When it comes to SEO content, the conversation is rising: can’t we just use AI? Can ChatGPT help me rank in Google? What about Jasper? Or one of the others?
Maybe. But it’s imperative to have a full understanding how these SEO content tools work and what their limitations are. Relying on AI for search engine rankings could be really risky business. While a lot of unknowns remain, here’s what we do know.
The “helpful content update,” the quality scores, the AI checkers — more and more is being added to Google that seem to indicate its ongoing preference for unique, original content.
(Read More content by people, for people in Search for an overview from the source itself.)
To date, any content generator fueled by AI is generative only in a limited sense. This means AI does not create original content in the true sense of the word.
What it “generates” is not creative but iterative.
It’s taking and aggregating all available data on the internet (admittedly, a vast set of information), then replicating it in various forms to align to a brief. It’s pretty sophisticated but it’s not perfect. It will be flawed but, more importantly, it will not be totally unique.
You’ll find this quickly if you try to give ChatGPT or another system the same command over and over. It will “generate” the same output. Google ranks based on uniqueness and downvotes sites for plagiarism.
(You’ve probably heard about this in education this year - check out Originality reports in Google for Education.)
Currently, AI content for SEO is kind of considered cheating. And Google hates cheating (or gaming the system). If your entire website or blog is written by AI, you’ll probably face some penalties in terms of search. Obviously, this is the opposite effect of what you were going for in the first place.
Note — Google’s position on this appears to be evolving. In spring of last year, their terms explicitly stated that they were against “automatically generated content.” As of the middle of last year, the guidelines say that people must avoid “automatically generated content intended to manipulate search rankings.” (To read more, see Google Search Essentials here.)
So, for a business, this is problematic. It means someone could literally copy and paste and recreate your website and you couldn’t do anything about it. You don’t own the content. You can’t defend it. It’s not your intellectual property because it didn’t come from a human’s intellect.
(In case you haven’t seen the headlines, you can read The current legal cases against generative AI are just the beginning for some of them.)
The GPT-3 from OpenAI is an example of a Large Language Model, or LLM.
Where they are now requires extensive pre-training, but this is shifting.
In the future, the idea is that these models will be capable not just of functioning after massive inputs, but in two futuristic ways:
Few-shot - Minimal training
Zero-shot - No training or input
So, eventually, these models will need fewer sources or even no sources to create good content. Eventually.
(There are tons of good sources on this, but I actually really like this random blog I found: Beyond Words: Large Language Models Expand AI’s Horizon - keep scrolling and there are a few really good pieces on LLM.)
If you want to understand how content AI tools are built and how they function, you have to understand two tech components:
We all think we know a lot about algorithms and, in essence, they’re pretty simple. Algorithms are just computation mechanisms that follow a set of rules.
The big benefit of an algorithm is it can be trained to handle more and more complex data over time.
The big shortcoming of an algorithm is it can’t really break the rules, at least not in an absolute sense. And good writing is all about breaking rules.
Here is some real feedback I provided when I was training a content algorithm for an AI tool:
There is disparity in subjective hyphenation - for instance, “time-management” versus “time management” - “critical-thinking” versus “critical thinking” - should be set to some standard (doesn’t matter which - either is technically correct).
There is often throwaway text on the last sentence of the last paragraph - unsure why that pattern would happen, but it does appear to be a pattern (not universally consistent, but frequent).
In the “skills” sections, it’s almost invariably repeating the word “skills” - for instance “Skills for Math Skills” - I’m shortening to remove the extra text but it appears to be an erroneous pattern.
I would say there isn’t an identifiable pattern (yet) to this, but there is a lot of passive language on some of these - which some SEO platforms would consider a problem. I’m swapping out a lot, but some is integral to the entire block of text, because of how it’s structured. If there is any way to make a rule about this, it would be a good idea to always favor active voice.
Natural language processing (NLP) is about linguistics. It’s the interplay between man and machine when it comes to language. The ability of a computer to process human language is, obviously, fundamental to any content artificial intelligence or similar tools.
If you want to explore more, this is a good article.
NLP is fundamental to language generation, which is what content tools like Jasper AI or similar seek to do.
NLP is something most SEO experts are very familiar with, because it’s inherent in on-page SEO best practices. We build topic clusters based on linguistic relation, or factors of linguistic proximity. In plain terms, “if someone used these words, these words also make sense.”
It’s a way of organizing and better understanding the patterns of human communication, and a system that has served as the backbone for search engines as we know them.
Those are two tech components, but a quick word about the user interface, because you’ll hear it tossed around in these convos: a natural language interface is the way the system and human communicate using natural speech.
You probably talk to one of these every day. Yep: Google Assistant, Alexa, and Siri are all examples of natural language interfaces. You can directly connect to and command an operating system without knowing code. Just using speech. It’s been a fantastic technology.
But of course, we all know it also has flaws, many times because we simply talk funny, ambiguously, or unpredictably.
Let’s talk about good writing. I mean, really good, feel it in your guts, can’t put the book down good writing.
What do Faulkner, Toni Morrison, and Isaac Asimov have in common? Nothing. Truly nothing. You could not mathematically associate any of their works in terms of psycholinguistics, style, formatting, or any other measurement. They are distinct. And they are all wonderful in their own way.
That personality and style factor is something that AI can only imitate based on historical precedent and existing fiction, nonfiction, or reference materials. Because, arguably, the great authors of this generation are rising as we speak. Their ideas are not published on the internet, but still circling in their own minds, getting richer and more relevant and better formed.
Once delivered, they may be fodder for the bots. But until then, they are authentically creative and developed in a way that cannot be fully explained by science.
AI can only tap the past and imitate what has been great before. It can only pull from what currently exists online. It cannot create the next era of greatness in human thought.
Fantastic writers have this Gestalt-like sensibility. They create work that is somehow greater than the sum of its parts.
All AI can do is assemble parts and there is no thought behind the ultimate mosaic or big picture. It is informative and useful in telling you which dot goes where, but it cannot achieve anything close to pointillism or a fully shaped work of art.
Writers understand break and flow and they understand this because they listen and read. Even though voice AI is getting better at the rhythms of speech and content AI is getting better at the flow of written word, it cannot come close to deciphering the underlying complexities of how we communicate.
The profound pause. The wilful misspeak. The meaningful manipulation of phrasing.
Another weakness of content AI has to do with all of the little instincts when we cherry pick our words to deliver to another person.
An example: one time, I was in a personal development workshop (I was like 10 — another story for another day). The man was talking about how he had a friend whose family owned a “chalet.” They loved it. Cherished it. Were proud of it. He was eager for an invite and once approached the owner at a party and said, “we’d love to come to your cabin sometime.” The man visibly recoiled at the use of this word, as though it was an insult or a pejorative. Needless to say, the guy didn’t score an invite.
Point being — the words we use are so deeply personal and so incredibly well-informed by our life experiences, family and region of origin, and wildly varied inputs, it’s impossible to know person to person whether a word or a phrase is right or wrong.
AI has to be trained not to replicate hate speech. It’s a far cry from being able to use micro-culture lingo or familial slang.
People with excellent interpersonal and communication skills pick up on this kind of stuff very well. Great writers are sharp sharp sharp and so good at this. We communicate differently to different people. And, compared to AI, we have a superior ability to know not just how to do it, but when.
Could SEO content be predominantly created by AI content systems in the future? Possibly.
Will AI be writing any lasting, enduring tomes of human thoughts? I’d like to think not.
The truth is, any content system is two things:
To become proficient in an SEO content AI tool is a useful skill for a writer. But it’s not the core skill of writing. That remains a craft that is employed by craftspeople who have put in the time and understand the incredible value of stewarding human ideas, detailing how and what we think in this given moment, and dreaming of other worlds, other futures, other stories.
I’ll leave you with this, and I may have mentioned it before — the AI content algorithm I trained had something called hallucinates in it. It would be little snippets of content or phrases that we could not trace back to an identifiable origin point. A little creepy. But, ultimately, still an echo of what we had fed it.
AI may hallucinate.
But only humans can dream.
Watch this content and our discussion on my YouTube channel:
Take a walk through time and around the globe to learn how storytelling shapes humanity.
Humor is an underutilized writing tool.
The TikTok heating feature isn't an urban myth. Here's what you need to know.