How to get a
conversation design job

Conversation designers shape how people interact with chatbots, voice assistants and AI agents. The work covers understanding user goals, designing dialogue flows, writing prompts and responses with the right tone, planning for misunderstandings (repair), designing handoffs to humans, supporting multiple languages and accents, and analysing transcripts to improve. With large language models, the work increasingly includes system prompts, tool-use rules, guardrails and evaluation sets.

The pitfalls are well documented. Nielsen Norman Group's research on chatbots finds that users get frustrated when bots misunderstand, give generic answers or block access to humans.¹ Microsoft's human–AI guidelines stress making clear what the system can do, supporting efficient correction and making it easy to dismiss or escalate.² Conversations carry legal weight too: Air Canada was ordered to honour a refund policy its chatbot invented.³ There's no reliable India salary series for the "conversation designer" title, but related roles indicate the range — UX writers typically earn ₹14–16 lakh⁴ and UX designers ₹12.6–13.9 lakh, according to AmbitionBox.⁵

Air Canada held liable for its chatbot's answer

2024

Microsoft guidelines for human–AI interaction

18

Typical UX writer pay in India (AmbitionBox, small sample)

₹14–16 lakh

Typical UX designer pay in India (AmbitionBox)

₹12.6–13.9 lakh

Table: How conversation designers are assessed (editorial summary)

Table: Designing conversational repair and handoff (illustrative)

Table: How conversation design evolved (editorial summary)

• A dialogue flow including errors and handoffs
• Before/after transcripts with analysis
• A voice prompt set with confirmations
• An LLM scope and refusal specification
• Resolution or satisfaction results
• A verified conversational and voice UX certificate

Show resolution, not just containment: "Redesigned support bot repair and handoff from transcript analysis; resolution and satisfaction improved; repeat contacts fell."

"How do you measure a chatbot's success?"

Task resolution, satisfaction and repeat contacts — not containment alone.

"How would you design for misunderstanding?"

Confirm, offer likely options, rephrase, and escalate after limited attempts.

"What changes for voice?"

Shorter turns, fewer choices, explicit confirmations, and handling noise and interruptions.

"How do you design an LLM assistant?"

Define scope, tone, refusals, tools and handoff, then build test sets and review outputs regularly.

"How do you handle multiple languages?"

Design in the languages users actually use, test with native speakers and support mixed-language input.

• "How is the bot's success measured today?" — containment vs resolution.
• "Can I read real transcripts?" — essential for improvement.
• "How does handoff to agents work?" — key experience.
• "Who reviews AI responses for accuracy?" — risk control.

Table: Roles related to conversation design and typical pay in India (AmbitionBox, October 2026; indicative)

Table: Common conversation design tools (editorial guidance)

• UX writer or support specialist — writing and empathy.
• Conversation designer — flows and dialogue.
• Senior conversation designer — complex assistants and voice.
• Conversational AI lead — strategy, evaluation, guardrails.
• Head of AI experience — AI products across channels.

1. Week 1: Read transcripts — Analyse a bot you use; list failures.
2. Week 2: Design — Flow with repair and handoff.
3. Week 3: Voice and LLM — Voice prompts and an LLM scope spec.
4. Week 4: Prove it — Test with users; take the Conversational and Voice UX Certification.

Related: UX Writing for AI and AI Trust and Transparency. Guides: UX writing for AI, AI trust and transparency and how to build a chatbot.

• Transcript reviews — finding failure patterns.
• Writing — prompts, responses and system instructions.
• Flow design — happy paths and repair.
• Testing — with real phrasing and users.
• Partnering — with support teams and engineers.
• Measuring — resolution and satisfaction.

"Should a bot pretend to be human?"

No. Be clear it's automated; deception damages trust and may break rules.

"How do you design proactive messages?"

Only when timely and useful, with easy opt-out and clear purpose.

Table: A first-90-days plan (editorial guidance)

Start with the transcripts. They show what users really say and where the experience fails. Fix the biggest failure patterns first — usually repair and handoff — and measure resolution, not just containment. Then build a test set of real queries to check every change. Designers who make bots genuinely helpful earn trust from customers and support teams alike.

Transcript analysis is the most valuable habit in conversation design. Set aside time every week. Sample conversations across outcomes — resolved, escalated, abandoned — rather than reading only failures.

Tag each conversation with the user's goal, whether it was achieved, where it went wrong and why. Common failure types include misunderstood intents, missing content, confusing responses, loops, slow handoffs and wrong answers. After 50–100 conversations, patterns become clear and you can rank them by frequency and impact.

Look closely at phrasing. Users' exact words show which intents need more training examples or which prompts need clarification. Note moments of frustration — repetition, "agent", all caps — because they signal where trust is lost.

Respect privacy throughout: use tools that mask personal data, and follow your organisation's consent and retention rules. Share a short weekly summary with the team, including one quote that captures the main issue. It keeps everyone focused on real user experience.

Before large language models, conversation designers scripted most responses and trained intent classifiers on example phrases. Now LLMs generate responses on the fly, which removes much scripting work but adds new design problems. Teams have to decide what the assistant should never say, when it should use tools or look up data, how it cites sources and when it hands over to a person.

The job is moving towards system design and evaluation: writing system prompts, defining scope and refusals, building test sets of real user queries and reviewing outputs regularly. Designers also work with engineers on guardrails and with legal teams on risky topics. In interviews, show you understand both worlds. Give an example of a scripted flow you'd keep for reliability, such as payments, and an open-ended area where an LLM adds value.

• Optimising for containment alone.
• No error or repair design.
• Blocking access to humans.
• Long voice prompts with many options.
• Not reading real transcripts.

Chatbot usability findings come from Nielsen Norman Group; human–AI principles from Microsoft's HAX guidelines; the Air Canada case from MobileSyrup's report; and India pay for related roles from AmbitionBox (read October 2026; no dedicated conversation designer data). Repair examples, assessment formats and history are editorial. Science of Apps publishes the free test linked here. Last reviewed October 2026.

• Conversation design — designing dialogue between people and systems.
• Intent — what the user wants to achieve.
• Repair — recovering from misunderstanding.
• Containment — share of conversations not escalated to humans.
• Handoff — transferring a conversation to a human with context.