option. Pretending conversations are linear is a usability bug — they're not.
Patterns that fail
- Infinite scroll without anchors. Long conversations get impossible to navigate. Provide section jumps or collapse old turns.
- "Thinking..." with no detail. Opaque waits feel broken. Show what the system is doing.
- Walls of text responses. Users scan, not read. Break responses into short paragraphs with clear structure.
- Mandatory chat for structured tasks. Forcing users to type what a form would capture better makes everyone slower.
- No hand-off option. When the AI can't help, the user needs a clear way to escalate, not to re-explain themselves three times.
Voice is different
Voice UX can't do quick-reply chips or inline forms. The patterns shift: shorter turns, explicit confirmation, audio cues for state changes, barge-in for interruption. See our dedicated voice AI post for the voice-specific patterns. The overlap with chat is smaller than people assume.
Testing conversational UX
Standard usability testing applies, with a twist: instead of watching users complete a prescribed task, watch them use the system for a real task of theirs. Note where they get stuck, what they re-type, when they abandon. Every real pain point is a UX pattern to fix.
Quantitative signals to track: turn count to task completion (shorter is usually better for transactional tasks), abandonment rate per turn, re-type rate (sign of failed parses), escalation rate. Weekly review of these signals catches issues eval infrastructure misses.
Accessibility
Chat UIs often fail accessibility. Screen readers struggle with streaming responses (text is announced as it appears, which is disorienting). Focus management is usually broken (new messages don't update focus correctly). Keyboard navigation is often incomplete. Build accessibility in from day one — ARIA live regions for streaming content, explicit focus management, full keyboard access to all actions.
Closing
Good conversational UX is the difference between a product users love and one they tolerate. The patterns are well-understood but rarely implemented consistently. Invest in them. The time you spend on structured elements, progress indicators, suggested follow-ups, and clean handoffs pays back many times over in user satisfaction and task completion. The AI is the easy part; the UX is where the work lives.