Product of the week
(For a summary, click here.)
John Eisched: Okay, just for starters, just give me a brief intro to yourself, your background. It doesn’t have to be your whole personal origin story, but just some idea – like a quick elevator pitch.
Samantha North: Yeah, sure. So, I’m about to launch an AI automation agency which will be called Emigre Systems and that’s coming up at the beginning of next month. I’ve been in SEO and content marketing for about 10 years or more and that sort of led me to this stage – using more and more AI; completing a PhD in data science, which also pushed me more and more towards A; and then finally realizing that businesses were crying out for automation. And when I’ve automated my own previous business it really woke me up to the fact that this stuff is not going away. It’s really going to make a big difference for business owners. So, right now, AI automation is my main business focus.
JE: And that leads in well to my next question. How does your PhD in computational data science – how do you put that to work in using LLMs? I’ve noticed that you’ve done a lot of posts about how to get SEO optimized content. You also just mentioned automation. So, just kind of in very broad strokes, how does your PhD help you in that line of work or with those specific tasks?
SN: If I’m honest the two things are quite different. I would say where it helped me is more towards sort of pointing me in the direction of that and giving me an introduction and allowing me to work on projects that opened my mind to this stuff. I did have to do a fair bit of statistical stuff with Python and R during that academic work which I didn’t really enjoy and I think I realized during that phase that I really don’t like coding and that no code automation solutions were going to be the thing I would focus on. So that was an important – to weed that part out because it would have been hell otherwise. The project for my thesis was about analyzing large social media data sets from Twitter and we were looking at tribal political conflict around issues like – Brexit was the main focus. So we used data from about three years over that vote and after it and we were looking at what the two sides of the debate, the culture war, how they were interacting with each other and what that looked like and how disinformation was potentially spreading on Twitter through those groups. So it was very topical and that part was valuable. But in terms of practical skills, I think because I’ve jumped so far into the no code camp now, I’ve left the Python and left the little , bit that I learn far behind me now. So thank goodness because I’m more of a journalist than a coder.
JE: Yes, I sympathize with that. But given your background in statistics, do you feel like you can see under the hood of some of these models a little bit better than others? Because a lot of what they’re doing is kind of doing an analysis of what’s most likely to come next or what is most likely grouped together. So, do you feel like maybe intuitively that might be kind of working in the back of your mind a little bit or has it shaped the way that you thought about the problem?
SN: I guess so. I mean, yeah. I mean, I’m not intimidated by them. I feel like I can master them with ease, whereas maybe a lot of business owners are still kind of intimidated by how to even interact with an LLM. But I think they’re at a stage now where once you dive in and start prompting, once you know how to prompt and get used to this interaction with this machine then it just becomes very intuitive with practice. So I don’t think it’s something that really requires an academic background. I think for me it’s just a nice to have and I’m lucky that sort of it pointed me in this direction, but really it’s a matter of just getting in there and practicing and prompting and learning more about how to prompt.
JE: Yes, that’s – I was also I also wanted to go into that as well. A lot of your posts go actually much further than the typical “How to Prompt” kind of thing, and you actually go into multi-shot prompting, or breaking the questions or breaking one prompt up into multiple ones and kind of putting that together. And it seems like you’re more focused on people understanding the tool and how to use it rather than just pumping things out. So is that like a niche that you found lacking? Did you feel like there wasn’t enough of granular detail in the way that people were saying that people should be prompting or did you feel like something needed to go a little bit deeper especially given your background and background in SEO.
SN: Yeah, absolutely. Yeah. One thing that really stood out to me was that at the beginning, the last couple of years, a lot of people were jumping in and using AI tools to create content for blogs and for websites and for SEO, right? But a lot of it was trash. And I knew that it wasn’t so much the AI tools fault. It was the fact that people were using very very basic prompts like, write me a blog post about dog training or something like that. And I knew already at that point that, you’ve got to give it a lot more than that. You’ve got to – you can’t expect it to do all of the work for you. It does the heavy lifting, but you do have to give it a decent amount of examples and research. And you should review the piece of content after it writes it for you. So, I think there’s this whole camp where people just do mass publishing without really much oversight and much human input. And I still don’t feel that’s there yet. Even now, two years later from when I started this stuff, it’s not there yet,
JE: There are some people who think that using AI to just turn out SEO stories for SEO is fine, because you’re using machines to write for machines. Where do you fall in that camp? I mean, even if you’re not like posting it on social media, just getting that sheer volume of pages out there. Do you think that’s of value? Do you think it needs to be tailored a little bit more?
SN: SN: I think, yeah, ideally we should be a little careful right? So, back in 2020 during my PhD, I actually consulted as a counter disinformation analyst online and although a lot of the stuff we looked at was social media we also looked at what was coming back in Google search results and what comes up at number one, two on Google is what people take as the truth, right? That’s because attention spans are so short now people don’t often dig beyond the front page and if there’s something ranking at number one that’s out-of-date or false or deliberately misleading then that can be very problematic. So I think having had that experience sort of before I really got into SEO properly, I always was cautious that we have quite a big responsibility if you’re going to rank something in number one in Google, then the whole world can be seeing this and taking this information as truth. So we really have a responsibility to make sure that it’s accurate and timely. That’s how I see it. Yeah.
JE: Okay and that leads into, how do you feel about the term “Generative Engine Optimization” or I believe you called it “Search Generative Experience?” There are a lot of acronyms floating around and I know that there’s not really agreement within the industry about what they mean or even if they’re all that useful and I’m just kind of curious about what your thoughts are about the difference between SEO being used for search, the difference between the Google searches, and the difference between querying an AI or an LLM in terms of the information retrieval. Do we know how much of a difference there is between the two? You’ve also mentioned things in some of your blog posts about getting listed in business directories and professional organizations in order to boost your SEO. Are those factored into responses that we get from AI or – I mean, it’s kind of everybody is like, “We’re dealing with a black box and sometimes it gives us what we want and sometimes it hallucinates and doesn’t.”
SN: Right. For sure. I haven’t got a definitive answer on that yet or all of those things like directories and anything that is giving weight and social proof, like a vote of confidence on the content reviews, particularly verifiable reviews on things like Google, I guess, but also Trust Pilot and that kind of thing. That’s going to be valuable because that is something that’s difficult to fake, right?
JE: Right.
SN: So that’s going to be a big factor. I haven’t quite figured out the exact factors that push things up on ChatGPT results as opposed to normal Google search. I don’t know if anyone has a comprehensive thing yet, but I do know, my own websites, I’ve noticed that stuff I have that ranks in page in position one is often coming up in the generative results as well, so . . .
JE: Okay.
SN: There’s definitely . . . yeah.
JE: Yeah, I know that Perplexity, they have published some stuff where they will rank certain sites as being more trustworthy than others. They’re not exactly clear about exactly which sites those are and I think it depends on the topic but there is some indication of what AIs and LLMs are paying attention to.
SN: Yeah. It’s still unfolding. I think that one, and I’ve moved away from SEO a bit in recent months, so not keeping as much of an eye on it, but I will be very interested if somebody produces a definitive guide to how to rank in these engines. That will be good.
JE: They keep changing them because I did see your video about a change in Google’s algorithm and how you survived.
SN: Yeah, I had some bad moments, but you know, it always seems to come back, but it can take quite a long time. I think the worst one was September 2023. I can’t remember even the helpful content update maybe. Yeah, they wiped out a lot of people. I had a blog on Mediavine at that time that was making good, nice money from ads, and then it got almost obliterated and so did so many other bloggers. So, that was a brutal one.
JE: Um, let’s see. You have said that part of your goal is to “remove friction from content creation unquote.” A lot of people approach content just by outsourcing it whether to a freelance writer or an AI as is often the case. How did you arrive at your approach where you’re not basically going to do the work for people but rather make it easier for them to do?
SN: Yeah, the importance of expertise is really what’s driving that when you’re creating content for a client, let’s say a tech company for example, there’s always this emphasis on interviewing subject matter experts – and you would know that as well, as a journalist – that’s so important. But in this way I’m doing it, the business owner can be their own expert and all they have to do is something simple like voice dictate something into a note and then let the AI handle building that out into a piece of content. So in terms of removing the friction, yeah, they don’t have to sit down and tap out an article by hand, which no one wants to really do anymore, right? They can just share their messy idea and then automation can pull that into the workflow and then build it out behind the scenes to a properly structured piece of content. And that I think is really removing the friction. It’s allowing those ideas that are fresh in the business owner’s mind to just get out there publicly without all the donkey work.
JE: Right, and the automation just kind of greases the wheels, keeps the gears turning.
SN: Yeah, it’s great for that. I shared one last week and today for blogs and YouTubetubes that just take the long form content and then go through four different routes and chop that up for different social media platforms. And that’s a really good way to use it because you’ve already got all the meaty stuff in the YouTube transcript or in the blog post, right? You just need something to chop those up and make sure they’re formatted to perform well on those different social platforms. And that’s a really nice use case, I think, for AI automation.
JE: And one thing that I think some people overlook is if you’re using an AI to generate content or to summarize things or to tweak it a little bit, you can just tell it to also include an SEO caption and a social media caption at the end, which kind of gets everything, nice and tidy together into one result.
SN: Yeah. Yeah. The main work is going to be, in building the blog post or the video, making sure they’re tied to keyword research, making sure they’re optimized for search. But I think once that big portion of work is done, the rest of it is really easy to get. You can do it in seconds, have it ready for Facebook. And Facebook, it’s not necessarily as important. I don’t think it’s as hard to get to pull in views on Facebook as it is on Google to rank number one in Google or on YouTube. That’s hard. That takes work.
JE: Another concern that some people have expressed is that it’s getting so cheap and so easy to generate all this content. Are we getting to the point where people are just kind of like going to be saturated with social media content, blog posts to the point where they just don’t want to check it anymore? I know some people already feel overwhelmed when they check LinkedIn. Do you think that there might be – for lack of, I wouldn’t say a data horizon – but do you think people are getting saturated with the amount of content that they have online? And a related question is whether or not, with everything automated, is there some sort of like digital dystopia in the future where it’s just a bunch of computers talking to computers and there’s just our little profile file pick next to it and that’s about it. I’m just kind of wondering. I’m trying to establish how far this could possibly go in terms of content generation.
SN: Yeah. How far could it go? You know, John, there’s always going to be an increasing swamp of trashy content, and that’s going to be that this is going to get deeper and deeper, I think. There’s always going to be that stuff for brands and . . .
JE: I think people are so obsessed with social media that they’ve forgotten about soap operas, which were the daytime.
SN: Yeah, I remember those – East Enders 2003. But I think for brands and individuals who want to go beyond that, yeah, there’s still space if you know your audience, you’re very specific about who you’re serving, you make an effort to find out what their problems are and what their goals are and you produce content that’s realy targeted towards that stuff. You engage with people in a human way. One thing I think is so important is okay my LinkedIn posts, right? All of those for the last three to four months have run through an AI structurer, right? But I never use AI to respond to comments because that’s where I think that it’s nice to actually chat with people, to engage with people, to build trust and camaraderie in the DMs and in the comments. So while I always post automatically now, I never comment automatically. And I think that’s an important distinction. It’s about kind of having that sense where to use the tool, where to use automation and AI, and where to be human. And there’s a lot of nuance in that.
JE: Right. Can you give me a few examples of what those nuances are? I mean, obviously the comments and the content creation, but maybe some other situations like people are turning into customer service chat bots and I’ve heard some nightmares about those.
SN: Yeah, I tinker with a few chat bots, but I don’t know. I haven’t really gone deep down that rabbit hole yet. Not sure how I feel about it, but in terms of what I know best, which is content. I mean, let’s say you’re writing a blog post about your specialist subject, it should have things like anecdotes, your opinions. It should have your firsthand experience of how something works that’s your unique process of doing the thing. And you can use AI to put the sentences and paragraphs on the page, but I think you should at least give some bullet points or something, into your AI from head, from your personal experience, so it’s got something human to work with. And I think in that sense, you can get away with automating a lot, but you have to have some human stuff in the first place for it to work with. Otherwise, it’s going to come across as very generic and bland. Not sure.
JE: Yes, and I think often the distinguishing factor there is the level of detail that you go into or if you’re able to describe physical processes that actual humans do in the real world. I’ve also noticed that AI doesn’t always use metaphors or understand them, so those are other things that you can put in there as well to kind of . . .
SN: Absolutely. Yeah. He uses lots of imagery, but in a very cheesy way, over the top. I have to tell it.
JE: Yeah, it’s nothing that I really wanted to remember.
SN: I’m always dragging those things out and deleting them. And I mean, I’ve got prompts now that kind of get past that, but sometimes they sneak in still. Mhm.
JE: So, as I mentioned before, I’m also publishing the full text of this and I want to know what your take is on how that might affect SEO or the results that people get from AI. Because, initially, I kind of wanted to do it this way because I’m a big fan of Interview magazine from the 80s and then I thought it was a very good way to get additional content out there for SEO and for our interviewees to get some additional information out there. So having said that, basically I’m asking you, “Is this a good idea or not?” Do you think it will have an effect?
SN: So, publishing a full transcript?
JE: Yes.
SN: Just verbatim, as is? Would that be within a blog post or how would you publish it? Just on a page on the website?
JE: Yes, “Here’s the full transcript.” “Here’s a link to the summary.”
SN: I mean, I don’t think that would if it’s not really pointed at a particular keyword, it’s probably not going to rank for anything. But that doesn’t have to matter. You can get traffic to it from other sources, right, without looking at the website. It’s difficult to say, but yeah, I think with this kind of interview content it’s not really that SEO friendly, because you’re not really answering a specific query right people aren’t going to be searching for “When was Samantha’s interview with My Marketing Pro?” It’s more timely stuff. So yeah, I would say I don’t know, publish the transcript and then promote it through LinkedIn to other social platforms. That’s what I would do.
JE: Yes, I am doing that. I’m also hoping that it’ll be beneficial to the interviewees if somebody ever looks up their name on AI or Googles their name. I don’t know where it’ll fall in the results, but it’s basically more content out there and I feel like it’s a bit more accurate.
SN: Absolutely. interviews are always good for building that brand authority and showing that we’re real people. So, yeah.
JE: Yes! So, what is one thing that you think most people miss when they’re using AI? Or is there something that . . .
SN: . . . I think they missed. I’ve got something. It just came to mind. It’s sandwiching the AI between two slices of humanness. So to elaborate, you start with your own ideas and all your – I don’t know – your opinions and expertise and your angle and all of that stuff. So that comes from the human, the starting point. Then, you let the AI do its thing, structuring, creating, building, all of that. And then the second human part is reviewing the content, checking for accuracy making sure there’s nothing crazy in there. And then you go ahead and publish. So, I like to do it that way. They do all the grunt work in the middle, but we have the human on each end to keep things under control.
JE: Yeah. Some people say “the human in the loop,” but this is kind of like the human at the start and the end. It’s more like . . .
SN: Yeah, I think it’s kind of the same concept. I posted about it on LinkedIn, I think yesterday or the day before explaining this. I just think, you can’t rely on them to do everything. You just need to be there as a guiding force if you want something of reasonable quality.
JE: They’re very fast and immediate assistant and in some ways a researcher to give you a first draft, but you still kind of need to do some revisions.
SN: Yeah. Yeah. Although, I mean, I’ve noticed that I haven’t tried 03, o3, the newest chat GPT. I haven’t had a chance to try that yet. That’s the most advanced model. So, that’s something to try. I usually use the newest version of Claude’s LLM for most of my writing. Reason being, I’ve always found that Claude, the way it responds is more human and less flowery. I find that there’s less forcing that I have to do to make it sound like what I want it to sound like, right?
JE: Yes.
SN: And I like that.
JE: Yeah, you’ll probably agree that Claude sounds a little bit more like a journalist than any of the other chat bots.
SN: That’s why I like it. Did I tell you I did journalism for a couple of years?
JE: Yes. Yes.
SN: That’s probably why I like it.
JE: I did some freelancing for quite a while and I went to school for it and I was an editor for a bit. So the next question might be a little bit difficult, but I’ll try to phrase it – the [Oxford University Press’s] word of the year this past year was “brain rot”, which refers to just scrolling through meaningless videos online, not really remembering anything that you’ve seen, and just kind of being in that doom scrolling daze. Do you have any suggestions for content creators about what they can do to counteract that?
SN: The doom scrolling. Yeah, that usually happens on Instagram. Avoid Instagram and Tik Tok. Content creators on which platform, though? Do we have a platform specific thing here? Because I think there’s a difference.
JE: Oh, the term wasn’t used to describe any specific platform.
SN: For me, I doomscroll most on Instagram, so I try and cut down my Instagram use. So, I don’t really doomscroll on LinkedIn because I find lots of valuable things on there that stop the scroll and I actually engage. So, I guess my answer would be, “Know your audience well. Produce content that is thoughtful and helpful to them to stop them scrolling and make sure it’s structured and hooked, especially the hook – make sure it’s structured in a way that gets their attention. Right? Hooks are important. They might sound like cheesy things, gimmicks, but they actually do work and that’s what’s going to stop the scroll.” That’s what I would say. That’s what works for me anyway. If I find something that’s resonant and helpful and I know that I need this in my life, then I will stop scrolling and I will engage with it. So, I can only assume the same goes for other people, too. Yeah.
JE: Okay, I think that’s kind of a good stopping point and we’re getting close to the half hour. Is there anything else you think to add?
SN: Anything else to add? I guess a bit of advice maybe for people from my own experience is that there’s so much AI stuff around at the moment, right? Every day, there’s a new thing, a shiny object, and a new tool, new model, whatever. I think don’t get too distracted and don’t feel overwhelmed and behind being left behind. You know, just pick the ones you like best. I’ve been using Claude for two years and I haven’t found one that I like better. I do sometimes use different chat GPT models. I’ve tried DeepSeek, but nothing has really stuck because I like Claude. I feel like I’ve mastered it. So, you know, pick a few that you like and just get used to them and don’t feel like you’re being left behind because there’s always going to be a ton of new stuff. And yeah, you’re not going to be able to keep up unless this is your daily work and even then it’s difficult.
JE: Yes, it is. I’ve kind of gotten to the point where I target everything and think that just don’t cover everything. Just kind of stay focused on the one tool that you’re looking for and see if it’s out there.
SN: Absolutely. Although I did discover an amazing tool recently called Lovable. Have you seen this one? Lovable.
JE: Yeah, I saw your post about that.
SN: Very cool. Yeah. So, I advise people – if they hate coding like I do, but they still want to ship things – check out Lovable because it’s been a game changer for me in getting things out there fast without all that procrastination and having to code, which is a nightmare.
JE: Yeah, I kind of have some questions about whether some of the other things that it’s capable of, but I can reach out to them for that. So, all right.
SN: Yeah. No, for Yeah, it’s cool. I like it.
JE: Okay. I guess we can stop it there. You got something else to add? Okay . . . .
SN: Anything else you want to ask?
JE: No.
SN: I think I mean there’s lots I could say, but is there anything else we haven’t covered that you wanted to squeeze in or brush?
JE: Well, I was kind of wondering about your thoughts . . . . You did mention your YouTube presence and all that. You did say something about making sure that people know that you’re an actual human out there and all of that. Do you feel like AI avatars are going to further jeopardize that level of credibility?
SN: Yeah, that’s a great question. I really want to build a workflow with AI avatars in it. And the only thing that’s been holding me back is that I don’t want to pay for yet another subscription. I got so many. But they’re very good. I would say my feeling is that they are better for shorter form content. Let’s say I got a HeyGen or a Synthesia subscription and I did a personal avatar, I would feel more comfortable using it to put out YouTube shorts and Instagram stuff short things rather than long instructional videos, for example.
JE: Or an interview like this.
SN: Yeah, I think maybe if I could get the avatar to show up to stuff for me. Would that work? And I could be outside in the sun and it could be talking for me. [Laughs.] Yeah, I haven’t really tried them yet, but it’s on my radar and it is something that I really do want to check out, because from what I’ve seen, they are quite convincing. So, yeah, it’s heading that way, isn’t it? What can we do? It’s heading in that direction, I think, for sure. So, we have to embrace the avatars.
JE: Okay.
SN: Cool.
JE: I guess we can end there.
