Somewhere in 2023, a group of researchers at Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi asked a simple question.
If AI tools are going to answer questions by pulling from web content instead of sending people to a results page, what actually makes a page get chosen?
They ran the experiment, published the paper, and gave the answer a name: generative engine optimization (Aggarwal et al., arXiv:2311.09735, later presented at KDD 2024).
That is the whole origin story, and it matters because it means GEO is not a term an agency invented to sell something new.
It is the name for a set of techniques that were actually tested against real generative engines, with real results attached to each one.
What the researchers actually found
They built a benchmark of ten thousand real queries, ran nine different content techniques against live generative engines, and measured which ones changed whether a page got cited.
Three techniques stood out clearly above the rest:
- Adding statistics. Swapping a vague claim for a specific number, percentage, or date
- Citing sources. Referencing a credible outside source to back up a claim already being made
- Adding quotations. Including one short, attributable quote from a real person
Here is the detail most people miss. Keyword stuffing, the oldest trick in classic SEO, actually made things worse in the same tests.
A model does not care how many times a phrase shows up on a page. It cares whether it can find one sentence worth repeating.
So how is this actually different from SEO
SEO and GEO are chasing two different prizes, which is easy to forget once both start showing up in the same content brief.
SEO wants a position on a results page, tracked by keyword and by domain. GEO wants a mention inside a generated answer, tracked by engine and by the actual question someone typed.
SEO still rewards backlink volume as one of its strongest signals. GEO cares far more about whether several credible sources describe the same fact the same way.
SEO tolerates a page that has not changed in years, as long as it still ranks. GEO consistently favors whatever was updated most recently, since many of these tools increasingly pull from live pages.
The gains are not spread evenly either, which is the part worth sitting with. Pages already ranked first on Google in the original study barely moved.
Pages sitting around fifth position saw their visibility inside generated answers jump 115% after the same rewrite. You do not need to dominate Google to win at GEO. You need one sentence a model can defend without hedging.
Where to actually start
If you want to try this on your own content, the order matters more than people expect.
- Pick five pages that rank somewhere between position four and ten, since that is where the research shows the biggest room to move
- Find the vaguest sentence in each one and rewrite it with a real number attached
- Add one credible source anywhere the page is already making a claim that needs backing
- Get one short quote from a founder, a client, or someone with real authority on the topic
- Check back in six weeks by running the page's core question through ChatGPT and Perplexity directly
None of this replaces the content calendar you already run. It is a second pass, aimed at a handful of pages, built for a kind of visibility a results page was never designed to measure in the first place.
Build your GEO framework with Ande Media
We apply these exact techniques, in the order the research actually validated, instead of treating GEO as a new label on an old SEO retainer.
- Find the vague claims already sitting in your content
- Add the specific numbers, sources, and quotes that make them citable
- Prioritize mid-ranked pages, where the research shows the biggest opportunity
- Track citation share separately from organic rank, month over month
- Report on which rewrites actually changed AI visibility
Build your GEO framework with Ande Media. Apply the same research-backed techniques to your own highest opportunity pages.
Frequently asked questions
Researchers at Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi, in a 2023 paper later presented at KDD 2024.
Adding specific statistics to a claim produced the strongest visibility gains in the original study, ahead of citing sources and adding quotes.
No. It runs alongside SEO and targets a different outcome, citation inside a generated answer rather than position on a results page.
Not necessarily. The first step is auditing existing content for vague claims and adding specificity, which any content team can start today.
