How to build an AI content system

Key takeaways

• A Claude plugin is a collection of skills, and the EdPlus team uses theirs to draft emails, blogs and ASU Online degree pages with brand, style and strategy built in.
• AI gives the team more time for strategic, creative projects, with a human in the loop for quality checks.
• You don’t need to be a coder or an AI expert to start. You just start asking questions.

In the first episode of Behind the Build, host Auryan Ratliff talks with Stephanie Morse about how her team built an AI content system from Claude skills. Stephanie runs a live demo, drafting an email and a full ASU Online degree page, and explains how the system frees her team up for higher-impact strategic work. She also shares how anyone can start building with AI, no coding required.

Watch on YouTube now or read Stephanie’s blog here.

Behind the Build: Season 1, Episode 1

Full transcript

Stephanie: I think that there's a big misconception with AI, and honestly, it's fair.

Auryan: When you talk about bringing AI into your workflows, it can sometimes be, "Well, AI's going to do all my work. What am I going to do?" I think we're not just getting text back from Claude, but it's actually building a whole interface for us to work with.

Stephanie: We can focus on the projects we love most and think will have the most impact.

Auryan: That's really exciting to hear. It feels more empowering than it was handing off your power to the machine to do it for you.

Auryan: Hello everyone, and welcome to the first episode of Behind the Build. My name is Auryan Ratliff. I'm the director for technology innovation R&D here at EdPlus, and I will be your host for this series. Behind the Build is a new initiative we are starting up to highlight and share more of the ways that we're using AI, not only to transform the way that we work, but to transform the student experience here at ASU, so we couldn't be more excited for this series.

I am so happy to announce our first guest for our first episode, Stephanie Morse. She is the assistant director for AI content systems and strategy here at EdPlus, and they have been cooking up some great work over in that department around how they're leveraging AI in their workflows. And so I will maybe hand it over to Stephanie. 

Stephanie, hello and welcome. Can you tell us a little bit more about your role and maybe some of the areas you all have been experimenting with AI?

Stephanie: For my role, as you said, I'm the assistant director of AI content strategy and systems. So on my team, I have content strategists who focus more on the content generation aspect and content creation, content strategy, writing, but then also I have people focused on content systems. So there's the internal systems piece and then there's the actual content creation, content strategy part, which is probably the external part that a lot of people see. And we're using AI to help with almost all of that.

Auryan: Wow, okay. So it sounds like you all are working — content is almost like a reductive term. You guys are doing content production across every aspect of ASU Online. You guys are doing a little bit of product design. You're doing a little bit of marketing. 

You all are doing comms to students. There are a lot of different areas where you all have to work, and so I'm excited to hear how you were actually able to build a system that maybe accommodates all those different ways of working. Can you tell us a little bit more about your AI content system?

Stephanie: Technically, it is a plugin, which is just a collection of Claude skills, and there are five main parts of the system. So there's the content-generation piece, which helps us get that first draft. There is the brand style that's baked in. There is also the specific style for the specific skills, for how a blog is different than an email, and that's baked into those specific skills. There's also the strategy piece and how we're baking in content strategy into each of the skills. And then finally, we have a piece for the governance and iteration of the skill to make sure that we can easily keep it updated and that we can easily iterate on it over time.

Auryan: Oh, I love that. Okay, I’ve got maybe a silly question. Can you tell me more? What is a skill?

Stephanie: A skill is basically a set of instructions that you give Claude, that you want to train it to follow those instructions over and over and over again.

Auryan: That's very nice. And you keep mentioning Claude, which I love, but I also want to clarify: a lot of what you all built actually works across many agentic platforms. So something like this could fit into Codex or Antigravity as well, correct?

Stephanie: We have experimented with Claude the most, but we've definitely done some things in ChatGPT. We've been trying to experiment more recently with using it in Codex, and even trying different parts of it in Slackbot. And then our knowledge base is in Guru, so even experimenting with some of their AI systems. So really anywhere that there's skills functionality, you can definitely adapt this or parts of this system.

Auryan: Platform agnostic, very flexible. That is great to hear. I mean, it sounds like it is a very robust system, lots of parts, lots of people working in there. How do you even get started trying to build something like this?

Stephanie: I think that there's a big misconception with AI, and honestly, it's fair because a lot of previous technologies, you had to learn the technology. I need to get in here. I need to read a manual. I need to learn everything. But I think one thing that's amazing with AI is you actually don't need to do that. You don't need to become an expert before you can start. You don't need to be a developer or a coder. You can just go in and start asking questions.

Auryan: Oh, that is so approachable. I think when people are thinking about these AI systems, it can feel a little daunting. It feels like every day there's a new feature, and you're like, "How do I keep up?" But I love your answer. In reality, you just start with a conversation, and this tool, I think what's really unique about AI, is it can teach you how to use AI in a way other tools can't. 

Stephanie: Yes.

Auryan: And so, it's really not that hard to get started, and you could build amazing, amazing things working that way. So that is so exciting to hear. Can you talk about where this first started? What was the first iteration of this product?

Stephanie: Yes, so we did build a custom GPT in ChatGPT, and I think that was one of the first things that really gave us the leap and the confidence to start imagining something like this. Once we started to experiment with Claude, it was the agentic capabilities, where it could do a lot more than the custom GPT.

Auryan: Sounds like ChatGPT was a perfect place to build this proof of concept. Hey, there's something here. AI can help with this process. You move to something like Claude, way more agentic capabilities, able to build much larger workflows, and so that sounds like a really natural transition. You know, this was all great, but what problems were you even trying to solve for your team?

Stephanie: So I think there were really a couple things that we were really looking at. So first was just the balance of the work that we have. So one thing that we always say with our team is we're unique. There are other teams like this, but I think we're unique in that we are a service team, and we're a strategy team. So I think one big part of it was making more time and space for the strategy and the creative projects that we really think will have the most impact, while still making sure that we can support all the other strategies and requests.

Auryan: You know, that is really exciting to hear. I think the narrative when you talk about bringing AI into your workflows can sometimes be, "Well, AI's going to do all my work. What am I going to do?" But it sounds like AI freed you all up to do more strategic human work that you weren't able to do before because you were getting bogged down with all these tedious tasks. And so, that's really exciting to hear. It feels more empowering than handing off your power to the machine to do it for you. Would you agree with that?

Stephanie: Yes. That's really been the goal of how we've approached AI — I love how you say empower — is to empower the team. So one, to empower us to scale our work and be more efficient. So yeah, we can make some of those tedious tasks faster or automate them, whatever the case may be. We can focus on the projects we love most and think will have the most impact. 

Auryan: Okay, so we've got a system, you all had a goal, you set out, you built all of this incredible tooling and all of these incredible processes, but what does it actually look like to be working with this plugin inside of Claude? Can you maybe take us in there and give us a sense of what it's like to work with the plugin?

Stephanie: Yes. So we are in Cowork. That's where we use our skills. I believe you can also use them in Claude Code as well, but we're going to start by doing /email, and you'll see here that it's calling all of our different skills that have email in the name. And for us, we know we're going to do email, we're going to select this email skill, but we have a few different email skills because some emails have very specific templates, so we pull those out into specific skills. 

Auryan: Got it.

Stephanie: But if you are like, "Oh my gosh, 23 skills, like that's super overwhelming," you can also just say, "Hey Claude, I need an email, blah, blah, blah, blah, blah," and you don't have to do the /email, and Claude should recognize, "Email. I need to use the email skill." And you don't have to do that part. For us, we know what skill we want to use, and you can do that if you want.

Auryan: Perfect. So if you know what you want, you can just go in and use it.

Stephanie: Yes.

Auryan: If you don't, Claude is always there to support.

Stephanie: Yes. So what is our email going to be about?

Auryan: Oh, that is a great question. Maybe we have a new club, a new Claude Club at ASU, and we're inviting new students to come and join our club and learn about agentic workflows and how they can integrate AI into how they're pursuing their studies at ASU.

Stephanie: Okay. Awesome. Well, let me summarize that.

Auryan: I'm already feeling at ease as you're typing this. You know, when you think about these AI systems, you showed all the skills. You're almost like, "Whoa, there's a lot to learn here." But in reality, you're just articulating what you want. And the standards take the hard part out of it.

Stephanie: Okay, so I have that we have a new club at ASU Online about Claude and how to use Claude for agentic workflows, and we want to invite new students to join our Claude Club via email.

Auryan: All right. Looks great to me.

Stephanie: Okay, let's go for it. And let's see how it goes since we only provided a few sentences of info.

Auryan: Mm-hmm. And I will already call out, I'm actually pretty surprised with the initial response. It looks like we're not just getting text back from Claude, but it's actually building a whole interface for us to work with. Yeah, walk me through what I'm seeing here.

Stephanie: Yeah, so this is a little bit of what I was talking about earlier, where the content strategy's built in. So we didn't provide a brief. We only provided a few sentences of detail. And so Claude is saying, "Hey, I don't think I have quite enough right now." 

And so maybe, unlike some other AI systems that might be like, "Well, I'm just going to go with this and take a shot," Claude is pausing, as we trained it to do, to say, "Hey, I need a little bit more info. I need a little bit more strategy here to really draft something robust, that's of quality, that's going to meet the standards that we have here for our content."

Auryan: That is incredible. Is that something you all put in there, or just naturally, Claude's doing that?

Stephanie: Well, I do think Claude naturally does that on its own, but we have trained the skills to specifically do that. And we have trained the skills on some of the different questions to ask. Who are we inviting to the club? We have a few options here: incoming new students, prospective students, current students.

Auryan: I think this is going to be for current students.

Stephanie: Okay. Okay, perfect.

Auryan: We start with something that is Zoom only. This is going to be for ASU Online students. We want everyone to be able to join, no matter where they're at. Something we'll do on Fridays, and we'll just say it's at 2 p.m. MST. And if it helps at all, we can also say it's all about learning the basics of Claude, talking about agentic workflows and sharing how maybe folks who are a part of the club are already using their tools to get other people excited.

Stephanie: Okay, so what should a reader do after reading this email?

Auryan: I think we need a proper call to action. Send them to an RSVP link where they can sign up for the event and actually get the Zoom link sent to them.

Stephanie: Okay. And then tone direction. Do we want enthusiastic and casual, warm and conversational, professional, or maybe something else? 

Auryan: I think we do enthusiastic and casual. That is my vibe.

Stephanie: Yeah, I agree. I think that fits for this.

Auryan: I love that. I mean, it already feels like such a curated experience. I love that there were areas for me to provide the context where it made sense, but then other areas it was just a really nice UI to choose from options that had already been defined, I'm assuming by your standards. 

Stephanie: Yes. 

Auryan: And so it is a very intuitive process for someone who's never done this before that all felt very approachable. 

Stephanie: So if anyone was watching, you would've seen that it did read like the brand guide, the brand context and then we have our email here. And I think that is also creating a Word document for us. That is another thing that's built into all of our skills. So we use Google Drive for all of our files, and any content we create goes in there. We review in Google Drive, Google Docs, all of that. So we always have it create a doc for us. So then we can start editing in there, put it in the proper place, start commenting, sharing, all of that good stuff. 

So it integrates into the rest of the places that we work. So we have our email here, and overall, I think it's looking pretty good. And I think this is what we're used to. That we get a really good first draft. We bring that human in the loop. We maybe make a few tweaks, and then off it goes.

Auryan: All right, we got an output. How's this look?

Stephanie: I think we're looking good. I think it definitely hit on a lot of what we shared. So it has the messaging. You don't need to be a coder or an AI expert. It says that you're hosting. Very excited for you and your new club.

Auryan: That is incredible to see, and just the ways that you'd actually want to work with a tool like this. So you showed us an email. It looked fantastic. I love the experience. An email's probably one of the shorter pieces of content you'd produce with a system like this. I'm wondering if you could show us what's one of your more complicated workflows or outputs that you would produce with the system? 

Stephanie: Yes. Yeah, I would love to. So we have our program page content which is a much longer workflow. So let's dive right in, and I can show you how Claude really takes us through this whole workflow on our end, and show you some of the more agentic properties that we have built into some of the skills.

Auryan: That sounds great.

Stephanie: Okay. Awesome. So same thing, we'll start with the slash, and we'll do degree page writer. Write a new program page for ASU Online for the advanced nursing practice education DNP. And then we have several briefs that we will be attaching here.

Auryan: Okay.

Stephanie: So we're going to add files. And we are attaching a few things here. There are three things, and one right here is the fact sheet. So this is something that comes from the academic unit, and it is the source of truth for the program. So this is the unit telling us a whole bunch of information about the program, kind of what they're looking to include, any differentiators, courses, things like that, what students will learn.

Then we also have a market research brief. This is from our portfolio and portfolio strategy team. So this is them giving some of that market research, competitor analysis, guidance on how we can position this program to help ASU Online stand out as we're writing the content to help us give that, hopefully, competitor advantage when we launch the page. And then finally, we have a search strategy analysis right here. This is from our search strategy, our SEO team. And this is where we're getting into the SEO, GEO, AEO, whatever you and your team are calling it these days.  

Auryan: All of the engine optimizations.

Stephanie: Exactly. Yeah. This is where our search team, they are giving us that expert analysis of exactly how to show up in all of the different engines from Google, LLMs, AI overviews, and giving that guidance of how we can optimize around certain keywords, FAQs, things like that.

Auryan: Awesome. All sounds like great data and important data to have when you're producing an output like this. So it makes a ton of sense.

Stephanie: Yes. And this is a lot of information already, so my prompt is really not going to include too much. Because the skill has a lot of information. We have a lot of information in the prompt, so we are just going to go. 

Auryan: I love a short prompt. We're getting right into it. Yes. You made it easy for people to use, so I think that's a good thing, not a bad thing. 

Stephanie: Yes. And then this skill will use all of the attached information to create a program page.

Auryan: I see we got an output on your computer.

Stephanie: Yes.

Auryan: It looks like the skill finished running. Walk me through all the things that it did in the output. How does this look?

Stephanie: So up here at the top, we have these pieces of information. A lot of this is important for the CMS upload. A lot of this is pulled in from the fact sheet, but some of it's pulled in from the SEO brief. We also have the start date that is pulled in from the academic calendar. So it pulled in, oh, this is the fall A start date is August 20th. And then up top here, we have the page title, meta description. This is all pulled in from that SEO brief, that search strategy analysis, and then it writes this short summary, the expanded details here, writes all of this based on the information that was provided in those different briefs. Pulls in the careers from the fact sheet, writes this intro blurb.

Auryan: I'm also seeing some custom formatting in here. The highlights, was that Claude, or is that something you all told it to do?

Stephanie: Yes. So this is actually our template for these pages. And that goes back to before we started using ChatGPT or Claude. So we trained Claude on this template. We told it we want these highlights. We told it we want this chart at the top that has this information. We want all this content in this order, in these sections, with this formatting, and it knows that and follows this exact template now every single time.

Auryan: That is already so impressive. All the steps and all the different areas that this is actually checking to produce its output. This is truly a multi-step agentic workflow. You're referencing dates. You're going into different documents. You're accessing different databases. This is really comprehensive. 

Stephanie: We did start chatting with Claude, and we were like, "Okay, what if it could create the doc for us? What if it could go out to the rankings tracker and pull in that information? What if it could do all these different things that were pain points?"

And, you know, we were chatting with Claude about, "Can you do this? Can you do that? Can you do this? How would you do this?" And it kept saying yes. And we kept building that and kept iterating on it, and now we have a system that is very agentic, where again — almost everything in the content creation process is built in here, and we just have that human in the loop for those quality checks. 

Auryan: Sounds like there's almost a mindset shift as well that helped enable this. This idea of, "Well, it couldn't do that," turned into, "Well, what if?" And that turned into slowly, "Hey, everything we're saying, what if it could do that?" It can do that, so now we're asking intentionally, "Can you do this? Can you do that?"

And so you went from someone who's very hands-off with the AI to someone who's really now pushing the limits in terms of what it's actually capable of doing, and I think almost that mindset shift feels like one of the most powerful aspects of everything you've talked about.

Stephanie: Now it really feels like the only limit is our own imagination.

Auryan: That feels so incredibly empowering. I mean, when you have a shift like that and you're asking yourself, "What's next? Our imagination is the limit.” What is next? Where are you all trying to take this platform next? What are the next set of features you're looking at? Like, what does the next year of this plugin look like for you and your team?

Stephanie: We have adapted a version of this skill for an enterprise project that we are working on with the Provost’s Office and a lot of partners across campus. We are working to rewrite, re-optimize all 1,400 degree descriptions for ASU degree search, so not just ASU Online degree programs, but now the campus in-person programs. 

Auryan: Just a few programs.

Stephanie: Yeah. Again, you know, a little over 1,000. And, you know, that wouldn't be possible really without something like AI.

Auryan: What has this impact looked like for your team? I mean, it's clear you guys are producing really high-quality content. You're looking at automations. You can do work at a scale you weren't able to do before. But how does this make a difference day-to-day for you and your team?

Stephanie: So I think one good example that is related to degree pages, and we actually used to have one person on our team, and really, almost their whole job was to write degree page content. There'd be other team members I'd have to help sometimes, and it'd really be all hands on deck because it's like, okay, what can we pick up? Because we have people who need to focus and dedicate so much of their time to this. And this year for onboarding using ChatGPT and then Claude, we didn't have to have one person who was solely dedicated to this.

Auryan: That makes a ton of sense. Another great example of Claude meeting you where you already work, and not having to redefine everything just to start working on a platform like this. So, great call out there. I love it, you got Claude working on the side while you're getting other things done.

Stephanie: Yeah, I can finally multitask for the first time in my life.

Auryan: Ah, I love that. Get that quote, "You can finally multitask." What do you want people to understand about how this tool was built?

Stephanie: So I think it's really important, and I really want to emphasize that, my team, we are not coders. I didn't even imagine myself working in tech for a long time because I'm like, "I'm not like a techy person." Like, I don't think anyone on our team would be like, "Oh, I'm like a tech wizard," right? So you don't need to be a coder. You don't need to be a tech wizard. You don't need to be an AI expert. All you need to do is go into Claude, go into ChatGPT, go into whatever your favorite AI tool is, whatever one you prefer to use, and just start asking questions.

Just start using it, start asking questions, and just start imagining what you would want your work to look like. What pain points do you want to get rid of? What things do you want to automate? What do you want to improve? And start dumping that in there. Start asking how you can make it better, asking, "Hey, can you do this? Can you do that?" And I bet more times than not, you'll get a yes, and you can start building from there.

Auryan: I couldn't think of a better lesson to end this on than to just ask. When in doubt, when you're curious, when you don't know, when you have that idea that you're just like, "I don't know how do I make this happen." Go to your AI tool of choice and just ask. And with that, I couldn't think of a better ending for the episode. Stephanie, thank you so much for coming on Behind the Build and being our first guest. We are going to be ending here. I hope you all learned something. If you're curious or you want to know more about these projects, please reach out to Stephanie or myself.

We'd be happy to provide that information, and thank you all for watching.