Mistakes Marketers keep Making with AI and How to Fix them

Okay, let's get things started, shall we? Hello everyone. Just getting things set up here. those that are joining in, please feel free to say hi to everyone. Yes. Hi. Hi, David. How is everyone doing on this very fine Friday for me? Thursday for you guys. Wait, no, it's Friday, too. Sorry. For you guys. It's Saturday for me. Anyway, we're Yeah. How's that third dom? It's doing great. AI is uh making magic for all of us. Wouldn't it be nice? Well, I don't know if it would be nice, but like if you could just actually have a third arm uh and biologically have no downsides, where would you attach it? Have you seen like uh you know videos of of people trying to imagine if uh people could actually have wings like bird wings and actually there's no space in the back for it. So, I think the answer is don't think too hard. AI will have an answer like it always does. Just like this picture, it it has a way of figuring that out. Uh yeah, no, we're just getting it started really quickly here. Uh I'm just making sure that we're all set up correctly. The live stream to YouTube is on. People are able to hear me clearly. Uh welcome welcome everyone to today's session on mistakes marketers keep making with AI and how to fix them because that keeps happening uh whenever we try to set up you know our AI project sometimes we keep coming back to the same mistakes and it can be literally very costly. So, I'm just going to put this I'm just rearranging my slides so that I can see everyone. Yep. Yep. Yep. Okay. I put the slides here. All right. Um All right, guys. I am your host, Constance, product marketer here at Hrefs. Today, I've put together examples of how we use AI the wrong way and tips on how to fix these like mistakes that keep coming back over and over again. I've actually pulled the best practices from across the HS marketing team, too. So, you're going to get a ton of great tips on how to, you know, use AI to get uh the right thing that you want out of AI uh sooner and hopefully with less tokens. So today I have actually prepared more material than 30 minutes. Uh but the bulk of the mistakes and tips to fix are in that main 30 minutes. So we will cover that first and then later we'll step into things about like overall approaches to AI and um like a lot of it is also my musings based on observations like of how people say uh they use AI in their organization and and where things could go wrong there too. So, the bonus is that if you stay to the end of the 30 minutes, I know some of you guys have very busy schedule, but here's the bonus is that if you stay to the end of 30 minutes, I'll be giving out uh three one month free trials for agent A uh to all nonHF subscribers. Why nonHF subscribers? Uh you'll see by the end of the presentation. So, first 30 minutes is what I prepared. is the main content and then we give out the three agent a free trials and then we will cover the Q&A for any questions that asked along the way. So I'll not answer them during the presentation. Okay, for the interest of time as usual if you have questions please use the Q&A function that Zoom has. It's under more and you click on more in your toolbar. You'll see Q&A over there. Please ask the questions there because I am operating uh alone today. So I may not like um as the chat keeps going up I may miss your question. So to make sure your question does not get misplaced and you know ask it in the Q&A function. Okay. And yes um if you ask of course uh this session will be recorded it's currently streaming online on YouTube and the link for the recording and today's slides will be sent to everyone who attends today's session live. Okay. And this question keeps coming up again and again. So if you see someone else ask this question in the chat, my fellow um attendees to today's live session, please help me answer these people who asked if this session will be recorded. The answer is yes. All right, without further ado, let's begin with the first mistake and it's one of the most obvious ones. Let's go. Mediocre output from AI. Okay, everyone be riding a like AI slot, but no one wants to read it. This particular screenshot comes from one of the blogs uh of uh a particularly famous website that uh said we, you know, used AI to generate lots of new content and managed to steal lots of traffic from their competitors. If you actually read their content, it's terrible because it's not particularly helpful. uh it doesn't uh even though it's saying I think this particular page was trying to answer uh how uh Excel works it's like not really related to what their business does but uh in its answer it had no screenshots it had no links it had no um like I um useful formatting super vague stuff this is easy to generate AI these days and producing more content means nothing if nobody reads it uh readers is I mean tune out once it's obvious you use you um AI. Have you noticed recently that everyone on LinkedIn has started to literally sound the same? They'll say things like it's not just this but also this. It's it's X but it's not Y. You know um full fully amazing stuff and and their vernacular of everyone really just starts to sound the same. We've evolved beyond the M dashes. you start to see a lot of them uh show up less. But we've not evolved beyond the soulless way of talking about something amazing uh when compared to how I think the earlier way of social media used to be like but I'm talking about like a long time ago. But basically people are so obsessed with the way something sounds that they forgot about how it these so-called best practices just make you all sound the same. And that's really not the way to stand out on social media. Um, so how do we not sound the same? Because at Hrefs, you think that with with quality that means that we're creating everything by hand, but actually a lot of what we do today does um use AI a lot. Uh the biggest example that we share is about Ryan's content engineering workflow which and he's at the head of our content uh team here at in HS marketing and he basically a lot of his content now is um generated by AI but not in one shot. He basically brings uh goes through the same like stages that you would drafting normally as a human but you know give specific instructions at each stage about how to start from researching about a a topic to creating a content brief to drafting an outline. So now at every stage AI does a lot of that work and humans have oversight to make sure that the content brief is good that the outline looks you know well structured and it covers all the points they're trying to structure you just become the orchestrator at each one of the steps but AI helps to make it so a like after you go through each of these steps uh like when creating your content uh the quality really does become better because it starts to fix particular problems of creating good content every stage just like how you would if you were to create an an article in real life you know like not real life but like as a human right I've linked some of his blogs here I think the earlier one um talks about how uh bad content you know does not take you very far here at the bottom here and then the the second is his one of his later uh blogs where he covers the process of how he goes through and automates each uh each step. So you get better content uh and you can publish more consistently and they frees you up to like actually explore and about the actual content ideas are creating cool things to blog about. Uh yeah, so AI doesn't replace expertise. In fact, experts are best at maximizing what AI can do because it's only if you know what it takes to create something that's really great that you know the it like the specific steps that are required to make sure that quality comes out consistently. And so um I think the the best people in the industry right now know how to combine knows which steps require like can be automated or can be uh created to to improve the quality of something versus uh actually needing humans to go in and and um apply creativity. So this is a post that was shared about Ryan's like post recently which I think is very relevant here. And uh a lot of good marketing of course is great content. So great copywriting is only one part of the story. In the end, you can't engage an audience if you don't have something unique and original to share. So use AI to give you the human more time to find and make cool things. And we'll cover more about that at the end. Uh hopefully if we have time. The second problem is spending too much time and tokens. Uh I think this may be a common experience for you for anyone that has tried your hand at creating your own apps and dashboards for more than a few times for very serious applications where it you know matters to create something good for maybe your team internally or for a client. Um my experience was that I would give AI a prompt and then I would spend most of the time trying to figure out why that thing uh is broken. I would be asking AI and screen like um sharing screenshots or pasting like error messages back into AI and say you said this was you know working but it's not uh and you know when it fixed one bug sometimes it was it managed to break something else. So the so-called nirvana of saving so much time or like becoming the 10x uh productive person uh really didn't happen at the beginning because I was spending so much of that time just trying to fix the the thing AI was building. Uh here's an example uh I uh of this help article studio. So I um do some work to help uh update and create the new articles uh that you see for the HF's help center. And some of them are, for example, we publish, we literally have updates to our platform every week. And it may not be the most obvious thing, uh, but of course, we need to make sure that all all our help articles are up to date. They're not. I know that's not the the case, but we're trying to improve the ways that we can do this. Uh how long did you think these you know like identify outdated articles improve existing and create new articles and and like have personas to review the articles. How long did you think this system take to build? The answer is a whole week. The first time AI just it gave a plan. I think it like even the planning part of like how much work it should take to to build something like this took like a couple hours and and uh until today it's still not working like 100% where I want it to be. If you're not a programmer, I think or if you don't really understand how software works, you will make a mistake in how big a scope you're asking something to be built even to a person. And only people to who understand like how software really fundamentally works or even have that experience will have a better understanding of like nah this thing is like more complicated than you than you think it is in your head. um in like in in the uh just the planning stage. So when you can ask AI to say okay script out a plan of how to build something and yeah it will ask you questions of like oh do you want um a database which database do you want? Oh you should use like MySQL I have no idea what this means because I don't have that background. Seven implementation stages is what it gave me. like each document is just the different parts that we have to build something to set up because I want to say oh you know if this tool is amazing I can give it to my colleagues so that everyone can build like consistent help articles together. It became such a massive project. Uh and like I waited for a whole week basically to see the first version of a broken thing that I now am still fixing. Right. So it is so important um to a plan first. I mean this is first part of like trying to avoid this problem is to give AI the full context of what you're trying to build like try to go into as much detail as possible and this requires you to really try and visualize what's going on in the workflow of what you're trying to build. So like um this was a different application I wanted to to run uh for one of the earlier uh webinars where I covered building a dashboard for using free API endpoints in HRS. gave it information about what metrics I wanted, how they should be calculated on the platform and things like that, right? Uh and the second part is don't do what I did and try to oneshot give the whole scope of a massive project. Uh especially if you don't know what you're doing the first time, start with one thing. I could have just started with uh build like drafting help articles as just the first outline which eventually just became like a skill. I didn't need a whole application at all. I just built a skill to help build like a new help article from scratch or like a a text dump, right? And then the best part about doing it this this way is a you get something that's actually working in the very beginning to help you even though you haven't automated the whole thing. And B, uh, if you automate each step at a time and make sure things are working one step at a time, stringing them together in a whole workflow becomes so much more successful at the end. Um, one of the tips from our teammate Made is that you should prototype first. So, uh, I should have prototype first. Start with something that's not necessarily working at the beginning, but try to just um visualize uh you know put in like a sample data set or uh start with a part where where some numbers are even faked just to test the idea and and see the workflow and give it to people in your team to see if it's useful or not. I think in this case he was trying to uh build like a mystery shopper. Yeah. Uh walking through a website or something like that, right? So, so and then you can use cheaper LLM models as well. So that not only saves you money in actually building thing, you can swap to a different model that is better at creating something that fakes an idea to test it. Uh yeah, use planning skills. I mean I did it for that application that I showed you guys earlier. Uh but it's important to have uh use a framework like or or like a skill set like superpowers like in Claude. Lots of people use this uh superower superpower skill to brainstorm how to implement something. should be asking AI questions about hey I want to do this my goal is to be able to uh give this project uh to be hosted somewhere so that everyone can use it or I want to make it so that I'm using this application by myself and I want it to be as lightweight as possible like basically I don't want to deal with too much functioning parts uh how what are the best ways to implement something like this and u something like a superpowers uh for claude code uh can help you answer questions about how this has to be built uh in agent A. For example, this screenshot was when I wanted to create a PowerPoint uh presention uh presentation out of a text dump, which is what I actually did for today's presentation. Uh, and it's a screenshot of agent A basically asking me questions about how this wants to be built, like what do you plan to put in as content or uh how how it should look for each slide and things like that, right? The idea is that you should ask as many questions as poss well not as many questions but like ask at least some questions understanding so that AI has an understanding of what you're trying to do. Uh, uploading files instead of writing mega prompts. So yes, uh AI lots of the popular AI models that many of you guys are trying to use or are using has a massive context window where you can dump in like a whole novel's worth of text every time you talk to it. But that's not necessarily the best way to give a structured uh instructions of how you want something to be built. So, uh it's much more efficient to uh have like explain what you're trying to do, but also upload helpful articles, uh not articles or like markdown documents or CSV files or images and reference those in in discussing like what you plan to to be creating in its prompt to AI. Okay. So, it's a bit more like effective way of distributing the information in a structured way to AI. So do give this a try. Uh and planning and practice will get you there. Uh that's just something that you have to do. You simply have to use AI more to personally understand what it is that you that you need to understand to get AI to work more effectively for you. Uh by accident, I actually had to build a webinar attendee analyzer application three times. So uh I met I it's because we were just deploying uh AI access to the entire organization. So I started with uh building it lovable. Had a painful time with that. And uh mostly because you know maybe I was just not experienced or dumb. I don't I don't think like other people are struggling with lovable as much as I was. But uh I I built it first in Lovable. took me a really long time and it was not so like effective as an application. Things were like you know broken a lot of times. I was just constantly waste like spending my time trying to fix things. And then uh I tried it again in cloud code where I built it locally on my machine and it took me a lot less time. And eventually of course then as we started to build our own platform uh like AI building platform agent A, I then tried to do it again on agent A and it took like just less than two hours. It was quite amazing actually. Um, so and honestly I think if you're especially if you're not sure about how the thing you're trying to build is needs to be fully visualized, start with pen and paper uh instead of trying to go straight into an idea. Sometimes it's the best to just plan out what exactly you want uh to happen on every step of the way and then you have an idea. Do you have like a more structured thing that you can give AI? Uh yeah, I see your your points David about um like attachments. U we can get to that at the end of presentation. All right, mistake number three. AI makes mistakes and struggles to fix them. This is like a just a like a continuation of what I was discussing earlier about how uh it difficult it is to give AI sometimes the the exact thing you want uh to be fixed and it trying to struggle with you trying to fix that thing that you're complaining about. An example would be when you have created uh a dashboard that uses MCP connectors from other applications and you have no idea why the numbers don't exactly match up. there has to be, you know, something that's just not right. And uh and yet if you just copy and paste this to AI, like they'll be like, oh, you know, they'll just try and figure out what what is wrong and not necessarily get it right all the time. Uh and for example when I was creating when I mentioned about the creating that dashboard uh that uses free API endpoints for like an earlier webinar like I think last month or the month before um I asked it to use HFS MCP and refer to HFMCP for uh explicit ways it like explicit uh for creating the widgets that were on that dashboard that uses endpoints from rank tracker and brand radar. and you know any of these platforms that we're familiar with. And as I was debugging to see why the whole dashboard was just not working, uh the AI came back and he said, "Oh yeah, I figured out that uh you know these these parameters don't work." So he was just like hallucinating these parameters. They're simply not supported by the API endpoints at all. So even though I gave the instructions to use the MCP, it's not always going to work out. It really depends on a what you tell AI and b how well documented the uh MCP connector is going to be. So what I ended up doing uh is I actually went into our main platform or the documentation we have this uh like on our platform you will see in the different tools that there's an API button and if you click the API button it'll give you the exact API like URL call that you need to use to uh see the numbers that you see on that page itself that you're on including like filters or uh parameters like like if you filtered uh for example an organic keywords report for a specific words or like a minimum number of traffic and you click the API button those um filters will actually be air there in the API URL. So I use that and I gave an AI that very specifically. I want this please use this um API call and then build with that. Right? So um these are and then I asked the to give me uh information about like okay now that you build this widget please share how how you've actually fetched this data and save it as like a tool tip. So now I do this for all my applications because it helps me debug what's going on and make sure that the numbers are calculated the way I want. Um yeah and you can also test this by actually you know copy and pasting the API URL itself into like terminal or whatever to just call it manually and see what what comes out to make sure that the error is is on your end and not like something happening with the API. There are other u lessons that I've gotten from our team materian Glenn here. One is that you can um save what AI has learned next time from when it's one after one of your big debugging sessions. So that and in all those lessons it would like look at the the history of you chatting with it and create a bunch of like tips that would be helpful to avoid um as mistakes the next time you try to build something or within the same project. Uh another one is also just to create that manually as well. So you you would have some learnings right and you should keep that fresh in your mind whenever you're building something so that you can that to the next project that you take it you have a list of um dos and don'ts that you should be telling to AI whenever uh you ask it to build something similar or like solve a similar problem. So write write stuff down basically either you ask AI to do it or you ask like you know you remember to do it yourself. Uh make make a way to save versions of your work. Uh sometimes AI does very silly things and it will or like it will create something that is working and then uh it'll mess it up the next time as you add like new features to it or there could be something that you you had in the past that you removed and now you want to get it back. Um that's why lots of programmers have a versioning system like GitHub or GitLab uh to be able to fetch an older version of something that you've built last time. Um it's a good practice to do it yourself as well. Uh and a lot of these platforms have some free version that you can use so that like if you create something along the way, you're just not able to get it right again uh with AI, you can always go back to the older version and pull it back out, right? Uh an overview file. So that's why a lot of projects have something like a readme or or like a document that actually explains what you were working on and what this project is supposed to do. And this is very useful if you're building working on something uh with AI and then you go away for a while and you come back and you have you either don't remember what's go what was discussed or or what you covered in a project or AI doesn't remember you've closed that session chat window and now you have to um you have to ask to go back from the beginning to figure out what it is that you built. Instead, if you have a a helpful document that already has that document in from the last time you're working on something, that becomes very useful to save time and tokens to to get back to the context of what exactly you were building last time, right? And what problems you trying to fix last time? Okay. Uh okay. All right. We are three minutes away. So, what I'll do is I'll quickly cover agent A, right? And then we'll cover this section when we have time. for for those people that have to step away they can and then I will u cover this section a bit later. So let me try to go to uh agent Agent A really is a platform that as we continue to use AI so much we decided that and all the best practices of how we should build things for marketers should be on one platform ourselves and we can then bring it out to other people as well. uh why should you use agent a one of the big things is that it has direct data connectors to the ATS platform and I don't mean the API I mean actual um built-in functionality from the platform that we don't always support with our API which has an abstraction layer and it requires you to uh understand I mean it requires our team to build a specific way in which you can pull data based on the the the project or or or like the the platform that people have asked for. So you're not limited by what uh our HS API supports it. You can use a lot more tools and a lot more ways of calling the data or at least AI has direct connectors to it. uh and a less much less hallucinate basically no hallucinations because it's so built in into AI into the AI of the platform and b uh using these connectors to build your dashboards or create your apps or whatever it is doesn't spend any of your accounts API limits uh your HFS accounts API limits so that's the extra point uh pre-built skills so uh we have a team of both marketers and engineers um updating this like long list of different skills that help you build uh useful dashboards or perform analysis for like AEO or or just SEO or analyze your uh communities that are mentioning your brand things like that um that are functioning and working and it's something that you can call as a skill anytime and you can preview it as well to see what actually does and then launch it for your own website and it's I mean this really will not only save you time it will literally save you both money and time because like you're starting with something that already work works rather than trying to start with an idea and then like fix it and debug it till you get to a working part. Now that's money that you and and time that you are spent uh saving for sure and of course like we have people who have deep expertise about uh what are the best practices to do SEO and AO building these skills. So like you're you're getting that sort of uh functionalities as well out of the box. um not only do you have the skills to create those dashboards or or analysis reports and stuff like that, there's also a lot of quality of life skills. So, this is another reason why when I was building like moving a lot of my AI projects to agent A that I spent a lot less time because there's a lot of uh built-in skills that you don't see. This is like you're not going to see this on the agent A platform. This is something that I've I've found um our team found out and pulled out. simple things like, you know, um fetching uh from the web or like being able to export what you have to a PDF or how to, you know, set up different jobs that are running consistently uh based on what you tell the agent to do. These these are things where if you don't have them sort of set up uh ahead of time and AI has to figure it out for the very first time, it's going to waste a lot of time and money getting it right. So um having those pre-built inside um without you worrying about it is is really really useful. Uh this is not unique to agent A but you can share your project to either members of your workspace like your team, your marketing team or uh anyone. So I've used uh agent A to build a platform or a dashboard of different apps as a showcase and I I shared a readme version a readonly version uh out as a public website. So I was able to share literally share a a demo to like attendees to a workshop that I that I was doing for HS Evolve Singapore last week. Right? That's something that you can do and and it's just a few clicks to set that up. So like it it saves a lot of um work with actually making something that that is useful for your team or for your your clients for example um up and ready uh within the platform. Uh you can choose different LLMs. So that's another thing that that we covered earlier about how you can save time and money is that you can switch to a specific LLM that's cheaper to do things like planning and prototyping or creating rough versions about things. And then like once you actually want to implement something for real and get the details down and functional, you can switch to more expensive LLM and you can just mix and match uh based on the task at hand to get sorry to get the outcome that you want, right? So it's just really convenient to do this in a single All right, we're at the part talking about free trials. Are are you guys ready? So um why I say that I will give out free trials only for people who are not subscribers because HF subscribers already get uh a one month free trial if you're a paying subscriber. If you go to uh agent A, let me just put that in the chat. Now you should see this this this message at the top that says you've been selected for a free month of agent A. Everyone selected actually as long as you are a a paid subscriber of Hrefs you will see this and you will get one month free then you can start a free trial. Okay so this is really for everyone who does not already have a paid subscription for hrefs right are you guys ready and for people who um for people who are not um like for people who participate in this I can check if you're a subscriber or not. So, please, if you're not a subscriber, please don't um participate because I can check that. All right, fastest fingers first. First three to say agent A please in the chat gets a free month of Agent A. Go, go, go. Agent A, please. Yes. Follow the instructions carefully. We have Lucas and we have Sara and Gloria. There we go. And that's it for the first three people. Um, you will be getting a try. Can you please uh private message me in the Zoom chat? So, select uh instead of messaging everyone, please select host and panelists to uh DM me your email and I will check to make sure uh that you are not a paying subscriber and I'll send you guys uh the the the code for the first month free. Okay. Yes, thank you for sending it. Uh yes, so far you need to send me your email. Sorry, David. Um, but we if you join more of our uh events coming up or like future webinars, we'll be giving out more giveaways every time. So, or you just have to subscribe for hrefs. Those are your choices. Okay. All right. Thank you everyone who um the three of you guys for sending in your stuff. Okay. So now go ahead into Q&A which I think currently only has one question. So we'll cover that one question and then I'll cover the last section of the presentation which shouldn't take more than like 10 minutes. Okay. Oh, Lucas said he's actually already subscribed. So the next person uh to say agent A please. I'm not gonna move. All right. David got it. David got it. So, so, so, um, so, uh, David, can you please, uh, private message me your email address, please, so that I can send you the codes. All right, awesome. Uh, all right, let's answer the one quick question that is actually from Sara that people have from this presentation, which is, where are Ryan skill files? He's not going to share you the actual files themselves for you to make. Like that's the that's the like the gold for the HS content team about how we create our our articles. He's not going to share the the actual uh I think you're talking about this one. Let's go all the way to the beginning here. Yeah, he'll show you how he built them, but uh he's not going to give you the actual skills. So, he didn't actually put them on on on the blog. Uh but I I think I think that's fair because otherwise it will be too easy to create what HFS creates. Well, maybe 75% of the way. Uh there will be a limited uh version of creating content or engineering content that will appear on HF's uh agent A as one of the skills later. Uh but you will not you're not going to get all of this. I think even that limited version though is is kind of helpful. I've I have made a very shortened version of this pipeline myself for for help articles and other uh content like case studies as well to create it doesn't I and of course I've adapted a a separate process out of this because it's not the same it's not like trying to do the same thing as a typical blog is it's not as vigorous. So so I was using that based on his uh uh the skills here. Um but yeah, I think it is useful to reference and try to build one yourself. You'll be surprised how far you can go now that the you know the thought leadership and and ideas are there if you build this yourself. yeah, what are some of the slides we are not able to cover first 30 minutes? Let's let's quickly dive into that. So it's it's just a quick section that I wanted to talk about uh like the whole idea of using like marketers and using AI and I did put the caveat at the beginning that this is a like personal opinion. It has a number of hot takes and it may contradict what some uh thought leaderships like thought thought leaders are saying on LinkedIn and other social media platforms, right? And um and like a lot of times you see AI being u really just pushed as as a way to make you like the the most amazing person ever. And like AI is is this magical box that uh can create uh amazing things out of thin air. But what I really see it is like a way that you should use it to free you to do creative work without you actually needing to understand how to uh program and do specialist specific tasks you know rather than thinking about how much of a 20x 30x person you have become it's more uh it becomes better to visualize it as like hey AI can do this like thing so what can you then do as a human being right Um, so good things that are that can be automated with AI are like like menial tasks like creating you know issues in in Base Camp or Jira or any of these like project management apps um sending like very um emails that do not you know that they're just good to have or creating uh short codes and other formatting things in your articles. AI is very good at doing these like specific tasks that you know are just they're very structured. They're very repeatable and they are in fact humans are not that good at um overall at creating at doing some of these repetitive tasks. So we keep making mistakes and stuff like that. Uh but AI is pretty good at this. Transcribing videos, uh extracting structured notes from long form recordings, like things that uh can be automated with a small script uh that uh we usually, you know, for for most marketers that don't have a programming background can't really do, but now we can do uh with AI. That's AI is great for that, right? Uh but it's not great at like developing a marketing strategy. And I don't mean like oh take this like uh discussion dump from a meeting and develop a marketing plan. I mean, AI is not going to have an amazing idea of how you're going to grow your business, you know, the next 20% or or access your audience, you know, uh or in a unique way that your competitors are not doing because it's just not very good at creating new ideas or or brainstorming new ways to approach uh that that marketing can be done. they literally it's built on everyone else's existing pre pre-made ideas in the past. So, and it's not necessarily going to get the nuances right for what your organization needs or what is working well for your industry. So, don't ask AI to do this, you know, from from the get-go and expect it to have great results. Um, reading and learning new things. So yeah, I I even though it is useful to use AI to maybe shorten and help you read articles or watch videos, uh I don't give personally I don't outsource all of it to AI because I think the very skill of being able to digest content that people create out there especially and and discern whether or not they are you know useful or not really is something that humans are good at doing and and it's a skill that I want to preserve and not like outsource that part of my brain. to to to AI uh deciding who you should hire and fire. I've just added this in because like I see now people are going oh like I've replaced my team with uh uh with AI or I have to decide how I should restructure my team and what they should do with AI. And I feel like these are just sort of it's not just how capable something some people are. It's it's all the nuances right of how to manage a team. It's it the AI is not going to be good at this. It's still not good at this. So, and do you really want AI to automate this part of your job for you? It's it's like you're supposed to you wanted AI to do like make a bad version of something that I think is a very uh expensive and very powerful capabilities that humans have. I feel like this this is the reason why lots of the the people at top are paid lots of money. You don't really want AI to automate this. I think it's not there yet. So uh another part is that I want to talk about is how AI really is useful at the parts that are the most boring. So this is of all the applications I built so far this copy optimizer like thing that I've built where I type in something I want to say in social media or uh for this webinar like if I wanted to create uh like just a short write up about what this uh upcoming workshop is about or if I'm interviewing a guest uh this application is the one I use the most. It's super boring. It's it's not something that you can um you know be really proud of on on LinkedIn or whatever, but it's the one that actually you know drives the most value for me, right? A lot of the flashy demos out there like if you see they're not getting into a lot of detail about how they're being built, they're probably not that impressive actually. Uh it's so so don't don't go for for things that like look super amazing or oh yeah, I'm going to really just spend all my time watching Netflix if I have this, you know, with my team. Uh this is the stuff that actually drives drives value for you. Um and and to add on to that, a lot of I just found all these people that are now adding uh this tagline to to their profiles or their resumes about with AI, they've become a 10x 20x. This doesn't mean anything. This this, you know, now that AI is here, it doesn't tell you how you are more effective at your job. And it it's it's not very concrete, but it drives a lot of FOMO for people who who are I think everyone uh in in the space when they're talking to each other in the community are really worried about being left behind. And I think like this is not this whole practice of trying to have this persona becoming so much more effective uh with AI is not very useful and actually adopting AI in a useful way and I don't think we should be hurting each other like this. I think it's AI can be such an amazing uh helpful thing to adopt that like just just creating this kind of uh fear really amongst each other doesn't actually help us adapt use AI more effectively in our jobs. Uh so rather than trying to follow this whole like trend about being this 10x person and trying to push this persona, it's so much more useful to actually start building and you know be um openly talking about like the mistakes you make and and like sharing your insights with others which is not new whether AI was there or not. Yeah. Uh many of the best users of AI are posting all day about it. Uh here's an example of Samo if you you many of you will know him from our YouTube channel who has created like this uh keyword clustering internal linking suggestions tool where not only you can see the different pages and how they're linking out to each other but but also um see the individual connections as well like it was such a cool thing that he has built is he's just busy trying to find like an effective way to to to solve a problem and AI is just a tool so like you know he's not posting about it on LinkedIn or or X or or whatever. Um he he is really building cool stuff out there. And this is an example of uh Eric's who uh leads our international team uh and how he's um tracking the performance of our content localized in different markets. He's also extended this to like our the paid ads that we run on different platforms or uh the the um activity we get on platforms that are not that are unique to that market. So um I think in I can't I'm blanking out with the examples right now but like not everyone lives on Reddit for example the the lot of the world is really on different parts of the world uh web and being able to combine that performance together into one dashboard will help you give a real understanding of how different audiences that don't use English um interact with the HR brand. Uh AI can be a pretty good teacher reviewer. So uh for example there was one time uh last week or actually like a couple weeks back when uh for HS evolve Singapore we were working with one of our sponsor partners to create a uh workshop title and description based on what he wanted to present uh and I it was he was sort of like struggling to get something that was really working. So I was helping him with that. I asked AI to critique it, not not just create like a, you know, just give a a few iterations and I'll pick the best one. So, I gave uh a portion of it was something that AI wrote and then I I like edited it and I gave it back to AI and say, can you just give me some um advice? And it did. And the advice was really good. It said, you know, your readability is very verbose. Uh, a lot of people are going to turn off before they're going to finish even reading the the end of the title or the description. You're missing some information based on what your the original content included, right? Uh, here. And then uh this is great because the actual suggested rewrite uh rewritten version based on the feedback was terrible. Like this was not it was not really uh following up on the actual advice that it gave, which is kind of funny. uh but I used the feedback itself to create the final version and uh which was mostly handwritten and it turned it was the final version that we we all liked and we uh published that on our platform and that's what our uh that that's what our workshop person followed up with at the end. Okay. So, so yeah like don't just ask AI to give you stuff. You can ask AI to be uh you know uh critique as a expert in so and so field or even uh get a few best practices that you've researched outlined together and build that into persona and ask AI to critique uh your work based on that persona that you've built together. Okay. Yeah, that's it. That was mainly what I want to present at the end. Uh are there any other questions? So, we're back into this that we saw earlier that people have for today. Anyone else have any questions? I'm going to remember to save some of this uh emails that people shared earlier. I think people are good. The slides will be sent to you guys. I think either tomorrow or the day after basically over the weekend uh uh in the follow-up email from you guys. Yeah, I think people are ready to go. Uh Lucas has one final question. What are the other benefits of using agent A? Uh so I I guess I I can quickly just go through the slides here again to talk about why agent A is worth your time versus like claude or some other uh general uh AI platform is the pre-built skills that are made for marketers that help you already create some um like perform analytics or create dashboards out of the box and we will keep adding to this as well. So like it would also start to include like workflows or connectors with your WordPress and other uh thirdparty tools that marketers use a lot uh and all these quality of life skills as well as like uh the an easy way to share what you've built and based on specific permissions. So you want to share it to a specific number of people, you can. If you want to share it to everyone as like a website, you can. Uh and choose between different LLMs. Those are the main things. Uh Jason D asks, "At the end of my prompts, I add any questions before answering." Oh, yeah. So, that's like a good tip. So, whenever you you ask AI to do something that's a little more complicated than a than a simple task, it's you can tell it to ask any questions to clarify the task. Yeah, that those are always good ideas. That's also what I do sometimes if I'm not too sure with how things should go, I'll just ask like, you know, please please suggest um questions that you may have based on what I'm trying to do. Yeah, I think people are good. Uh people are really uh there have no more questions left. Uh if you close out this uh today's uh session, you will get a survey for you to maybe share some feedback about what you experienced today. Please fill out that survey. It should take only a few minutes to give me tips on how I should improve sessions. You can also give some ideas about what you'd like to see next. Uh the next time we do this, there's always more requests for uh information about how to do a AI or how to use AI or um you know best practices around this case studies and stuff like that. We did uh like we walked through a case uh of of building a dashboard last time and like we would like to hear from you guys what else you'd like to see. More specific ideas will be super helpful. Um yeah. So then I will go ahead and end today's session. Uh thank you so much for joining us today. I'm going to head off to bed. It's 1:00 a.m. for me here. So have a happy Friday and great rest of the week and I'll see you guys in the next live session. Take care everyone. See you. Bye-bye.