"Sorry, I didn't catch that." Of everything a chatbot says, that line probably gets read most often by people who are already annoyed. They typed something reasonable, the bot missed it, and the next sentence decides whether they try again or leave. Designing that moment well matters more than polishing the happy path, because the happy path mostly takes care of itself. The post belongs to remote conversation designer jobs, pays a $75,000 annual salary in USD for full-time work, and leaves the designer free to live in any part of the world. You would map what users want, write the dialogue that gets them there, build and test it, then keep fixing it based on what real people actually typed.
Every flow begins before any copy exists. Your first job is to work out the goals people bring to the assistant and turn them into intents the system can recognise. That means reading tickets and old transcripts, then grouping the messy ways people ask for the same thing. "Where's my parcel", "it never arrived" and a bare tracking number pasted with no greeting may all belong to one intent, or to three, and the choice changes everything downstream.
You will also decide what the bot should not attempt, since a narrow assistant that hands off gracefully beats a wide one that guesses. Part of the conversation designer job description here is drawing that boundary and defending it to product teams who would like the bot to do a little of everything.
Once intents are agreed, you script the dialogue: greetings, questions, confirmations, and the branches that handle every way a reply can go wrong. The fallback is where the craft shows. A good repair strategy changes on each attempt instead of repeating itself:
You will write slot-filling prompts that ask for one thing at a time, confirmation lines that read back what the bot understood, and error messages that admit fault without grovelling. Voice projects add constraints of their own: no buttons, and a listener who can hold only a few options in memory.
Persona work is less about a cute name and more about consistency. You will write a short persona guide covering how the assistant greets, apologises, jokes (if it jokes at all), and talks about money, health or account problems. Tone should flex with stakes. A playful line is fine when someone asks for opening hours. It is wrong when they report a stolen card.
Prototyping comes early and often. In Voiceflow you will build clickable conversation paths that stakeholders can actually talk to. Figma is where you lay out the full flow map, annotate edge cases, and show how chat bubbles, quick replies and cards will look inside the product. Being comfortable with both tools is part of the core conversation designer skills this role expects.
Here the design stops being a drawing. In Dialogflow CX you will set up pages, routes, intents with training phrases, entity types for things like dates or order numbers, and the event handlers that fire when the bot hears nothing or understands nothing. You need not be a backend developer, but you should understand how parameters move through a flow so fallback copy fires at the right point. Testing means throwing awkward inputs at the simulator: typos, two requests in one message, sarcasm.
After launch, the real users teach you what the first version got wrong. You will review conversation logs to find where people drop off, which intents get confused with each other, and which fallback messages show up far more than they should. Then you rewrite. Sometimes the fix is a training phrase; sometimes the prompt asked two things at once. Moderated user testing fills the gaps logs cannot show, like a tester hesitating over an ambiguous button label.
Many assistants now mix scripted flows with replies generated by a large language model, which shifts the work rather than removing it. You may write system prompts that set persona and limits, define guardrails for topics the assistant must decline, and decide when an answer has to be grounded in approved help content instead of improvised. Some moments, like identity checks or payment confirmations, should stay fully scripted. Testing a generative bot means probing it with adversarial and off-topic messages, then tightening instructions and retrieval sources when it wanders.
A background in UX writing, linguistics or design is the starting point, along with 2 to 4 years spent creating chatbot or voice experiences that real people used. Hands-on skill in Voiceflow, Dialogflow CX and Figma is expected, and so is excellent written English. A Conversation Design Institute certificate counts in your favour, though it does not replace shipped work. Whether a conversation designer course or certification is worth paying for depends on where you are coming from: a linguist with no product experience may gain a lot from the structure, while a UX writer who has already shipped bot copy may learn more by building a few prototypes.
Because the role asks for prior bot experience, it is not a fit for complete newcomers. Anyone searching for entry level remote conversation designer jobs should treat this one as a target to grow toward.
Your portfolio of sample flows matters more than your conversation designer resume. Reviewers want to see your reasoning behind the diagrams. Strong samples usually include:
Hypothetical projects are fine if labelled clearly; under NDA, rebuild the logic with a made-up business. Interview rounds for this kind of role often include a take-home exercise, so expect conversation designer interview questions about repair strategies, how you would measure a flow, and how you would push back on a bad request from product.
UX writers already know microcopy, voice guidelines and working with designers. Conversation design adds turn-taking, branching logic, intent modelling and the discipline of planning for misunderstanding. The UX writer polishes words on a screen; the conversation designer scripts an exchange where the other side can say anything. If you want to learn how to become a conversation designer from that starting point, rewrite a bot you use every day, prototype it in Voiceflow, and test it with five friends.
The demand comes from teams building AI assistants: software companies adding chat to their products, banks and telecoms running support bots, retailers, healthcare services, and studios making voice apps. As language models write more of the raw replies, the work tilts toward prompts, guardrails, evaluation and deciding where scripts still belong. That is a change in what the job involves, and nobody can promise how any one career will go, but someone still has to decide what a good conversation sounds like.
That $75,000 USD figure is the conversation designer salary per year for this position, paid as salary rather than per project. Apply with a short note, your resume and a link to two or three sample flows. Describe a fallback message you rewrote and what happened after. Applying costs nothing at any stage.