I started working on voice assistants as a usability challenge in 2019 after one of the students in my User Experience (UX) Principles class asked, “I naturally speak in Spanish and English within the same sentence. Why can’t Google Assistant handle that?” This compelling question motivated me to explore the intersection of conversation design and user experience. I have been studying conversational UX ever since.
Conversational UX/conversation design entails writing, revising, and testing the dialog used by conversational agents that talk to humans. These agents range from those with a VUI (Alexa, Google, Siri) to chatbots (Woebot), embodied conversational agents (Replika), and the chat interfaces of large language models (ChatGPT, Claude, Gemini). Writing effective dialog or prompts for these systems is challenging, even if you already have experience in conversation analysis. However, experience dissecting conversations gives you an advantage, allowing you to start in the middle of the learning curve with knowledge about concepts like response variations and pacing.
Conversational UX made me wish I knew more about linguistic anthropology than was included in an introductory anthropology course, and my foray into writing for generative AI and chatbots has required me to expand my knowledge quickly. In my current ethnographic work on Conversational AI, I often look up concepts like “adjacency pairs” (which describes two connected turns in a conversation, such as “How are you?”/ “Fine”) just to understand workplace debates. Unlike me, many computer scientists and engineers I collaborate with have formal training in linguistics.
The job market for UX professionals is changing. Several applied anthropologist colleagues have suffered layoffs because their organizations undervalued their work. If you have lost your job in UX, you are probably rolling your eyes at an article about AI. But AI needs you. AI is not especially equitable or accessible; anthropology can help it do better.
Many linguists have discovered conversational UX, and linguistic and cultural anthropologists can also excel in this work. AI is less frightening if you are its voice, and conversational agents work better if the people creating them are sensitive to the relationship between culture and conversation.
(Some) Basics of Conversational UX:
- Persona crafting: Personas are sketches of someone or something’s personality and demographics, and bots need one too! A good bot persona helps you decide on the tone you want to take and the “voice” your conversational agent has. Use your user research skills to determine your audience’s needs. If you create a chatbot for a children’s learning website or app, your bot will speak differently than if it is a transactional agent on a banking website.
- Scriptwriting: This means planning out the two sides of a conversation. Flowcharts help define possible responses to user questions for conversational AI. I know a few conversation designers who started by drawing these in physical notebooks and have now moved to spreadsheets, Miro/FigJam boards, or specialized software as the technology has changed.
- Prompt writing: If you are working with a large language model like GPT, you will focus on prompts, which are commands you give the AI. For example: “Ask the user a question about the weather.” The more you test your conversations, the more detailed your prompts become.
- Testing: My favorite method for testing conversational agents is Wizard of Oz (WoZ) testing. The second is dialog improvisation. Much of the testing will be you and your colleagues running through dialog repeatedly and swearing at errors!
Learning:
Conversation Design Institute: Consider one of their courses, especially if your employer will pay for it. They are experts at conversation design training and do it well.
Software:
Voiceflow: Voiceflow is a great place to learn how to write a chatbot for fun. They have many helpful tutorials, which I use in my UX for AI class.
Conferences:
Anthropologists interested in conversational UX might find one of these conferences (and others) useful for networking, learning about new tools and methods, and how to break into the industry.
ACM: CUI (The Association for Computing Machinery’s Conversational User Interfaces Conference). This and CSCW (ACM’s Conference on Computer-Supported Cooperative Work and Social Computing) are great venues for catching up on significant new academic research.
South by Southwest (SXSW): This conference features excellent conversational AI and UX panels and is particularly suitable for tutorials and workshops.
Unparsed: While it is unclear whether they will have a conference in 2025, their site is worth bookmarking for the next one. Along with Agentic, this is one of the primary industry conferences.
Agentic: Originally called “Voice & AI,” this is another notable conversational and generative AI conference. Like Unparsed, attendees primarily work outside academia, so it’s an excellent space for workshops and case studies.
If you are still in school and like writing and technology, I encourage you to find a linguistic anthropology class. Courses are also available online, or you can ask your network for a syllabus to follow. Anthropology needs to make a case for its continued relevance, and this is a significant way you can contribute.
The post How can linguistic anthropology help you work with AI? appeared first on The Anthropology Career Readiness Network.