Hello and welcome to SEO is not that hard. I'm your host, ed Dawson, the founder of keywords. People use dot com, the place to find and organize the questions people ask online. I'm an SEO developer, affiliate marketer and entrepreneur. I've been building and monetizing websites for over 20 years and I bought and sold a few along the way. I'm here to share with you the SEO knowledge, hints and tips I've built up over the years. Hello and welcome to SEO is not that hard. It's me here, ed Dawson, your host, as usual, and today I'm going to be talking about AI agents. And are these the next wave of AI and automation that we've got coming through?

So you know, it's not that ago only a year or more ago maybe, that ChatGPT sort of burst onto the scene and really captured the imagination, particularly of the SEO industry, because all of a sudden, you could generate lots of content very quickly just by prompting a large language model such as ChatGPT to produce an article based on some, yeah, a short prompt, and saw a huge proliferation of, yeah, ai spam content and lots of spam sites that sort of soon leapt up that were sort of generating automatic content based on people, scraping people's asks and answering questions, and obviously then Google tried to combat that and come back and hit back at it, um, but it's clear that obviously the ai is not going anywhere. It's something here to stay. But these first llms had a few issues. While they looked really amazing in the first instance that wow, it can create this content dead quick and it's sort of plausible you know, plausible at first glance content there are issues with it, okay. So let's sort of think about what these issues are so compared to what an agent might be, so an LLM.

Basically, you have to give it a prompt and it will give you a reply and then that's it. So it's kind of like a one-shot deal. It's just one prompt, one reply and then you've got to prompt it again, get a reply, and this is what you might call a single-turn interaction, so it doesn't do anything multi-step, it will only do one step at a time. So you say, prompt it with a question, it will give you an answer. Then you might prompt it again with another question based on that answer and you know it might start to build up a conversation. But you might have noticed that when this happens you know the conversation it can kind of forget things you were talking about. It doesn't necessarily have a great memory, in some cases no memory at all, so it's essentially what's called stateless. So the model doesn't retain any knowledge of previous interactions unless you specifically prompt it in a way to keep reminding.

And I remember using this back, using LLMs, back maybe a year or so ago doing some coding, because as well as obviously producing English or any kind of natural language content, there are also very good LLMs at producing code and I was using it to help speed up a project. I was doing some clustering work and I was experimenting it to help speed up a project I was working on. That was when we were doing some clustering work and I was experimenting with clustering and I was just using ChatGPT to help accelerate some of the coding process. What I found was at times it would sort of trip over itself and forget where it was and where you were in the interaction and what you were actually working on, and that can be frustrating, obviously, because you're working along and then all of a sudden it gives some new code that can break everything, which then means you've got to go and reverse that that situation and start back again. So that fact that they're um don't really retain a project. Knowledge is an issue, okay, so you're still whether you're using it as an accelerator. You're not really using it to build a sort of a product or a project and have it remember where you are and it develop a knowledge around that project and what you're doing. The other thing, obviously, that the lms had issue with was, yeah, they couldn't sort of deal with external tools or integrations, so they couldn't, for example, talk to apis. It's only recently I've seen chat gpt able to actually really go and search the web to ask you questions. It would you know it wouldn't go and find further information, basically. So you were tied into the, the level of knowledge it had when its language model was built. So these were the issues that you're facing. When you see the LLMs, you know, maybe a year or so Now, what I'm now seeing that we're moving towards is agents, AI agents.

So the difference Between an agent and, say, just a large language model is, while the agent may well be based on a large language model so the ones I've been using you still need, say, a ChatGPT account or a Claude account that hooks into it what an agent will do is address some of those problems we've just talked about. So it will have a memory, it will be able to perform multiple tasks, it will be able to integrate with third-party tools and information, it will do multi-step problem solving and it will work with much more autonomy than just the large language model would. And there's a few interesting tools that have come out and I've been experimenting with recently. One that's getting a lot of noise is Cursor Cursor AI come out and I've been experimenting with recently. One that's getting a lot of noise is cursor cursor ai, and that is a development environment based on visual studio code, which lots of developers use but which has been forked from the visual studio code code base and has had an ai agent basically added on the side. So it allows you to basically start to build code talking to this AI agent that's hooked into the development environment and you can start by telling it what you want to achieve. So, rather than just starting with very basic code, you can just say you can explain the project you're trying to build and it will tell you what files to create, what code to put in those files, and it will then, once those files are created, it will start to tell you what code to put in those files and it will then, once those files are created, it can start to tell you what changes to make and it'll even do those changes for you if you ask it to. And then you can sort of step-by-step build a project by talking to the code base, essentially via the agent, and the agent itself will then make changes to the code base for you. So you don't actually have to code and make direct changes yourself and you can if you want, but you don't have to and it will do a lot of the work for you.

And I did experiment with this a couple of weeks ago. I just had it connected to the google search console, api and pull down data and it explains that what api wanted it to talk to. Um. It even explained to me how to get the api, api keys and stuff for that. I sorted those details out for it. It then could talk to api. It built a web interface, um, and yeah, it did some quite fancy things really quickly within a couple of hours. I built a reasonably complex app now it wasn't production ready at that point, built a reasonably complex app now it wasn't production ready at that point and I asked, pushed it a bit far and asked it to do a few things. It got confused and it broke the whole system. But as a starting point, where I basically just gave it an english natural language description of what I wanted to achieve, I connected the google search console, api, download results and sort them in various ways. It was actually quite impressive and the thing that impressed me the most was when, you know, the first code it came up with was really slow because there's a huge amount of data. I just said it's really slow. Is there any way we can speed this up? And it put in a whole load of threading and you know it was speeded up the whole process by a good factor of, I don't know, 10, 20, 30, maybe even 40 times. You know it really went. A process that was taking minutes to download all of a sudden was taking seconds to download. It was really, really impressive. So that's curse ray I.

Another one that's launched just a week or so ago is called replit. Um. Now if you google replit, uh, r-e-p-l-i-t. Replit works on a similar basis. It's uh gives you a, a development environment, which in their case, is through a browser with an agent on the side, and with this one I was playing about with it and I said to it I want to create a plumbing website, a website for you know, a rank and rent. I said a rank and rent plumbers is going to be given their services and I just put in a little bit of text that I got from chat GPT, in terms of what the problems are then we need to be solved. I'd also got some keywords, some data from keywords. People use the questions people ask.

So to sort of give it sort of an idea of the kind of content I wanted to put on there and said let's create a static website for it. And this agent, I think, than Cursor's agent, because it actually created all the files for me. It sorted everything out, it did all the integrations, all the dependencies and all I had to do was just talk to it and explain, as I went along, what changes I wanted to make to the web page sections to add and other functionality like that, and it did a really good job. I mean, occasionally at times it would add new content in and remove existing content that you didn't want to move in. But I just had to say to it this section is now disappeared or you're missing. This bits out of sections and it will go back and go sorry and it would work it out and fix it. Um, but again, it's still not perfect. It was didn't do it exactly as they want to do it. In the end it turned out. It used a lot of Python in there and I didn't want Python on to do a static page, but as a proof of concept at the moment in terms of how good this is likely to get, I was really impressed because if we go back to what ChatGPT was like a year or so ago and compared to what it is now, it's much more sophisticated.

These agents and these tools with agents are only going to get better and better and better. So it's not strictly SEO but obviously as part of seo you have to create sites and anything that helps you create sites and helps create content and speeds up your process is going to be good and these agents are going to become more and more prevalent. You know I can see um agents becoming more widespread at the moment. These ones I've seen so far are very, very concentrated around coding, but I think we're going to see more and more agents that come about that are going to help people automate and have agents run processes for them in their business that currently they're not able to do.

So I'm thinking about things, say, for example, you could have an agent that was specifically there to, a, say, carry out seo audits, make recommendations, make changes potentially an agent an ai agent that could be there to help you create a content plan and you can talk to it. But it isn't just hallucinating facts from a large language model. Like I've said in previous podcasts, don don't just use ChatGPT, for example, to create a keyword research or a content plan, because it's not actually basing anything. It does on real data. With an AI agent that is actually hooked into proper data sources like, say, the keywords people use, api, like data from, say, heyhrest, like data from Google Search Console, where you could give it the live data sources for the topics that you're interested in, for the subjects you're interested in, for your actual site, for your actual analytics, to give it actual context on what you're wanting to do An SEO or AI agent like that could be really incredibly powerful and do work for you.

That A you might have had to pay someone a lot of money to do before and they might not have done it very well, and also can give you much more sort of on time on specific, much more powerful, targeted interventions that you need to make, compared to doing it any other way. So I think that this is an exciting space and this these, this concept of agents is something to really watch out for, something I'm something to really watch out for and something I'm going to be looking out for and making much more use of it in the future. So hopefully that's given you something to think about. Do go out to Google AI Agents. Look at some of these things like Cursor, look at Replit and, yeah, watch what we try and do. Maybe we'll try to do something with keywords people use and how we can look at, you know, using agents in some way to make our processes much more user-friendly and much more powerful. But, who knows? Watch this space.

Anyway, until next time, I'll see you in the next episode. Before I go, I just wanted to let you know that if you'd like a personal demo of our tools at Keywords People Use that, you can book a free, no obligation, one-on-one video call with me where I show you how we can help you level up your content by finding and answering the questions your audience actually have. You can also ask me any SEO questions you have. You just need to go to keywordspeopleusecom slash demo where you can pick a time and date that suits you for us to catch up Once again. That's keywordspeopleusecom slash demo and you can also find that link in the show notes of today's episode. Hope to chat with you soon.

Thanks for being a listener. I really appreciate it. Please subscribe and share. It really helps. Seo is not that hard. It's brought to you by keywordspeopleusecom, the place to find and organize the questions people ask online. See why thousands of people use us every day. Try it today for free at keywordspeopleusecom To get an instant hit of more SEO tips. Then find the link to download a free copy of my 101 quick SEO tips in the show notes of today's episode. If you want to get in touch, have any questions, I'd love to hear from you. I'm at channel5 on Twitter. You can email me at podcast at keywordspeopleusecom. Bye for now and see you in the next episode of SEO is not that hard.