GEO (Generative Engine Optimization), the Future of SEO (?)
A new study has found on-page optimization techniques for AI generative engines to find and reference your content in their answers.
Six researchers from multiple universities (Princeton University, Georgia Tech, The Allen Institute for AI, and IIT Delhi) collaborated on a study exploring what AI search engines like to surface in response to users’ queries.
The main idea is that generative engines are going to replace search engines but there’s still an “optimization” strategy needed to help publishers get found.
Hence the name of that strategy is going to be Generative Engine Optimization (GEO).
Thanks for reading Ann Smarty's Search & #AI Digest! Subscribe for free to receive new posts and support my work.
To save you time, the basic ideas behind this research are:
Take the database of queries and URLs responding to those queries
Build your own AI generative engine (based on what BingChat’s output looks like). Correct me if I am wrong here. They also used Perplexity.ai to verify their findings.
Optimize content from those URLs using various tactics (more on these below)
See if their own engine cites them more after these optimization efforts
The queries varied based on the topic and intent (most were informational) but overall the winning optimization tactics were:
Cite sources from authoritative sources (especially for factual queries)
Quotation addition (e.g. from a known expert / entity)
Statistics addition (especially for queries in the law and government niches)
I must admit, I am pretty fascinated by the idea of human beings trying to decode what a machine thinks is good for them.
And frankly, this is a good start to understanding how to surface our and our clients’ content in AI-generated answers.
However, I do have a few problems with this research:
This is based on a self-made AI generative engine - so the study assumes they will work the same
Google’s SGE patent talks about how Google’s AI snapshots work and how little they have to do with the typed query. Google’s SGE algorithm includes related queries and also personalizes snapshots based on each user’s search history and interactions. Read more on this here: Google’ SGE (AI Snapshots) Patent and the Future of Search (?)
How about this bold statement from the research:
This emerging technology, which we formalize under the unified framework of generative engines (GEs), has the potential to generate accurate and personalized responses and is rapidly replacing traditional search engines like Google and Bing. Generative Engines typically satisfy queries by synthesizing information from multiple sources and summarizing them with the help of LLMs. While this shift significantly improves user utility…
What is it based on?
Is there data proving that AI generative technology can, in fact, provide accurate answers?
I saw data about AI hallucinations and providing generic answers (“mansplaining”) which are also filled with bias on many levels. But not much about the accuracy of their summaries and claims.
Is there data on these engines actually replacing traditional search? I mean beyond this: No, ChatGPT isn't stealing Google's search market share
It might be changing some searching behaviors (with time, these never change fast) but this is going so far as to say that Google’s answer boxes or featured snippets have prevented people from searching.
It is also six highly educated researchers making claims without basing them on anything which makes me wonder how much of a research that is.
Room for manipulations
Finally, if enough publishers figure out those easy on-page tactics to trick a generative engine into believing their content is better, I see a flood of manipulations leading to lower-quality results.
It is all about surfacing real value (yes, Google seems to be on the right path here trying / struggling to find actual expertise and first-hand experience).
Share YOUR thoughts
I encourage you to read the research and its SEJ coverage and let me know what you think. We are all figuring it out together!
This is the research itself: GEO: Generative Engine Optimization
Also, this graph from the paper was pretty good: A nice way to explain what we are going to deal with:
This conversation is definitely needed. Figuring out ways to be found by future search/generative engines is going to be our job. So studies like this will continue being under my radar (subscribe to be updated!)