Helping bots talk: A conversation with Aubry Ellison

Aubry started as the first Conversation Designer on LinkedIn's Content Design team two years ago. She's drawn to the iterative and complex nature of building valuable conversation systems for users. With the rise of GAI, she's increasingly interested in the content problems to be solved to make quality conversations. Outside of work, she loves watching Bravo, being an aunt, and spending time in the mountains.

Queenie: To get started, can you tell us a little bit about your career journey: how did you start as a content designer? And then how did you make it over to conversation design, the specialty in which you currently work? 

Aubry: I started as a middle school teacher and got burnt out by that. Then, I went into academia and I got burnt out by that really quickly too. But while I was in grad school, I worked for a test prep company. I was helping students prepare for the GRE there. Honestly it was just nice working in tech. There were a bunch of benefits, people were nice. And that got me thinking about what my life could be if I had a job that paid well and treated me well. That was a big motivation. 

After grad school, I got whatever technical writing job I could find. It was writing contracts for the government. Then, I found this job at Meta that was focused on their community help center page. And I was like: ‘Hey, I used to be a teacher and I can write. I think I'm a good fit!’ That was my way into tech writing! 

From there, I went into conversation design. Facebook had a customer support bot. I wanted more variety, so I just volunteered for the customer support bot project. In a weird way, I did conversation design before content design!

Queenie: I love that—you were really clued into what you liked and didn’t, and let that lead the way. I know it's a very different job market now compared to when you were applying to your first content design job at Facebook, but I’m curious: did you just cold apply or did you engage in any networking to help you get your foot in the door?

Aubry: I did cold apply. I spent a lot of time looking at job descriptions and figuring out how to pitch myself.

I think that as writers, that's an advantage we have. We're good at communication, so if you’re coming to content design from outside the field, it’s important to nail your story: why are you qualified even though you maybe have no experience in that exact thing?

But yes, I did cold apply and I think there's just a lot of luck involved in that, you know? 

Queenie: Absolutely! You also said that you got interested in working on the customer support bot and just took that project up at Facebook. So before you made that foray into conversation design, did you do a lot of learning or upskilling in conversation design to set yourself up for success? Or did you just learn on the job?

Aubry: I stepped into it head first. And honestly, I’d recommend that same approach to other people who want to do conversation design too. If you're already working in content and you already know a lot about language, you’re going to be the most qualified person on these projects. There’s no doubt about that. A lot of people on the project come from project management backgrounds and such. 

I was very confident about that, about being qualified to write the content—and it was conversational content about the help center content I was already writing! So I was already a subject matter expert of sorts. After jumping into that project, I did a lot of research and I would also set up time with people at Meta who were already doing conversation design to learn from them and get their recommendations.

So, for anyone interested in this work, I think the easiest way to get into it is to volunteer for projects at the company you're already working at. 

Queenie: Do you have any advice for folks looking to move into conversation design apart from getting in and doing the thing? Do you have any resources you could point people to?

Aubry:

I really like the book Conversations with Things. I think that's my favorite book on conversation design. It really broke down the vocabulary for me and provided many ways of thinking about the work involved, such as designing personas and intent mapping and such. 

I also think it’s really important to learn about the tech involved, like natural language processing. I’d say it’s perhaps even more important than learning the design skills. You need to understand how to train for understanding the user and how to train towards better copy, especially now with natural language generation. So I would recommend that you really focus on getting to know the technology that you're working with. 

Queenie: Do you have any resources that people can use to start digging into that tech piece?

Aubry: Conversations with Things goes into natural language understanding, but not so much generation. I honestly don't have a great resource for natural language generation. I feel like it's so new and everyone's just collectively figuring it out. 

Some things you just have to learn by doing. But it also helps to read about how GPT-4 was created—the white papers are really interesting, I think. 

Queenie: That leads directly into my next question: how do you upskill? How do you keep up with the latest developments in conversation design?

Aubry: I really like to follow other conversation designers on LinkedIn and Medium and see what they're writing about. So many people are talking about conversation design these days, but I will say that not all of them are talking about it from a content perspective.

I do think it's important to create a network of people who are thinking about conversation design from a content perspective—Kathy Pearl is one of my favorites. She's been on Google Assistant for a long time and is now on Bard. She writes a lot and has a lot of talks that help me learn about the technology, but that also help me center myself by asking: ‘OK, what role can I play in this technology?’ 

Queenie: Let’s talk a little bit about the work you do as a conversation designer at LinkedIn. What does your day to day look like? 

Aubry: It’s very interesting because there are very few content people at LinkedIn. Right now, I'm really focused on setting up high-level conversational patterns that other people, like designers, can reuse. So that looks like meeting with lots of designers on different projects, understanding what patterns they're trying to set up, and then coming up with a format that’s ideal for them to use and that can also improve a lot of systems at once. 

But I think generally, the day-to-day of a conversation designer is centered on a lot of partnership—especially with technical partners. More than what a content designer typically does. It’s really about understanding the system that you're designing for and understanding how to do iterative training versus focusing on start-and-go projects, which is more in line with what content design does. 

Queenie: On the topic of partnerships: I feel like it's intuitive that the word people will be good at conversation design, but do you feel like your partners in other disciplines readily acknowledge and understand that? Or do you still feel like you're trying to convince people that you're the expert with words? 

Aubry: I think you deal with the same stuff as content design does, where people are like: ‘Oh, I know how to write, so I can just throw something in here.’ There’s a lot of negotiating around that. 

Another challenge I’ve faced at LinkedIn is that there’s such an overlap between conversation design and interaction design. As a result, you have to do a lot of work educating interaction designers that the interaction is inherently language and we need to be approaching it differently than the other interactions they’re used to designing. All this to say, I think there’s a lot of education that comes with conversation design. 

Queenie: What are your top tools that help you do the work of designing conversations?

Aubry: It really depends on where you're building. When I was at Meta, we had an out-of-the box system. So we didn't even work with engineers. I could just go in and iterate and train the system—that’s how I’d do intent training. 

At LinkedIn, we don't really have a lot of those tools, so we end up creating documentation for engineering. To that end, I think the most impactful documentation is creating conversational flows and diagrams to document and explain the logic. And then also creating sample conversational copy, which I think should just be done in a Word document. Partners can then go in and comment on that. It's similar to drafting long-form copy in a way, and sometimes it's just easier to close Figma and work off of a Word document.

If I'm working on something that has a lot of pre-written responses, then Excel is best. I know it’s not sexy, but it's the easiest way to store a bunch of responses. 

Queenie: At LinkedIn, did you get to do all of the work around defining the bot’s personality, its voice and tone—that kind of ground-up stuff—or did you just get in and write the conversation scripts? 

Aubry: When I first got to LinkedIn, I was specifically hired to work on the help center bot. They were excited to work with me and open to me setting up the persona, voice and tone. I really got to define the conversational structures pretty independently. 

Since then, it's really sort of exploded here at LinkedIn—we suddenly want conversation everywhere! And that's been a challenge. Just being one person with so many stakeholders working on these projects and trying to advocate for conversation design. There needs to be many types of stakeholders on these projects, but there also needs to be enough content folks in the room to really advocate for quality responses.

Queenie: Absolutely. Have you found any good ways to say that without saying that? Any good ways of resolving the friction that can emerge there—or at least managing it if not resolving it?

Aubry: I think being really proactive helps. Improvement tasks are a great way of doing that—taking other people's mocks, rewriting them, and articulating why the interaction needs to be fundamentally different.

That’s honestly how I got into conversation design at Meta. I got into the group doing this work, and at first they were like: ‘We got this.’ So I said: ‘OK, but can I just come to a Q&A?’ They said yes, and I gave them pages of feedback that made them think more deeply. And eventually, they said: ‘OK, you can work on this project with us!’

People really need to tangibly see how you can improve an interaction. Show them what you can do. Take what they’ve done, redo it, and ask them: ‘So, what do you think?’ 

Queenie: That’s helpful. At this point in your career, would you recommend that people strive to be generalists in content design—do the broad ‘menu of tasks’ such as information architecture, microcopy, vision work—or should they become specialists and dedicate themselves to a particular track within the field? What would help them future-proof themselves and also grow in their careers? Do you think it depends on seniority?

Aubry:

I’d say that people should be generalists! My opinion is that you should do as many different types of projects as are interesting to you. I guess if you really want to be a specialist, then pursue that, but I do think that we over-categorize ourselves as writers.

For example, I feel like I'm qualified to do longform technical copy or content design or conversation design. Specializing in one thing and only thinking about that for the rest of my career doesn't sound fun to me. 

Queenie: So, Aubry, you’re a remote employee at LinkedIn. I’m curious: do you think that has, in any way, impacted your career? Has it constrained your networking opportunities or your ability to grow in your career?

Aubry: I feel like content designers are the most pro-remote work people in tech. I personally don’t feel like this choice to be fully remote has negatively affected my career. 

Content designers are also so distributed. When I was at Meta, I was just running from conference room to conference room. I was never sitting with a group of people building a relationship with them and designing stuff together. That just didn't happen. I think that's something that we often think happened before COVID, but at big companies, it just really didn't. 

Queenie: What advice do you have for early to mid-career folks who are thinking ahead about their content design or conversation design careers?

Aubry:

I think my biggest piece of advice is: nobody’s going to care about your career as much as you do. And if you don't go in and tell your manager: ‘This is what I want to do, I want to get promoted, and this is the timeline I want to get promoted in,’ nobody’s proactively going to be thinking about all that for you. In tech, you have to be willing to evangelize yourself.

If you want to move up or get the project opportunities you want, go into your 1:1s and don’t be afraid to say: ‘I want to be on this project. I want to be promoted.’ 

Within the first six months of a new job, I always make sure to ask questions like: ‘What does promotion look like? What’s a reasonable time frame for me to be promoted? What do I need to do to ensure that I'm on track for a promotion within that time frame?’

It also helps to get really good at identifying what you’re good at. Then you can go in and say: ‘Hey, I'm really good at this. Where can I do this more?’ Or: ‘I really want to do this. Where can I do it more?’

Queenie: I hear you. It’s important to be your biggest advocate. Let’s take a bit of a macro lens now: What do you think the job market in conversation design is like these days?

Aubry: I think it's faring well. We saw this at LinkedIn, where there was this perception that GPT would make content plug-and-play, which is just deeply not the reality. In fact, it’s quite the opposite. You need more content people in order to truly do it well. And my prediction is that we’re going to see more layoffs, but when things start to fail because of a lack of content people, we're going to see them opening jobs back up. 

I think that the writing jobs that are going to open up are going to be centered on training LLMs, both for traditional UI copy and for conversational copy. 

Queenie: Can we dig into that a little more: what do you think the future of content design is?

Aubry: I think content design is going to be more iterative and focused on training AI. There’s also going to be a bigger focus on thoughtwork and principles, I think—developing principles to train the AI to do certain things, that type of work, rather than writing exact copy. 

I do believe that it’s going to look very different from: ‘I look at these designs and I write some copy and then the project's done.’ And yet, I do think that that'll always exist on some level. AI isn't just going to be out there labeling buttons perfectly, you know?