Written by Brook Schaaf

This past Friday, I attended what was billed as the first-ever conference on Generative Engine Optimization, aka AI SEO (the term has its detractor). Attendees seemed largely to be—no surprise—SEOs and SEMs. The most in-depth and expert speaker may have been Jason Clinton, the Chief Information Security Officer for Anthropic.

Clinton described how LLMs (the preferred term of the day) operate at a level that transcends the written word, reaching into a semantic, conceptual cloud ineffable to humans—then effortlessly descending into any language (why its translations are so good). This is admittedly a poetic, though serviceable, interpretation. The Platonic concept of the Form comes to mind, as though the LLM reaches up through the ether to touch it, grasp it, then bring it back down to earth, which also brings Prometheus to mind. And what do you know? My own LLM agrees (the flattery alone is worth $200 bucks a month!). 

Continuing in the vein of Greek mythology, it seems to me that we’re observing the formation of Hubris 2.0, Hubris 1.0 having been Google’s claim it couldn’t be gamed, disproven time and again.

Clinton noted that the engines are so fast, they’ll run through, say, all 300 results if there are 300. Keep in mind that the answer engines still buy search responses from the search engines. Early on, results were “very dumb,” especially because Reddit is “unfortunately” a signal. But now, they look for contrary evidence (e.g., bad as well as good reviews). Naturally, I asked how fake information can be filtered out. Clinton acknowledged this to be “a problem” and pointed to improved bullsh*t detectors, which depend on improved pattern recognition, which depend on…more training data. 

Every speaker who addressed the topic agreed that SEO is not dead and that organic and paid links from Google and its smaller competitors still account for the vast majority of leads. But answer engines are growing quickly as traffic sources, and the new modality of lengthy, unique queries is anticipated to overtake the blue links sooner or later. There was also consensus that ChatGPT is the mindshare winner in this category.

There was no strong prediction around monetization, save perhaps that it might anchor to the subscription model. This is conceivable, but perhaps not lucrative enough, given the high processing costs and large base of free users. This points back to affiliate-style opportunities, like OpenAI’s deal with Shopify. 

But there’s more. If the detector needs more training data and the data sources are ranked as platforms, sites, and individual users, then higher-ranked sources will be valuable for their mentions even if they don’t drive any traffic. This may thwart a commission, but terms for a paid placement can still be negotiated with input from the next generation of monitoring companies like Peec AI, Anvil AI, and Bear AI.

The problem of AI slop is going to continue to grow. One presenter showed off copy-generation prompts that were better than anything I’d seen before. According to a study published late last year, AI-generated content was estimated to have grown to over a third of content on Medium and Quora (and I don’t buy the low 3% number for Reddit). 

It’s not clear how this particular problem can be resolved except by looking for analogous signals to the ones we use right now, if not the very same signals. Thus, this new “GEOgraphy” sure seems like it will take the shape of the one we’re already familiar with.

The post The Shape of the New GEOgraphy appeared first on FMTC.