The game has changed, and quite recently, too.
Generative engine optimization (GEO), AI Overviews (AIOs), or just an extension of SEO (now being dubbed on LinkedIn as Search Everywhere Optimization) – which acronym is correct?
I’d argue it’s GEO, as you’ll see why. And if you’ve ever built your own large language model from scratch like I did in 2020, you’ll know why.
We’ve all seen various frightening (for some) data on how click-through rates have now dropped off the cliff with Google AIOs, how LLMs like ChatGPT are eroding Google’s share of search – basically “SEO is dead” – so I won’t repeat them here.
What I will cover are first principles to get your content (along with your company) recommended by AI and LLMs alike.
Everything I disclose here is based on real-world experiences of AI search successes achieved with clients.
Using an example I can talk about, I’ll go with Boundless as seen below.
Tell The World Something New
Imagine the dread a PR agency might feel if it signed up a new business client only to find they haven’t got anything newsworthy to promote to the media – a tough sell. Traditional SEO content is a bit like that.
We’ve all seen and done the rather tired ultimate content guide to [insert your target topic] playbooks, which attempt to turn your website into the Wikipedia (a key data source for ChatGPT, it seems) of whatever industry you happen to be in.
And let’s face it, it worked so well, it ruined the internet, according to The Verge.
The fundamental problem with that type of SEO content is that it has no information gain. When trillions of webpages all follow the same “best practice” playbook, they’re not telling the world anything genuinely new.
You only have to look at the Information Gain patent by Google to underscore the importance of content possessing value, i.e., your content must tell the world (via the internet) something new.
BoundlessHQ commissioned a survey on remote work, asking ‘Ideally, where would you like to work from if it were your choice?’
The results provided a set of data and this kind of content is high effort, unique, and value-adding enough to get cited in AI search results.
Of course, it shouldn’t take AI to produce this kind of content in the first place, as that would be good SEO content marketing in any case. AI has simply forced our hand (more on that later).
After all, if your content isn’t unique, why would journalists mention you? Bloggers link back to you? People share or bookmark your page? AI retrain its models using your content or cite your brand?
You get the idea.
For improved AI visibility, include your data sources and research methods with their limitations, as this level of transparency makes your content more verifiable to AI.
Also, updating your data more regularly than annually will indicate reliability to AI as a trusted information source for citation. What LLM doesn’t want more recent data?
SEO May Not Be Dead, But Keywords Definitely Are
Keywords don’t tell you who’s actually searching. They just tell you what terms trigger ads in Google.
Your content could be appealing to students, retirees, or anyone. That’s not targeting; that’s one size fits all. And in the AI age, one size definitely doesn’t fit all.
So, kiss goodbye to content guides written in one form of English, which win traffic across all English-speaking regions.
AI has created more jobs for marketers, so to win the same traffic as before, you’ll need to create the same content as before for those English-speaking regions.
Keyword tools also allegedly tell you the search volumes your keywords are getting (if you still want them, we don’t).
So, if you’re planning your content strategy on keyword research, stop. You’re optimizing for the wrong search engine.
What you can do instead is a robust market research based on the raw data sources used by LLMs (not the LLM outputs themselves). For example, Grok uses X (Twitter), ChatGPT has publishing partnerships, and so on.
The discussions are the real topics to place your content strategy around, and their volume is the real content demand.
AI Inputs, Not AI Outputs
I’m seeing some discussions (recommendations even) that creating data-driven or research-based content works for getting AI recommendations.
Given the dearth of true data-driven content that AI craves, enjoy it while it lasts, as that will only work in the short term.
AI has raised the content bar, meaning people are specific in their search patterns, such is their confidence in the technology.
Therefore, content marketers will rise to the challenge to produce more targeted, substantial content.
But, even if you are using LLMs in “deep” mode on a premium subscription to inject more substance and value into your content, that simply won’t make the AI’s quality cut.
Expecting such fanciful results is like asking AI to rehydrate itself using its sweat.
The results of AI are derivative, diluted, and hallucinatory by nature. The hallucinatory nature is one of the reasons why I don’t fear LLMs leading to artificial general intelligence (AGI), but that’s another conversation.
Because of the value degradation of the results, AI will not want to risk degrading its models on content founded on AI outputs for fear of becoming dumber.
To create content that AI prefers, you need to be using the same data sources that feed AI engines. It’s long been known that Google started its LLM project over a decade ago when it started training its models on Google Books and other literature.
While most of us won’t have the budget for an X.com data firehose, you can still find creative ways (like we have), such as taking out surveys with robust sample sizes.
Some meaningful press coverage, media mentions, and good backlinks will be significant enough to shift AI into seeing the value of your content, being judged good enough to retrain its models and update its worldview.
And by data-mining the same data sources, you can start structuring content as direct answers to questions.
You’ll also find your content is written to be more conversational to match the search patterns used by your target buyers when they prompt for solutions.
SEO Basics Still Matter
GEO and SEO are not the same. The reverse engineering of search engine results pages to direct content strategy and formulation was effective because rank position is a regression problem.
In AI, there is no rank; there are only winners and losers.
However, there are some heavy overlaps that won’t go away and are even more critical than ever.
Unlike SEO, where more word count was generally more, AI faces the additional constraints of rising energy costs and shortages of computer chips.
That means content needs to be even more efficient than it is for search engines for AI to break down and parse meaning before it can determine its value.
So, by all means:
- Code pages for faster loading and quicker processing.
- Deploy schema for adding context to the content.
- Build a conversational answer-first content architecture.
- Use HTML anchor jump links to different sections of your content.
- Open your content to LLM crawling and use llms.txt file.
- Provide programmatic content access, RSS feeds, or other.
These practices are more points of hygiene to help make your content more discoverable. They may not be a game changer for getting your organization cited by AI, but if you can crush GEO, you’ll crush SEO.
Human, Not AI-Written
AI engines don’t cite boring rehashes. They’re too busy doing that job for us and instead cite sources for their rehash instead.
Now, I have heard arguments say that if the quality of the content (let’s assume it even includes information gain) is on point, then AI shouldn’t care whether it was written by AI or a human.
I’d argue otherwise. Because the last thing any LLM creator wants is their LLM to be retrained on content generated by AI.
While it’s unlikely that generative outputs are tagged in any way, it’s pretty obvious to humans when content is AI-written, and it’s also pretty obvious statistically to AI engines, too.
LLMs will have certain tropes that are common to AI-generated writing, like “The future of … “.
LLMs won’t default to generating lived personal experiences or spontaneously generating subtle humour without heavy creative prompting.
So, don’t do it. Keep your content written by humans.
The Future Is A New Targeted Substantial Value
Getting your content and your company recommended by AI means it needs to tell the world something new.
Make sure it offers information gain based on substantive, non-LLM-derived research (enough to make it worthy of LLM model inclusion), nailing the SEO basics, and keeping it human-written.
The question now becomes, “What can you do to produce high-effort content good enough for AI without costing the earth?”
More Resources:
Featured Image: Collagery/Shutterstock
