You Ask, I Answer: Creating Content for Search Engines?

In today's episode, Stephanie asks, should I be creating content for search engines or for people? Kind of the same thing. In the old days of SEO, we would focus very heavily on things like uh individual words or keyword phrases that we needed to create content around that we needed to create uh you know text for that took advantage of the way that search engines used to work in terms of how they identified what a page was about, how they identified what uh what were the most important words on a page. With the advent of the more sophisticated language models in search engines, that those tactics have largely gone by the wayside. Not because we've stopped using those words or phrases, but because of the way the engines themselves can understand language.

A few years ago, uh 2015, uh, we started to see uh the advent of these new uh models called transformers, not anything related to the um the toy line, that can process text in very different ways. And probably uh one of the more important and impactful frameworks uh that came out of this was uh one called uh bidirectional encoding representations for transformers, which we now know as uh Google's BERT uh model. And what these do differently than uh the way uh search engines used to process text in the old days is they understand at the sentence and paragraph level, uh, almost at the document level, as opposed to the word or the phrase level. A lot of the techniques that SEO folks um who haven't gotten up to speed um still focus on are very much still at the word level, and the word level is not as important anymore. It is about sentences, paragraphs, and uh context.

With these new models, they're able to read a sentence and predict what the next sentence logically should be. Uh, they're able to predict what a sequence of sentences are and then generate a uh an understanding of what the next uh uh paragraph logically could be about. Obviously, this technology has a great deal of uh power when it comes to things like question answering. So uh if you were to feed a uh uh a question to uh one of these models, uh they can come up with an what is a likely to be a solid predicted answer. Now, what does this mean?

This means that uh these models that are trained on an awful lot of text and tuned uh using things like Google's search quality rating guidelines, uh, can look at a page and say, how closely does this page, this page's content match uh what is predicted, like what should logically be here. Uh if this is a page that starts off with a sentence about coffee shops and coffee, it should be looking for a discussion of, you know, these are the hours that were open, these are the drinks that we serve, etc. And the further your content deviates from what the model understands, the less likely it is that the model's going to think that your page is relevant, right? So when we talk about creating content for search engines of people, that it really is the same thing because you would not take a page that you were putting on uh a website, say a coffee shop, right, and start putting like cooked ham recipes on there, right? Um you would not take a paragraph that was about when your coffee shop was open and insert you know uh random facts about you on there.

You shouldn't. Because as a human, you would read that and go, that makes absolutely no sense. What who put this page together? This is terrible. Um understanding how these natural language processing models work helps us understand how we should be writing content.

We should be creating content for people. What are the logical things that you as a person would want to know? Um, what are the logical next things that would occur in a piece of text? Think about it. If you are writing a page about your coffee shop, you've got a page of frequently asked questions.

What are the logical questions that should be on there? What time are you open? How much do things cost? How do I drive how do I get there? Um, you know, how do you treat things for allergens?

And by building a page that would satisfy a human who wanted to know all the things that logically you'd want to know about a coffee shop, you're also satisfying um these modern search engines that have the ability to uh predict text um and guess essentially how how aligned is your page with the things that it is predicting should be on there. Now, is it perfect? No, and are there uh constant updates? Yes, but the way these algorithms work, there's not as much drift as long as they have that that uh solid training data set, which again comes from two things. It comes from Google Search Quality Rating Guidelines where quality readers go out and rate existing pages as to how uh how well they display things like expertise, authority, and trustworthiness.

How uh what kind of quality uh is a page? Is it a high quality page with a lots of uh original unique content? Um is it does it demonstrate expertise? Does it demonstrate um topical relevance? Those are all things that are in the quality rating guidelines, which I strongly encourage you to read.

Uh Google publishes them. The other thing that Google has is a a group called Pygmalion, which is I believe uh you know two or three dozen um linguistic experts, you know, PhDs in linguistics who are constantly helping understand the mechanics of language and how the search queries that are being processed are being are essentially are are or how correctly they're processing language and ways to deconstruct language because one of the challenges of content on the web is a whole bunch, but one, it's t a technically a hot mess. Um, but something that is authoritative, authority varies wildly between um different industries and disciplines. You know, what's should be authoritative text, say in surgery, is going to be very different than authoritative text in plumbing, right? You're not going to expect a plumber to write at a graduate school level on their website, but they may have great content that makes them the popular choice for answers about to plumbing questions.

So a lot of the these models on the back end have to contend with these different topics. So what do you do with this? Well number one, you create content for people. And the best way to create content for people that's relevant is to actually ask people what questions do you have. Right.

Um go into your customer service inbox and just fish out every single question and uh for any question that occurs more than like two or three times guess what you should have content on the website about that because it's clearly a question that people have that they want to know more about um go into your uh your meeting notes with all your customers and all your clients and pull out the questions that they keep asking that people keep asking over and over again. Again, that's content that should be on your website. And when you create stuff that is aligned with people and reflects reality um you're going to do better in search engines because that's what Google is in is bringing in from all these other websites that are out there and these models will come to essentially a consensus like this is authoritative based on the training guidelines based on the the expert models and based on the the user behavior signals that Google can see when people search on search engines. If you know if this site reads like it should be authoritative but everybody who clicks to it clicks back to Google in two seconds like yeah something's wrong there. Um as opposed to you know people click to a website and they stay there for 30 minutes.

Like okay they they must have found what they were looking for. And those data points go in as well. So that's how to create content in the modern era. Focus on what people want, uh, what they need, what questions they have, anticipate the questions they're going to have in advance and put that content up there. Um, do a show like this.

I mean, uh, this this whole show that I've been doing for three years now is all about answering questions that real people have because it's how search engines function. If you got follow up questions, leave them in the comments box below. Subscribe to the YouTube channel and the newsletter. I'll talk to you soon. Take care.

Want help solving your company's data analytics and digital marketing problems? Visit trustinsights.ai today and let us know how we can help you.