AI: The Next Big Thing That “Kills” SEO

Over the past year and a few months, ChatGPT broke onto the scene in such a big way and was immediately perceived to be an existential threat to Google in its search engine dominance. It is obvious why this is: OpenAI’s ChatGPT is one of the first, most reliable artificial intelligence systems that has highly adept natural language functionality and can answer questions very intellegently and produce work of very high quality that is equivalent and sometimes better to work produced by humans. It became rapidly obvious that if it can answer questions reliably, this could in some ways be superior to humans submitting keyword queries to a search engine to get a list of web content that may possibly provide what they’re seeking. ChatGPT produced one of the things that had long been a goal of Google’s itself.

Back in 2004, Google’s first employee and Director of Technology, Craig Silverstein, predicted that users would be able to talk to search engines, and that they would be able to talk back. He had also told John Battelle that “I would like to see the search engines become like the computers in Star Trek. You talk to them and they understand what you are saying.” Search engines in the early days dreamed that they could provide services like the artificially intelligent android in Star Trek, named Data, provided to characters in the show: ask a question and he would access all the collected knowledge of mankind and provide back a good answer. Twenty years later, the generative AIs are now doing this.

But, the question is: does AI also pose an existential threat the SEO professionals?

To some extent, such a question gives me a deep sense of deja vu. Way back in 2006, I wrote a blog post predicting that “SEO may be eclipsed by user-centered design,” and I’ve already felt vindicated about that prediction a few times over as Google progressively announced ranking factors over time that focused upon web design elements that impact user experience. People have long predicted the demise of SEO, because there is an argument that eventually the search engines could become so sophisticated that websites would not need particular help in order to achieve appropriate rankings in search result. Generative AI raises this specter once more, as it seems magical, and people are not fully considering what is going on off-stage in order for the magic to happen.

Generative AI is the emergence of a paradigm-shifting tipping point for a lot of industries and jobs. It can reliably produce mathmatical formulae, answer questions, and write lengths of articles on virtually any subject. Copywriters and article writers are now greatly at risk as articles that were previously produced in content marketing strategies can be more rapidly produced at lower costs by AI. Generative AI can also write programming code as directed and produce workable apps and programs.

Google itself has already long incorporated the basic elements of AI into its search algorithm. Various specific functions have long used neural nets and machine learning. In 2019 I wrote an article about how Google had likely begun using an AI ranking algorithm based on a classic machine learning method to incorporate human ratings of search results quality to generate many abstracted models for quality scores for types of webpages associated with specific search keywords. However, the quality scoring ranking algorithm incorporates all the existing ranking signals into generating these models — so, search engine optimization could still improve rankings.

In September of 2023, Google rolled out the Helpful Content Update (“HCU”), which resulted in a great many small or indie publisher websites losing rankings severely. No doubt some of the affected sites were using black hat SEO techniques (a.k.a. spam techniques that go against Google’s rules), but many were built with very basic SEO or were not particularly/intentionally optimized in a technical sense. Many of these sites relied upon Google being fairly forgiving of a lack of laser-focused SEO techniques, and they gained traction through solid content marketing — through the production of articles that were appealing and which incorporated good, topically-relevant keywords. The overall ranking change experienced by these websites seemed based upon significant changes in how Google weighted ranking factors. Very mixed responses from Google seemed to simultaneously indicate that Googlers were somewhat confounded by what had occured, while also indicating that the websites involved could not particularly re-optimize in order to recover from the effect.

In late 2023, Google also shifted to favoring major online forums where information and questions were crowd-sourced. Reddit was particularly seen to have benefited in search rankings with that change — quite likely intentionally as the company had signed a secretive deal with Google to provide the search engine with its content for the sake of training Google’s AI. There is some likelihood that Reddit also had demanded and received improved search visibility (i.e. rankings) as part of this deal, as Reddit was primping-up for a strong IPO.

However, even a year prior to the HCU and well before Reddit’s astonishing increase in search rankings across many keywords, users and SEO professionals had been noticing poorer search result quality. It seems likely that Google’s nascent use of AI in its search rankings was not working as well as should have been expected. The upset among small online publishers following HCU and increasing numbers of observers pointing out poor search results finally resulted in some movement at Google to attempt to improve its search quality once more. Some opined that the company had shoved aside search quality purists in favor of more aggressive monetization — the argument that “enshittification” had occured with Google. It also seems likely that the push to leverage AI in the ranking algorithms could have resulted in some degrees of unforseen consequences. Google search evangelists and webmaster central spokespersons did not seem to know what should be recommended in the wake of the HCU — an indicator that the abstracted ranking models that I theorized in my article about the machine learning method at Search Engine Land made it impossible for search engineers at Google to say precisely why some websites improved or declined compared to others.

The truth of it is that with the incorporation of machine learning methods, different types of webpages for different types of keywords could have particular ranking signals incorporated or not incorporated for them — so, the signals used most for a restaurant could differ considerably compared to the signals for a medical article about a disease and its treatment. Spread this over the many thousand or even millions of types of webpages and keywords and every particular query and relevant webpages could have different ranking signals and weightings applied to them.

The scenario I’m describing makes SEO much more complicated and harder to predict.

However, any AI ranking systems still have all the same ranking signals available to them. Optimizing for those many classic signals will still largely work, even without knowing exactly which ones will be most influential to Google.

Complicating the search scene, Google began rapidly rolling-out some generative AI contents at the tops of search results in an effort to remain relevant compared with ChatGPT. This creates a spector of potential “one-result” search results where the search engine might only respond back to a query with one single, ideal result that satisfies the users’ query. The mystery behind how Google’s AI may generate these responses further leads SEO industry personnel to fear that there is no way to optimize for the seemingly-intelligent system. But, the truth is that the AI may now need to sift through a million relevant webpages for any given content in order to deliver the best answer. And, the classic ranking systems are still what is used to find the best answers for many or most queries. Those classic ranking systems rely on all the classic signals available in order to feed information to the AI.

Thus far, other than providing a list of things to do for SEO, AI cannot particularly analyze a website for how well it has performed search optimizations. It cannot provide a list of specific optimizations that a particular website needs to do, and it cannot tell what strategies a website may not yet be doing in order to achieve better rankings.

As search is still needed to provide top sources for AIs to use in answering questions, SEO remains useful and necessary for connecting web content with the systems that will ultimate provide generative answers to people.

SEO is here to stay a while longer.