AI & SEO: How Much Should You Trust It?

In today’s digital age, search engine optimization (SEO) is crucial for businesses to establish an online presence and attract organic traffic. But with the rise of artificial intelligence (AI), how does it impact SEO and your content strategy? Is AI a friend or foe when optimizing websites for search engines? In this video, we will look at AI and SEO and the critical question, “How much should we trust it?” 

Transcript:

The AI Craze

As many of you are probably aware, AI has made leaps and bounds over the last few years, but the big question is how much we should trust this new technology. AI is actually not something new because even spell check is a form of AI. What is new are the new language models and the advancements in the area of generative AI. But AI has been part of SEO for quite a long time with BERT, RankBrain, or MUM.

Open AI is also not new, it’s been around for a while, and the GPT project had multiple iterations, and that’s where we get the fun and cool one that everybody likes to talk about today, known as ChatGPT.

ChatGPT and Unrealistic Assumptions

How can we really get our assumptions in order? There are a lot of unrealistic assumptions when it comes to ChatGPT and other generative AI tools. We have to remember, first and foremost, that these are simply tools. Many SEOs are getting lazy, and they’re just trusting AI when it spits out answers, and they’re just doing what the AI says.

Almost every single SEO tool now has some sort of an AI function with it, whether it be WordPress plugins,  tools that we use for our research, or specific tools built just for this new AI-advanced SEO. As I said, this is something that we’ve been doing for quite a while in the world of SEO, most predominantly in the area of semantic search, but now we’re starting to see AI-powered tools bridge out into other areas. In keyword research, content development, and similar areas, people put in prompts, click buttons, copy and paste, and think that they’ve done something that the search engines actually want.

Recently, Gary Illyes from Google Search put out a public service announcement about Large Language Models (LLMs) that touched on our unrealistic assumptions when it comes to AI.

He says that “LLMs have a very high wow factor, but they have no clue about your website; don’t use them for diagnosing potential issues with it. Also, remember that LLMs will hallucinate; pressed in the right way, they WILL give you information that’s completely detached from reality because the predictions on the word order make sense,”

These language models, artificial intelligence, are meant to give you a response to your input. The input is going to dictate the output, and they are trained on large language datasets and large amounts of information, but they’re not trained specifically on your business. They may not actually know what they’re talking about when it comes to your specific site, niche, and the people you’re trying to market to.

If you just go with the assumptions that are coming out of these language models, you can be way off and actually can hurt your SEO instead of improve your it. Remember, these are tools, so we need to understand the tools within the right context.

AI Examples

ChatGPT

Let’s get into some examples of this. First, we can start here with ChatGPT. I asked it what does SMA marketing do? SMA Marketing is my agency. If I go over to Google and I type in “SMA Marketing,” it’s an existing entity within the Google eco-space. I have Google My Business, websites, YouTube channels, and LinkedIn. I’ve got images showing up here, all labeled for SMA Marketing.

There are some others that have a similar name, but the reality is SMA is known as an entity. Google and search engines know that we exist, and we have existed since 2009. This isn’t a new entity that these language models would know nothing about. But let’s go back to OpenAI for an answer. It says that SMA Marketing is a fictional company. 

First of all, we know that that is not correct. ChatGPT says that it can’t provide specific information. But then it goes on to make some assumptions and say, “Well, if it did exist, maybe it would do these types of things.” While it’s trying to give me an answer about my specific business, it’s actually not giving me anything relevant because it doesn’t know who I am.

A lot of times, these tools will be used for diagnosing SEO issues, so you could go and ask it a question like, “Give me all the pages on SMAmarketing.net that talk about text message marketing.” We don’t do text message marketing, but let’s see what it comes up with. ChatGPT doesn’t even know that we exist. We’re not a fictional company. We have a domain that exists at SMAmarketing.net.

If I was trying to use ChatGPT in this instance, it actually wouldn’t help me much because it doesn’t know about my company. I understand that that’s not exactly what ChatGPT is about. Chat GPT is more of a generative language question-and-answer type of model.

You have to understand the usage of these types of tools as well. That’s another issue I see when people are talking about integrating these types of technologies within their tools, but they’re not used for these specific tools.

I’ll get even more specific and type “Give me the title tag for…” and I put in a specific URL. It’s not giving me the title tag, and this would be something pretty easy to do. It’s not the type of model that we’re looking at, but as you can see here, this is not going to be very helpful.

If I was trying to use ChapGPT to diagnose some issues on my site, it won’t even crawl my site or take a look at my site because that is not its use. ChatGPT is for chat, interaction, and engagement like that.

What people usually do is give the command, “Write me a blog on SEO,” and they hit the button. ChatGPT creates AI-generated content about SEO, they copy and paste it into their website, and they feel really good about themselves.

It’s a problem because it’s pulling information from somewhere, taking information from other people’s sites, and it’s basically making a new version of it. But it’s not citing anybody. There are no sources being cited here, and nobody is getting credit for this information. It’s just plug-and-play.

We could obviously go through this whole thing and pick it apart as well. If you just copy and paste this article on your site, you’re doing a pretty bad job because this is exactly what you’re not supposed to do with AI.

Now you could take this as a starting point, but you would not want to grab AI-generated content, copy and paste it and say, “Well, look at me. I just wrote a blog in two seconds.” As we can see, ChatGPT knows some things, like broad things here, but it doesn’t know everything.

In fact, it’s going to have a lot of difficulties when you ask it very specific questions, and even more difficult when you get into complex areas where it may not be an expert. You need to be cautious. Use this as a tool because that’s what it is, and then use the data from the tool to help you make better decisions but always check and verify.

I think a good line is from a man named Ronald Reagan, who was the president of the United States. He said, “Trust, but verify.” Use the technology but verify the information.

GPT3 Davinci

We’re going to take a look at GPT3 Davinci. I’m going to ask the same question, “What does SMA Marketing do?”

Now, in this case, it’s not getting it totally right. Again, it says, “SMA Marketing is a type of digital marketing that focuses on social media channels.” There is social media marketing, but that’s not what SMA Marketing does. SMA marketing is a business. Let’s see what happens if I separate these and run the code again to see if the input determines the output. I get the same answer, so it’s not correct because we’re not a social media marketing agency.

I can also take some of the questions that we asked over ChatGPT. This should be a pretty interesting result. This is a different type of language model, and it can be used in a lot of different ways. This is just giving me a list of titles, and it’s not actually giving me pages.

I’ll ask for page URLs and see if it does anything. I’ll tell you one thing – those articles do not exist on our website. It’s trying to give me an answer, but this is what Gary Illyes was talking about when he says AI can hallucinate.

It’s trying to give you an answer and is tasked with the job of providing you an answer – whether that answer is correct or not.

It gave me a list of URLs. This would be awesome if it worked, but none of these URLs actually exist on my website. We don’t talk about text message marketing on my website, and it’s telling me what I want to hear. This is an area where you have to be careful using AI to diagnose specific things on your website.

Benefits to AI

Target Keywords

Now where is AI pretty good? It’s helpful in many aspects of SEO. I like to use it for keyword extraction. Here is a prompt, and I’ve got some content from a client’s website. I can ask the tool to give me the relevant keywords from within this page. It’s going to give the main keywords on this page. There are quite a lot of them, but go ahead and check them out.

Title Tags and Meta Descriptions

Another benefit is AI-generated title tags. Type the command, “Create some title tags for a page around this specific topic,” and we can receive the response here. This is a good way to generate title tags, and if you’re just tired and you need some help, that’s a good way to do it.

These are great uses for AI. We can see that AI can make up stuff, but we can also see that it can be helpful if you use it within the right framework and you give it the right boundaries. We can use it for meta descriptions as well or even provide an outline.

Outlines for Content

Ask for an outline for content creation, and this is a great use for AI. You give it your prompt, you tell it what you’re looking for, and then you allow it to give you a baseline to start working with.

As I said, SEO tools are not magic wands. That’s what AI is. It’s a powerful tool, not magic. AI and large language models can be very helpful with things like keyword research and clustering, content optimization, content ideas, and research. But it can’t do everything. You need to inject your expertise and insight into your pieces of content in order to get the right results and produce high-quality content.

Content marketers can’t just blindly trust AI-based SEO tools. That goes for any tool, including keyword research tools and those for SERP analysis. Don’t just trust the SEO tools. Go look at the search results. We need to make sure that we’re using our own expertise and our own understanding of the industries and the websites we’re working with in order to get the best results.

I hope you guys learned something new from this video. If you have any questions, please contact us! I’d love to continue the conversation with you. We will see you again. Happy Marketing.