[Amber Samdahl] 11:03:17 Well, welcome everybody as the uh Zoom fills in here. My name is Amber Samdahl. I'm the Director of Design & Innovation at PBS Wisconsin, and I am part of the Public Media Innovators Group Big thank you to NETA for supporting our work. [Amber Samdahl] 11:03:34 And on behalf of my public media innovators colleagues, David Huppert and Chad Davis, we're really excited to be sharing the webinar today with our marketing communications PLC friends as well. So thank you to everyone. [Amber Samdahl] 11:03:47 And the Marcom group as well. We'll hear from them in a minute. And just, you know, thanks all for joining today. We have a really exciting webinar. We have four presenters. [Amber Samdahl] 11:03:58 We're going to hear all sorts of different connections to AI through digital, social strategies, marketing strategies. We'll hear about custom GPTs. What is a custom GPT? You'll hear about Claude AI. [Amber Samdahl] 11:04:13 A really cool tool called Opus AI and how you can actually take away some of these tools and prompts for you to use at home. [Amber Samdahl] 11:04:23 Please feel free throughout. The webinar to put questions in the chat as we go. We'll have four presentations and then we'll have a dedicated Q&A time at the end. [Amber Samdahl] 11:04:31 So feel free to put questions in the chat. And make sure you select to everyone in the chat because sometimes it defaults to host and panelists. [Amber Samdahl] 11:04:40 So make sure you select everyone. And I'm just going to share a couple of links before I turn it over to Erik Ernst. First, here's a link to our Public Media Innovators site where you can find a recording of this webinar within about a week and we'll share resources, chat. [Amber Samdahl] 11:04:57 Transcript and the recording up on that site within a week. [Amber Samdahl] 11:05:01 And then my shameless plug for next month's webinar. For anyone who is interested in games and particularly if you're skeptical of public media. [Amber Samdahl] 11:05:13 Participating in the games world, we have a really, really interesting webinar coming up in June, bringing in a researcher from AARP who specializes in gaming research for the 50 plus audience. So this is really exciting. She's done some incredible [Amber Samdahl] 11:05:31 Work looking at work looking a potential audience that we're overlooking. This is a really good place for public media to be thinking about for our future. So I encourage people to register for that. [Amber Samdahl] 11:05:43 As well. [David Huppert] 11:05:45 Amber, tell them about next week real quick about Tech Summit. They can find us there. [Amber Samdahl] 11:05:48 And then next week, yes, absolutely. There's annual meeting, which several of the people on this webinar today will be participating in and presenting at. [Amber Samdahl] 11:05:58 And also a tech summit. We will be doing, participating in an AI discussion at Tech Summit next Thursday, as well as a session looking at games from different stations across public media. [Amber Samdahl] 11:06:12 Check us out there as well. Let us know. Drop a note if you will be there. Happy to connect. [Amber Samdahl] 11:06:18 And now I'm going to hand it over to Erik Ernst, who's the chair of the Marketing Communications PLC. [Erik Ernst] 11:06:26 Thanks, Amber. Yes, I'm Erik Ernst, Associate Communications Director at PBS Wisconsin and Chair of the NITA Marketing and Communications Peer Learning Community. [Erik Ernst] 11:06:35 All of us on the Marcom PLC are just so grateful to be collaborating with our innovators colleagues on this webinar that will offer Some real life examples of how the innovative AI work that our colleagues are doing across the system could inform and inspire work at our own stations. I want to invite you to watch for more webinars from both of our PLCs on this and other topics over the next year. [Erik Ernst] 11:07:00 Topics that affect all of our work. And those webinars are only made more useful by your feedback. So please reach out to us, myself, Erik Ernst, our marketing and communications PLC liaison Tonya Weber, with ideas that you have around how these opportunities can best serve your work moving forward. In the past year, we've done sessions similar to this. We've done sessions on [Erik Ernst] 11:07:29 And telling our station stories, we've done open office hours where we just gather together as groups to talk about topics of the day. And so we welcome anyone who works in the marketing and communication space to join us for those sessions. As Amber mentioned, like any discussion around AI, we expect there's going to be many questions and we welcome those in the chat. And we'll be collecting those as we go along and getting to as many of them as possible. [Erik Ernst] 11:07:53 At the conclusion of our presentation. So I'm not going to hold up us getting to the real content today any further. So I'll hand back to Amber to do our first introduction. [Amber Samdahl] 11:08:03 Thank you, Erik. All right. I'm excited to introduce our first presenter, Jacqueline Lo is the Director of Interactive at South Florida PBS and has been doing some really incredible And in innovative work at her station. So I'm going to hand it over to Jacqueline to walk us through. [Amber Samdahl] 11:08:19 Thanks. Thanks for joining us. [Jacqueline Lo] 11:08:21 Thanks, Amber. All right, I'm going to share my screen. Hopefully it works. [Jacqueline Lo] 11:08:30 Can everybody see my screen? Awesome. Okay. [Jacqueline Lo] 11:08:36 So yeah, today I want to talk about a custom GPT that we built specifically for our content team. [Jacqueline Lo] 11:08:46 A custom GPT is essentially just a custom chat bot that is used for a specific purpose or a specific job. [Jacqueline Lo] 11:08:54 Really, that's what it kind of comes down to. [Jacqueline Lo] 11:09:01 And so I'm just going to talk about briefly about why we built it. We have a lot of content that we're putting up on digital platforms Every day. And it's coming from different places. We have, you know, producers, editors, you have social media coordinators, and they're they're all [Jacqueline Lo] 11:09:19 Prior to this, they were all kind of doing their own thing in terms of like where they were generating the metadata. [Jacqueline Lo] 11:09:27 How that metadata was getting transferred from team member to team member. So it was a little bit of a a mess honestly we had like an incomplete metadata. Some things weren't being done that should have been done. There was an inconsistent voice and format. [Jacqueline Lo] 11:09:43 Throughout the content team and a lot of our producers and editors kind of just basically was an afterthought because it's really it could be very time consuming creating all this metadata for all these digital pieces. [Jacqueline Lo] 11:09:58 So using the custom GPT, we were able to have a better consistent voice and format, much quicker turnaround for all this metadata. So producers and editors could send it to our social media folks. [Jacqueline Lo] 11:10:14 And not have to worry about it. And we have internal social toolkits that we deliver back and forth. And it made those internal social toolkits more robust, a lot more organized. And so everybody was kind of on the same page. [Jacqueline Lo] 11:10:31 And so what RGPT does is it basically takes specific inputs. And we typically use transcripts as kind of like the assets that builds all the metadata. [Jacqueline Lo] 11:10:43 So we take transcripts. We also do video descriptions, literally just a text description. And we add that into the GPT and then it will output all of these specific assets. So YouTube descriptions. [Jacqueline Lo] 11:11:00 Video titles. It will give you five suggested video titles, hashtags, timestamps, which is really crucial for YouTube because You can create chapters in YouTube and that helps with SEO. It helps with discovery social copy based on the platform. So it will optimize copy based on Facebook, Instagram, TikTok. [Jacqueline Lo] 11:11:23 You'll also generate thumbnail text. So if you're creating custom thumbnails, it will give you ideas for the text to put in the thumbnail. [Jacqueline Lo] 11:11:31 And then social engagement questions. So if you are trying to interact with your audience and you want to generate more social content, it will also generate engagement questions based on that asset. [Jacqueline Lo] 11:11:44 And so here's just a brief of like how it flows. You add your asset. You can select your outputs. The default is that it will create all of this for you unless you want something specific. [Jacqueline Lo] 11:11:55 You generate it. And then if you want, let's say additional video titles or you wanted to optimize something further, additional prompting The big one is making sure that you're editing everything. You never want to copy and paste something straight from the GPT. You want to edit, making sure that there's no typos at the [Jacqueline Lo] 11:12:13 People's names, locations are correct, and things like that, and then post. [Jacqueline Lo] 11:12:19 So I want to show you quickly. What that looks like. So this is what the GPT looks like. [Jacqueline Lo] 11:12:31 We have four boxes here. So in case somebody does not want to prompt, you know, you could literally just say, please create metadata from this transcript. [Jacqueline Lo] 11:12:41 Copy and paste the transcript or add the file here Or you can just click on this box that says create metadata from a transcript click on it. It's going to ask you, hey, can you please upload like a transcript? I will generate all of this for you. [Jacqueline Lo] 11:12:57 And so I conveniently have a transcript Oh, God. [Jacqueline Lo] 11:13:07 Copy this. You just literally paste it in. You paste the transcript you hit enter I was going to think for a second. [Jacqueline Lo] 11:13:18 This was just a… a short like one minute transcript from one of our stargazer interstitials And so he'll see it starts generating YouTube video titles podcast titles. So we do have some podcasts. So it just that are that live on YouTube and other places so [Jacqueline Lo] 11:13:36 It gives you that. Youtube descriptions, and you'll see there's custom, we have URLs set in place and I'll show you how we generate that. Again, timestamps You always want to make sure that timestamps are accurate. I do notice that sometimes GPT, ChatGPT will hallucinate sometimes when it comes to this. [Jacqueline Lo] 11:13:58 So just double checking that the timestamps are accurate. Hashtags per platform so you get youtube Instagram and TikTok hashtags thumbnail text options. Again, this is a text that you would use to create a custom thumbnail. [Jacqueline Lo] 11:14:14 Social copy optimized per platform. So your Facebook, Instagram, and TikTok copy And then social engagement questions. [Jacqueline Lo] 11:14:21 So that's it. I mean, this just really helped us again with being more consistent, having all of the producers and editors being able to quickly turn around metadata And it made our social media team, it's much easier for them to just go in and just copy and paste and start scheduling since we have so much content. [Jacqueline Lo] 11:14:45 So one other thing before I finish, I just wanted to show you how to build quickly how to build this. It's fairly simple, actually. [Jacqueline Lo] 11:14:54 So if you're in ChatGPT and you're in your normal kind of dashboard here. [Jacqueline Lo] 11:14:59 You want to go to explore GPTs. And then there's our create icon. And then once you hit create it's going to take you to take you this page here. [Jacqueline Lo] 11:15:12 Which looks like this. So you're going to see a create and then a configure. And these essentially do the same thing. Create is something if you want to, it's based on prompting. So there's nothing super technical, you're literally just creating the GPT through conversation, through prompts. So if you want to create something fast, you want to test something out and see if it could work. [Jacqueline Lo] 11:15:34 And this is a great way to do it where you don't have to have too much customization and you want to do something really quickly. That's what the create is for. And then configure is where there's more customization. You can do a lot more refining of your GPT. So obviously you can put the name, description. [Jacqueline Lo] 11:15:52 Instructions is where you would add your prompt. And so this is the base prompt we used to create the GPT. [Jacqueline Lo] 11:16:00 And just a couple of things to note with this is that the GPT prompts, we really emphasize optimization per platform and SEO and aligning with current trends per platform best practices. And we really emphasize that in the prompt so that it would give us the best output for each platform. [Jacqueline Lo] 11:16:24 And then for specifically, we have for YouTube, we're always adding our socials at the end of all the descriptions. And so instead of having to write it out every single time, the GPT will provide that for us. And we, sorry, we specify that in the in the prompt [Jacqueline Lo] 11:16:45 And then if we keep going, there's conversation starters. This is where you add those little kind of like I guess essentially like prompts for the GPT in case somebody doesn't really know how to engage right away So you do that here. And then let's say you have a brand guideline, you have a style guide for your organization or anything like that. You can actually upload that here. [Jacqueline Lo] 11:17:13 And in the prompt, tell the prompt to refer to the style guide when outputting all of this copy. [Jacqueline Lo] 11:17:19 So that you're in line with your brand and your style, guys. [Jacqueline Lo] 11:17:24 And you can do that here. And then other capabilities, you can attach it to the web so it can search the web if you need it. Image generation, all of that can also be added. [Jacqueline Lo] 11:17:36 So that's pretty much it. That's our GPT. [Amber Samdahl] 11:17:43 Thank you, Jacqueline. We have collecting questions, which we'll gather together at the end, but really quickly. [Amber Samdahl] 11:17:51 Can other stations use this custom GPT? [Jacqueline Lo] 11:17:54 Yes. So this one in particular is public and I'm going to actually add the link in the chat for anybody who wants to use it. Obviously, it's geared towards our organization. So like our links are going to populate and things like that. But just to get an idea of how to use it and play with it, by all means, you can definitely go ahead. [Jacqueline Lo] 11:18:15 And go ahead and use it. [Amber Samdahl] 11:18:18 Well, and we will share out at the end We have collected together notes from all of our presenters and links and contact information for them if you do want to follow up with questions. So we'll be sure to share that at the end. [Amber Samdahl] 11:18:31 And we're collecting all these great questions to address. As well for our Q&A section. So thank you, Jacqueline. [Jacqueline Lo] 11:18:40 Thank you. [Amber Samdahl] 11:18:40 We will move next to Mark Riechers, who is a digital content editor for PBS Wisconsin. [Amber Samdahl] 11:18:47 Who's going to talk about his work in Claude. [Mark Riechers] 11:18:52 Hey, everybody. I'm going to share… two tools today, actually. And it's a lot of it dovetails with how custom GPTs work. [Mark Riechers] 11:19:02 Claude has a similar… system, they just call them projects. [Mark Riechers] 11:19:08 But in this case, these are tools that are geared toward helping editors like me, coaching producers who are mainly focused on their video production. [Mark Riechers] 11:19:17 Helping them write better descriptions of their videos Thinking about things like alt tags or things that they might not necessarily think of when they're posting images as parts of their stories and things like that. [Mark Riechers] 11:19:30 I think the really thing, the important thing to keep in mind with integrating AI tools into your workflow is that you're solving a specific problem and you're kind of starting with a human input and you end with a human output. [Mark Riechers] 11:19:42 So the AI is really just helping kind of standardize things, speed things up, make things more efficient. [Mark Riechers] 11:19:48 So I'll start with this tool That I will was graciously shared by my um My boss, Tim Schneider. [Mark Riechers] 11:19:58 I'm hoping that my screen share works okay. [Mark Riechers] 11:20:07 And you are hopefully seeing GPT and not my time card. [Mark Riechers] 11:20:14 So in this case, it's a pretty simple input. You have an image and you need an alt tag of it and we're trying to train people on what's the difference between alt tag or a title tag or things like that. So the input is just going to be [Mark Riechers] 11:20:29 Any image. So you can just drag that right in. We'll upload it. [Mark Riechers] 11:20:36 This is very similar to the custom GPT we just saw it's just basically being instructed, like, if I give you an image Identify it. Here's a set of best practices around what we want in an alt tag. [Mark Riechers] 11:20:49 So, and this already kind of underscores the sort of the human output piece you know um older adult holding orange flowers. [Mark Riechers] 11:20:58 I don't know that we would necessarily like label someone's age in an alt tag. So that's why it's important to review. [Mark Riechers] 11:21:06 But this is the kind of thing where, you know, we go from a lot of stories that might not have any alt Tags, any text on the images to being able to pretty much standardize every image on our site has a really good alt tag on it. [Mark Riechers] 11:21:19 So that's a really simple tool. So what we do with Claude is a little bit more complex. [Mark Riechers] 11:21:25 We're essentially transforming transcripts like we just talked about. Jacqueline's presentation. [Mark Riechers] 11:21:31 Into… We're really focused on Media Manager for this product. [Mark Riechers] 11:21:36 Product for this Claude project. So what it's going to do is three things. It's going to it's going to take transcript data. [Mark Riechers] 11:21:50 Any working copy that a producer might already have. And then SEO data and help us kind of coach edits on how to improve something for discoverability and for clarity and even just basic things like AP style. [Mark Riechers] 11:22:04 We can start with a transcript. [Mark Riechers] 11:22:13 So just being given the transcript file, again, similar to the custom GPT, Claude has these. [Mark Riechers] 11:22:18 Basically like a set of expectations and the role that it's supposed to be playing and what I expect as a deliverable. [Mark Riechers] 11:22:24 So right now it's reading the transcript. And this initial document is essentially it's brainstorming possible titles, short descriptions, long descriptions and SEO keywords. [Mark Riechers] 11:22:37 For media manager. I find this… point in the process is actually just best for showing you what bad titles might look like. [Mark Riechers] 11:22:48 But it kind of jumpstarts the brainstorming process of Like you can present some options to a team that might be struggling with naming something. [Mark Riechers] 11:22:55 You can start thinking about What are the keywords that you want to integrate from the transcript itself? [Mark Riechers] 11:23:01 You can also do stuff like because you have the access to the full video, this is a transcript of a documentary. [Mark Riechers] 11:23:08 About lighthouses. I want to be sure. [Mark Riechers] 11:23:15 To mention all the lighthouses we actually visit. [Mark Riechers] 11:23:26 Doc. [Mark Riechers] 11:23:30 And in this way, you can kind of have a dialogue about how you might want to tweak descriptions and how you might reference things from inside of the transcript. [Mark Riechers] 11:23:47 Sometimes it takes a while to think. That one doesn't feel… that complicated but [Mark Riechers] 11:24:06 So this box where it says thought process, this is um Claude is using, it's called a chain of thought model. It's essentially talking to itself to kind of work out the best way to solve the problem. [Mark Riechers] 11:24:17 So this is kind of… It's going through the dock and trying to find all the named lighthouses. And now it's going in and rewriting these brainstorming descriptions. [Mark Riechers] 11:24:38 So now let's say a producer has handed in some copy And we want to see if maybe we can refine it. [Mark Riechers] 11:24:47 We do a lot of our copy composition in Airtable. You can just straight up post a screenshot from Airtable. [Mark Riechers] 11:24:53 And it's set up to say like, okay, this means you want to revise something. [Mark Riechers] 11:24:57 Rather than brainstorming. [Mark Riechers] 11:25:06 And what this is going to do is essentially take SEO best practices and kind of critique what we already have. [Mark Riechers] 11:25:15 And maybe make suggestions and it'll do a side-by-side comparison. It's just taking a little bit of time. [Mark Riechers] 11:25:25 So I find this to be pretty useful when working with producers because you can make these kinds of suggestions of like, hey, maybe be sure to mention Lake Michigan And then the producer might say like, well, Lake Michigan is not like, we can be more specific than that. So again, you're kind of jump starting the conversation around what should be in the final copy. [Mark Riechers] 11:25:44 But it also lays out its reasoning for why an edit would make sense. So like in this case, naming lighthouses Using terms that have some high SEO value. [Mark Riechers] 11:25:56 And then the final piece that this can do, I think, is actually maybe the most interesting. So we subscribe to a service called SEMrush. [Mark Riechers] 11:26:03 It's a search engine marketing platform that gives us a lot of like search volume and search intent information. [Mark Riechers] 11:26:09 So I'll take the keywords that Claude generates run an analysis on them in SEMrush and then share the screenshot back to Claude. You could do this with Google Analytics or like any data tool. Claude is flexible enough for that, but my instructions are kind of cued to what we have available. [Mark Riechers] 11:26:29 And now again, it's the chain of thought. It's reviewing everything that's in this screenshot. [Mark Riechers] 11:26:37 It's pulling the search intent and the volume and things like that. Now it's drafting a third report Which is a keyword report And… This is usually a good time that I go and get coffee or things like that. [Mark Riechers] 11:26:55 When it's running. [Mark Riechers] 11:27:02 Okay, so in this keyword report, this is laying out. So this is basically taking, and this is another thing I really like AI tools for. It's essentially taking information from varying platforms and really like quickly formatting it for what you need in a moment. So in this case, it's like, it's been told like, hey, give me a list that's easy to copy paste into [Mark Riechers] 11:27:21 A platform like Media Manager, but also give me I was going back and trying to revise some stuff. [Mark Riechers] 11:27:28 But also give me a breakdown of like what are the unique based on the data from SEMrush and SEMrush like your own analysis, what are the unique keywords we might use What are the high, medium. [Mark Riechers] 11:27:40 Keywords based on specific categories. But then this is something I take to producers all the time. [Mark Riechers] 11:27:47 This is essentially just recommendations of like verifying like, hey, is your title good or could it use some tweaking based on the data Are there words you should be sure to include in the short description, sure to be included in the long description. [Mark Riechers] 11:27:59 These are all suggestions and that's always how I frame them. It's not something where we're trying to cram a bunch of keywords in every description. [Mark Riechers] 11:28:05 But it's this kind of thing where you can be kind of a data informed and kind of show people how it might enhance their work as you're going. [Mark Riechers] 11:28:15 So I think that's the long and short of it. I will provide instructions for the Claude project. I can stop sharing my screen. [Mark Riechers] 11:28:23 And then I can actually share a link to the the alt tag writer. And thank you again to tim for being willing to share his work. [Mark Riechers] 11:28:38 There you go in the chat. [Amber Samdahl] 11:28:41 Thank you very much, Mark. Thanks for sharing and yeah, I encourage everyone to take a look at that. [Amber Samdahl] 11:28:47 Yeah, keep putting your questions in the chat. We will keep moving along because we've got more great examples to share. Next up is Jonna Kelley. [Amber Samdahl] 11:28:57 Who is the marketing director at KSPS and PBS. And I'd love to invite her up to share about her work. [Jonna Kelley] 11:29:06 Thanks, Amber. I'm going to attempt to share my screen. Can everybody see my screen? [Jonna Kelley] 11:29:15 I'm going to talk a little bit today about how we use Can't. Oh, shoot. [Amber Samdahl] 11:29:19 Can't see your screen yet. [Jonna Kelley] 11:29:28 How about now? Thank you. Sorry about that. So I'm going to talk a little bit today about how we used a custom GPT to create an Gary Stokes Speak, which is for PR and engagement Yeah, we just needed… conversational AI to connect, inform, and inspire. [Amber Samdahl] 11:29:29 There we go. Got it. [Jonna Kelley] 11:29:54 So we created Gary Stokes Speaks. Because we needed to communicate our mission consistently across platforms. [Jonna Kelley] 11:30:04 While maintaining Gary's specific style of communication. He's very warm and sincere and is a community leader in our region. [Jonna Kelley] 11:30:14 And I saw potential to scale personal engagement, especially for community outreach donor support and advocacy messaging This has been coming in very handy the last few weeks as we're all battling the federal funding crisis? [Jonna Kelley] 11:30:31 And this GPT was designed to reflect Gary's trusted tone. Part neighbor, part storyteller, part advocate for public media. [Jonna Kelley] 11:30:40 So how we configured it, we built this in OpenAI's chat GPT platform. I think Jacqueline had such a wonderful description earlier in our presentation about how to do this. [Jonna Kelley] 11:30:53 So reference that in the recording. We configured using a persona brief. [Jonna Kelley] 11:31:00 And for that persona brief, it was pretty detailed. I wanted to include real leadership trades for Gary and recurring values so In the persona brief, I asked GPT to speak as someone who was really passionate about education and cultural access grounded in regional storytelling and public service and speaks like a neighbor and not a corporate corporation. [Jonna Kelley] 11:31:27 I also gave specific instructions to emphasize our mission values, community programs and tone And then I integrated a lot of Gary's internal messages to make sure that it sounded authentically Gary. [Jonna Kelley] 11:31:47 So that meant that I combed through some of the messages that I've received from Gary over the past couple of years. [Jonna Kelley] 11:31:53 Some of them were pretty lengthy and were like to all staff, but some of them were like just more anecdotal short messages that he sent to me, but these ended up being anchor phrases that GPT uses naturally when it writes stuff for me. [Jonna Kelley] 11:32:09 And then I added guardrails to ensure that the tone remained hopeful, inclusive, and community focused. [Jonna Kelley] 11:32:18 I did spend a lot of time adding guardrails to this because it was so important that we got it exactly right. So guardrails include um Persona reinforcement instructions so I used phrases like speak with the warmth of a community leader who sees every viewer as a neighbor. [Jonna Kelley] 11:32:38 I added tone boundaries and filters. So parameters to avoid alarmist and accusatory or overly corporate language. [Jonna Kelley] 11:32:48 And then I added bias mitigation and inclusive framing So some of those prompts were avoid assumptions about audience demographics, default to inclusive people first language. So as an example, instead of this GPT using terms like viewers, it might use families across our region [Jonna Kelley] 11:33:12 And then I also… wanted to make sure that we were not engaging in political commentary or polarizing topics. So we can speak about things like public media funding or policy in general, but this GPT avoids taking political stances or critiquing [Jonna Kelley] 11:33:32 Specific public figures or parties And then after we configured it, we needed to refine it. So that meant that our team did some real world testing. So we asked Gary to write a blog post. And so he wrote a couple of paragraphs for his blog and then [Jonna Kelley] 11:33:51 We asked the GPT to help Gary write this blog post. [Jonna Kelley] 11:33:57 And then we created feedback loops, which is really just the marketing team And Gary. So we sent him what GPT wrote with this GPT wrote, Gary's Gary Stokes speaks and then asked him, does this sound like you? And then he refined accordingly um [Jonna Kelley] 11:34:16 So he adjusted responses to reflect local context. You know, he added things like using the actual names of towns and the communities we serve. [Jonna Kelley] 11:34:29 We added Inland Northwest references, local programming, and familiar community partners. [Jonna Kelley] 11:34:36 We made sure to add community partners that were longstanding. So like Spokane Public Schools or Spokane Public Library. [Jonna Kelley] 11:34:44 To just make sure that this will remain authentic longer. And then we added internal context access. So like Jacqueline mentioned, we uploaded things like notes from our board meetings or our voice and tone guide. [Jonna Kelley] 11:35:03 To help refine this GPT. [Jonna Kelley] 11:35:11 A couple of best practices for using Gary Stokes Speaks. We use it for Drafting messages to members, viewers, and donors, answering community questions. [Jonna Kelley] 11:35:20 And talking points, emails or scripts. Recently, we use this to help us write some scripts for public media giving days. [Jonna Kelley] 11:35:28 We do not use this for factual or legal statements, replacing human judgment. [Jonna Kelley] 11:35:36 It's a support tool, not a decision maker. I think Mark talked about that very well during his presentation. [Jonna Kelley] 11:35:44 And a couple of tips for using an AI for PR engagement is be clear in your prompts an example that I'll use is write this as Gary during a pledge drive. [Jonna Kelley] 11:35:58 Which we did recently and this worked and helped. And let it inspire, but always review and personalize before sending or publishing. [Jonna Kelley] 11:36:07 There has never been an instance that we didn't need to personalize something after AI has generated it for us. [Jonna Kelley] 11:36:19 I also included some more best practices in notes that Amber will share at the end of this presentation, but Yeah, that's how we're using AI for PR engagement with our community. [Amber Samdahl] 11:36:36 Great. Thank you very much, Jonna. It was really fun. Fun to see. And yeah, we'll be sure to share out All the great tips and details that you've shared with everyone. [Amber Samdahl] 11:36:51 We will be sure to get that to everyone at the end here. And next up, I'd like to welcome Mary Anne Lane, who is the Director of Educational Content at Georgia Public Broadcasting. [Amber Samdahl] 11:37:02 Welcome, Mary Anne. [Mary Anne Lane] 11:37:03 Thanks, Amber. Hi, everybody. For my portion of today's presentation, it's going to be a little bit different. So we're not feeding prompts into any kind of AI system. Instead, we're going to take existing content that you probably already have and we're going to format it into a shorter version. So vertical videos for YouTube. [Mary Anne Lane] 11:37:29 Tiktoks, reels for Instagram, all that good stuff, and basically clipping Content that you already have and thinking smarter and not harder. So the tool that I have to show you all today is called Opus Clip AI. Just real briefly, because I know people are going to ask, there's four different versions of it. There is a free version. [Mary Anne Lane] 11:37:56 The credits, when you see credits pop up, that just means minutes. So you have up to 60 minutes per month if you want to test it out and explore a free version. [Mary Anne Lane] 11:38:08 Our team has pro, but we split it across all of our divisions. So I work in the education department, but we also share it with everybody else within the organization. [Mary Anne Lane] 11:38:20 But if you're looking to just kind of, you know, get your feet wet. [Mary Anne Lane] 11:38:26 The starter is good, although pro looks to be better because it's cheaper right now. [Mary Anne Lane] 11:38:30 But this is just the breakdown of what the different options are. But there is a free version if you want to explore it. [Mary Anne Lane] 11:38:37 So this is what the homepage looks like. You'll see up here, this shows me how many minutes I have left, so it helps track the number of credits that you have. And so you can kind of keep an eye on things. [Mary Anne Lane] 11:38:52 And see how it's being used. Yeah. Oh. [Amber Samdahl] 11:38:55 Mary Anne, sorry, we can't see your screen. [Mary Anne Lane] 11:38:57 That's right, because I didn't share it. Just jumping right in. [Mary Anne Lane] 11:39:01 Sorry, y'all. There we go. How about now? [Amber Samdahl] 11:39:03 Thanks. [Mary Anne Lane] 11:39:08 Okay, perfect. So just real quickly, there's three different paid versions and then there is a free version. So you can explore from there. [Mary Anne Lane] 11:39:18 But this is what the home page looks like. It works with Rumble, YouTube, Vimeo, Zoom, Twitch. [Mary Anne Lane] 11:39:27 Wherever your content is living, you can post it into Opus clip and then it will clip from there. [Mary Anne Lane] 11:39:35 I'm going to show y'all an example of a video that we have. It's eight minutes and 46 minutes. [Mary Anne Lane] 11:39:43 Seconds long so not the right kind of length that we're looking for for these short versions. [Mary Anne Lane] 11:39:51 So you just drop your link And let's say that you have kind of I know a lot of documentaries like to open with like really beautiful scenery and that may take up 30 seconds or so. You can clip it. [Mary Anne Lane] 11:40:07 Your clip and then so that you're not using up those credit minutes. Same thing if there's something on the end that you want to shave off or if there's a long introduction and you just kind of want to get to the meat of things. [Mary Anne Lane] 11:40:20 You can trim it using this tool here and that helps save credits. [Mary Anne Lane] 11:40:26 You can also search by keywords and separate them by commas if there's a particular topic that you're wanting to look for And then also, let's say that you have a promo that's formatted for interstitial. [Mary Anne Lane] 11:40:43 And you actually don't want it clipped and it's the right kind of length, but you just want it reformatted from 16 by nine to nine by 16. It will do that for you as well. So you can click on the don't clip. [Mary Anne Lane] 11:40:55 Section. You can also set up templates. So you can add in your own color features, how you want your captions to pop up, because that's another great feature of Opus Clip is it will auto generate captions. [Mary Anne Lane] 11:41:10 And then you can kind of design what that looks like. So you can set up templates or you can just go to presets and there are several options that you can choose from. [Mary Anne Lane] 11:41:21 And you can kind of see what those presets look like there. [Mary Anne Lane] 11:41:25 And then there's also different aspect ratios. So if you want to do nine by 16, a one to one or 16 by nine, you can select that too. But because I mostly use this for our social channels. [Mary Anne Lane] 11:41:41 Which I manage our education YouTube, Facebook, and Instagram profiles, as well as x um But the video content is really helpful to have that nine by 16. [Mary Anne Lane] 11:41:53 So then you would go and click to get your clip. I've already clipped this in this case, so I'm not going to do it here. [Mary Anne Lane] 11:42:03 Save those credit minutes, but I'll show you what it spits out afterwards. [Mary Anne Lane] 11:42:08 So it's taken a nearly nine minute video and then it has given us 14 different options to choose from, which is pretty cool. [Mary Anne Lane] 11:42:18 You have a variety of different lengths. So some are slightly over a minute versus others are 18 seconds, 15 seconds It will, in its generation, it will give you a rating, which I think is very helpful. This will tell you [Mary Anne Lane] 11:42:38 Your virality score. So it's looking for things like hooks and the images, the image quality, the engagement for your audience. And so it just kind of gives a numeric score, which is really helpful. [Mary Anne Lane] 11:42:56 Sometimes I kind of think, well, 99 is like really hard. Let me take a look at this and let me compare it to some of these ones that may not be as high. [Mary Anne Lane] 11:43:06 I always encourage people testing this out for the first time to take a look at multiple clips. [Mary Anne Lane] 11:43:13 Because sometimes they're, you know, it's AI, it's on a human. It does make mistakes. And sometimes the ones that have like the lower scores end up being the ones that I think would perform best. [Mary Anne Lane] 11:43:27 I will show you all. This is on Peanuts. George is the number one peanut producing state in the country. So we have lots of content on peanuts. Maybe there's a peanut festival right around the corner or National Peanut Day. And I want something that's timely. [Mary Anne Lane] 11:43:42 It's, you know, when Jimmy Carter passed, he was a peanut farmer So that may be something I want to tap into as far as a trend. So taking something that exists, clipping an Opus clip is super useful. [Mary Anne Lane] 11:43:57 With the numeric score, it will also show what the different grades are. [Mary Anne Lane] 11:44:02 It will spit out a transcript. And then my favorite feature is that you can actually go in and edit. I recommend that for all the clips, especially when you're dealing with South Georgia accents, there's always something that is either misspelled or isn't punctuated correctly. So you can go in and you can directly edit [Mary Anne Lane] 11:44:26 The word, the text. You can also change the highlight. So let's say that I want the University of Georgia highlighted because that's an important piece of this story. [Mary Anne Lane] 11:44:36 I would go in, change the highlight And again, this coloring comes from a preset that I've already included so that automatically spits it out in yellow and and green, but you can change those colors as well. [Mary Anne Lane] 11:44:56 You can also come in and you can clip it. So let's say that I want it shorter. You have a timeline down here. So it's very easy to drag and clip your content. You can also extend the clip, let's say that it cut off a really important piece that you wanted to incorporate it, you could extend that as well. [Mary Anne Lane] 11:45:21 Another thing too that I think we come across often when we're reformatting something that's built for 16 by 9 and changing it up to nine by 16 is things will get cut off Especially when you're dealing with lower third titles, all of a sudden you have this big chunky text at the bottom and you can't fully read it. You can have, there's multiple different options that you can do to change the look of this. [Mary Anne Lane] 11:45:44 So you could drag. You could drag the screen to reorient it You can also add in your logo or replace the image entirely. [Mary Anne Lane] 11:45:58 But I could also change it from fill, which is going to make it fill the entire screen. [Mary Anne Lane] 11:46:04 And I could do fit, which is going to take it back to its original format and blur that background. [Mary Anne Lane] 11:46:10 And then you can also switch it up. So here. The content that it has spit out has automatically made those adjustments from fit to fill to sometimes it will do split if you have multiple people on camera. [Mary Anne Lane] 11:46:25 But that's just a very easy way that you can go in and change that format. [Mary Anne Lane] 11:46:31 And then you can also track the person. So if there's somebody that you wanted to focus on, if the person is moving, you can have You can have the clip focus on them. [Mary Anne Lane] 11:46:44 And I'll show you all real quickly what this is determining the proper time. Can you hear this? Can I hear this? [Mary Anne Lane] 11:46:51 And it's based on the percentages of harvestable pods. It is part science. [Mary Anne Lane] 11:46:57 It is hard, hard. The University of Georgia developed a chart. [Mary Anne Lane] 11:47:02 That we can scrape the outer layer of the peanut holes And by looking at the color of the So again, you can see how it's mixed and matched the fill versus the fit. And then again, it will do split strain as well. [Mary Anne Lane] 11:47:20 But yeah, there's just really simple edit tools. It's helpful for those who are maybe not super comfortable with video editing. [Mary Anne Lane] 11:47:27 But it's also helpful for those who, you know, you may be a one-man band and you're managing content for your entire station. [Mary Anne Lane] 11:47:36 And you're also running six different social media channels. So it will match the um It will match what you're looking for and then also just kind of give you suggestions on video titles and descriptions. [Mary Anne Lane] 11:47:55 But if you wanted to do more, you could also go in and you could export it into an Adobe file so that you could maybe there's not as many bells and whistles that are built into Opus and you want to do a little bit more. [Mary Anne Lane] 11:48:09 You can go and export it into a project file You can also add in additional text. [Mary Anne Lane] 11:48:18 Transitions, sometimes it will cut something out and blend two things together and maybe that blend isn't super seamless. So you can add in a transition These are some like beta functions, the AI hook, and then incorporating AI b-roll I haven't really explored it yet [Mary Anne Lane] 11:48:42 It just, I feel like the beta functions right now are just very I don't know, touch and go. [Mary Anne Lane] 11:48:51 You can definitely explore that. You can also embed music and there's music that is free to utilize. You don't have to worry about licensing and things of that nature. [Mary Anne Lane] 11:49:01 And then you can also establish a brand template. So that was That first piece that I was talking about, you can determine what the look and feel is going to be for your text, for your colors, whether or not you want it to pop up on screen, have multiple lines of text, one line of text. [Mary Anne Lane] 11:49:18 Incorporate emojis, not incorporate emojis. There's a lot of variance in there. [Mary Anne Lane] 11:49:26 But yeah, we've been pretty successful with some of this. I do think that like when you are looking at your platform, however you can best speak to that platform typically performs really well so like going in and actually scripting out something that has a hook that's good for TikTok or leaning into a trend that's going to get more views. [Mary Anne Lane] 11:49:50 But if you're short on time or you just don't have the bandwidth to produce a ton of content. This is a great tool for that. [Mary Anne Lane] 11:50:02 And I'll show you all just an example. This was from I guess it was last year. [Mary Anne Lane] 11:50:09 There was the eclipse that came through Georgia. And this was a piece of content that we had originally produced back in, I think it was 2017, was the previous solar eclipse um so The content was pretty much the same as far as like, how does an eclipse happen? So we just repurposed that and then published it for last year. So it got a lot of good engagement, did really well on YouTube Shorts. It got over 57,000 views. So it was super simple and probably took [Mary Anne Lane] 11:50:45 Like five minutes of my time. So yeah, so it's great, especially if you have uh documentaries, long form videos, videos that are old and you don't know what to do with them. There's a lot of variance with it, but [Mary Anne Lane] 11:51:00 I definitely encourage you all to check out the free version before you dive into a paid version. But the paid version is very cost effective. [Mary Anne Lane] 11:51:17 For Opus. [Amber Samdahl] 11:51:19 Thank you very much, Mary Anne. All right, I'll turn it over to Erik, who's got a whole long list of great questions. [Amber Samdahl] 11:51:26 And invite all the panelists to come back on. [Erik Ernst] 11:51:26 I do. Yes, I do. And please, we'll try to get through as many of these in the 10 or so minutes that we have here. I did want to start with one, though, that I think it was prompted directly at Mark, but we'd welcome input from each of the panelists. And that is, how do you determine [Erik Ernst] 11:51:44 If a platform like Claude or ChatGPT or some other platform is the right choice for your needs. What are the things that you look for on those platforms? [Mark Riechers] 11:51:54 Yeah, I'm curious to hear… how other panelists chime in on that but um A lot of it depends on the task. So Claude… I've just found, and I've tried some, sometimes I've tried porting custom instructions between different models just to see how they perform. [Mark Riechers] 11:52:11 Claude has just kind of most consistently been a really good writer. [Mark Riechers] 11:52:15 I'd say it's easiest. I said in my chat response, it's good at coaching. It's really good at explaining In kind of gentle terms, like why this or that should be changed or things like that. [Mark Riechers] 11:52:28 Chatgpt. Is a lot better for analytical tasks, things. One of the big things that comes up a lot, I feel like we've probably I bet there's a fair amount of people here who deal with media restoration or media preservation. [Mark Riechers] 11:52:41 Doing things like taking pictures of physical media and then being able to convert that into metadata that needs to go into a spreadsheet. Like ChatGPT is really good at that sort of thing. [Mark Riechers] 11:52:52 So some of it, though, is kind of like, you know. [Mark Riechers] 11:52:55 Realistically, if your station is going to pay for one resource or another. [Mark Riechers] 11:52:58 Like kind of just seeing what the model you have access to can do and then playing around with things around the edges. [Erik Ernst] 11:53:07 Anyone else have thoughts on that? On the panel. [Jacqueline Lo] 11:53:09 Yeah. Yeah. I just like to add that I agree with you, Mark. I do think that Claude is a better writer. [Jacqueline Lo] 11:53:16 And for us, we just kind of As an organization chose the enterprise version of ChatGPT and that's why we're using it. But I do think that Claude is excellent as a writer, better than ChatGPT. And I do think that ChatGPT is great at [Jacqueline Lo] 11:53:32 The analytical side data like i i definitely agree with you on there. So I just wanted to But, you know, put in my two cents. [Jonna Kelley] 11:53:41 I think for us too, chat GPT was easy to invest in and it was the easiest for people to understand because there is a lot of fear around using AI and like what's morally okay in this area. And so we just kind of pick the thing that sounded the most familiar and that was most accessible. [Jonna Kelley] 11:54:02 And it does integrate with a lot of tools that we already use. [Jonna Kelley] 11:54:06 You know, but I did get some tips from Mark because we also use Airtable. And so I was taking furious notes, listening to everybody [Erik Ernst] 11:54:25 So I think to that point of people being scared and familiarity, there was one question for you, Jonna that uh I spoke to that a little bit and that question was, are audiences aware that the copy generated from the tool that you described is not the specific creation of the person, quote, speaking? And how do you [Erik Ernst] 11:54:45 Think through that and how do you address that in your space? And that's in line with some of the other questions that we received around you know whether you're transparent, when organizations are transparent around the use of use of AI. [Jonna Kelley] 11:54:52 Yeah. [Jonna Kelley] 11:55:01 Yeah, that's a great question. And so I think that needs to be determined by your team. You have to talk about it openly and like, what are we going to do here? And we have never published anything ever that wasn't, you know, filtered through a human. So we use AI as a tool. [Jonna Kelley] 11:55:22 And there's this quote that I really love. We all attended a um a masterclass as a group here at the station around AI and how to use it. [Jonna Kelley] 11:55:32 And there was this phrase that I think about often is that AI is consider it brain prosthesis. So it's not replacing you or your thoughts. [Jonna Kelley] 11:55:42 It is enhancing what you can do or maybe ideation So yeah, we would never publish anything without having a human filter first. [Jonna Kelley] 11:55:53 And talk to your team, figure out what the rules are around this. We talked to other stations to get their input about that. Arizona was a great resource for us. [Erik Ernst] 11:56:04 And I think that aligns, I'd open up to other people, not, you know, not that specific question was for John. And I know a few people have answered similar questions in the chat here. But what are the approaches that your stations are taking to around those questions of transparency? [Erik Ernst] 11:56:28 I know that, as Mark nodded to at PBS Wisconsin, these are conversations that are still in progress and that are continuing to happen. And in our case. [Erik Ernst] 11:56:41 As a joint licensee that's also partially attached to the state and partially attached to university, we have different regulations that we need to and expectations that we need to align with with those. And so I know each station will have different approaches to that, but I think that [Erik Ernst] 11:56:59 Transparency is, you know, trust is something that that is at the forefront of public media and and i'm hopeful that like here at PBS Wisconsin that these decisions are being made with that leading in the forefront. And I think that that's a very possible thing to do. [Mark Riechers] 11:57:17 Yeah, in our case, we have an ethics policy that lays out how we think about using AI tools generally. [Mark Riechers] 11:57:24 And then we have… disclosure requirements around basically if the public is being handed something that was directly created by a generative AI system, then we disclose that It was. But if we're using it as an internal tool, then we don't always disclose its use. [Erik Ernst] 11:57:43 I think other people have weighed in similarly in the chat if anyone has anything additionally to add, feel free. [Erik Ernst] 11:57:57 If not, there was a specific question for Mary Anne, and I think that you um You shared some of this, but I wanted to get back and make sure that it didn't get missed. In the end, how clean are the edits that you're seeing? The questioner asked no upcuts or down cuts. Does it fade or cut to and from black? Are those options you can choose? Are you able to expand on that a little bit? [Mary Anne Lane] 11:58:19 Sure. Yeah, you can build in. There's that transition feature. So you can do a fade to black on the front or back end or in between clips. [Mary Anne Lane] 11:58:31 Generally, I would say that they're pretty clean, although sometimes I see like a the next word kind of lingers in where it's been trimmed. So you always want to go back and clean up the captions and But generally speaking, yeah, I would say that it's relatively clean. Like others have mentioned, again, Opus Clip is a tool. And so you always need to have that human element. [Mary Anne Lane] 11:58:57 To look at the captions, make sure things are spelled and punctuated correctly. And then also just from an editorial perspective that the open and that the end is where you want it. I know that there's been some questions about like the ethics surrounding uploading [Mary Anne Lane] 11:59:17 A link to Opus Club and then that being used as like a trained feature for AI. [Mary Anne Lane] 11:59:23 I think that's a really important concern and comment and is not one that we've really had an extended conversation about. So I'll definitely be talking with leadership about how we utilize it. But in this case with Opus, we're not manipulating anybody's voice. We're not changing anybody's likeness. [Mary Anne Lane] 11:59:46 But yeah, I guess it's one of those things that like once you're out on the internet Can the AI use it or is it like once you feed the beast, it's like, oh, this is Now I'm using it for training. [Mary Anne Lane] 11:59:58 Yeah, those are really valid questions, but we do not inform people when we're clipping it just because we we are on usage rights are pretty much in perpetuity and how we see fit as long as we're not manipulating what the person is saying. [Mary Anne Lane] 12:00:16 Or the purpose. [Erik Ernst] 12:00:18 Excellent. All right. So we are very close to time. I want to fit one last question. Speed round. You all have like 10 seconds to answer this. Daniel asks, did you all learn to use these tools on your own or did you get some professional training so the way I would prompt it for the speed round is what is your best one tip or suggestion for people who want to sort of get their feet wet and put their hands in the sandbox of AI. And I'll start with you, Mark. [Mark Riechers] 12:00:45 I would say… talk about AI at work. I think having, I have a very supportive supervisor, so we have a lot of experiments that happen in-house, but I'd also definitely encourage you to subscribe to one of these services on your own. [Mark Riechers] 12:00:59 And learn how it works. And that way you can kind of get the literacy in the tools and how they work and how they're evolving over time. That also gives you a little bit more flexibility to like try something for a little bit and then [Mark Riechers] 12:01:10 Hop over into a different system before your institution makes a big investment in something. [Erik Ernst] 12:01:16 All right, Jana. [Jonna Kelley] 12:01:19 I really highly recommend that masterclass that I referenced earlier. It was awesome. It helped our organization understand how and why to use AI and what it actually is. [Jonna Kelley] 12:01:32 Specifically with chat GPT and also I agree with Mark. We subscribed to like the business tier of chat GPT. So there is better functionality. You can turn things off. [Jonna Kelley] 12:01:44 Yeah, that [Erik Ernst] 12:01:47 All right, Jacqueline. [Jacqueline Lo] 12:01:50 Yeah, for us internally, we do um All staff meetings where we bring in somebody from the AI world and they kind of discuss like what's going on. And I think that's how it kind of like jump started these conversations internally at the organization. So yeah, I mean, try, if there's some like-minded people on the staff that are interested in this, try to like have some, some, some talks and get together and speak on it. [Jacqueline Lo] 12:02:16 And then my other piece of advice is just to jump in. I literally just went on ChatGPT, went on Claude, went on to all of these things and just started playing around with it and seeing what the capabilities were, seeing what was possible. [Jacqueline Lo] 12:02:29 And that's kind of how I started. [Erik Ernst] 12:02:32 Excellent. All right. And Mary Anne [Mary Anne Lane] 12:02:36 Yeah, I mean, I think everyone's kind of hit the nail on the head as you don't have to dive in head first. You can take little baby steps and kind of get a feel for it. And I think everybody's shared opportunities to to do something pretty simple. But yeah, we also have a, at Georgia Public Broadcasting, we have an AI team. [Mary Anne Lane] 12:03:00 And then we all come together, learn new tools, and then discuss ethics and help define that line of define that line [Erik Ernst] 12:03:13 Excellent. Well, thank you to everybody who joined. Amber, is there anything else that you would like to add? I saw that you had your pitch because you already have another webinar planned for next June. [Erik Ernst] 12:03:24 I want to extend a great thank you to all of our presenters and an additional Thank you to our friends in the innovators group who helped us collaborate on this. [Amber Samdahl] 12:03:36 Yes, thank you everyone for participating and for joining and for the collaboration. I'll put in the chat again one more time the link to all the great notes links and topics discussed here today. So check that. [Amber Samdahl] 12:03:51 Out as well. And yes, there will be recording posted in about a week from now so you all can rewatch and re-enjoy. [Amber Samdahl] 12:03:59 The discussion. Thanks, everyone. [Erik Ernst] 12:04:03 Thank you.