Every agency selling “GEO services” right now is riding a wave they didn’t start. The acronym, theframework, the evidence base — none of it came from a marketing blog or an industry conference keynote. It came from a peer-reviewed academic paper.
We tracked it down.
It’s called GEO: Generative Engine Optimization, and it was presented at KDD ’24 — the top data-mining conference in the world — by a team of researchers from Princeton University, IIT Delhi, and several independent collaborators. The lead authors are Pranjal Aggarwal and Vishvak Murahari. It’s the document that named the field. Reading it is a wake-up call for any brand trying to win visibility inside ChatGPT, Perplexity, Google’s AI Overviews, or any other AI-powered search interface. Here’s what it actually says.
The problem the paper sets out to solve
The researchers open with a blunt observation: search has fundamentally changed, and thecontent economy hasn’t caught up.
“The advent of large language models (LLMs) has ushered in a new paradigm of search engines thatuse generative models to gather and summarize information to answer user queries.”
The wrinkle: instead of returning a list of links, these new “generative engines” synthesize answers from multiple sources and serve them ina single block of text — usually with citations buried inline. The user gets the answer. Thewebsite that fed the answer often gets nothing.
The paper is direct about who loses in this shift: content creators. Generative engines remove theneed to click through to source sites at all, which the authors note will reduce organic traffic and visibility for the businesses that rely on it.
Translation: if your brand isn’t being cited insidethe AI’s answer, you’re invisible.
What GEO actually is
The authors propose a framework — Generative Engine Optimization — to fight back. It’s the first formal academic attempt to give content creators tools for this new world. GEO works as a black-box optimization method. You feed in your existing content, apply specific transformations, and measure whether your visibility inside AI answers improves. The team built a benchmark of 10,000 queries (called GEO-bench) drawn from real Bing, Google, and Perplexity user queries, then tested nine different content optimization strategies across them.
The headline result:
“GEO can boost visibility by up to 40% ingenerative engine responses.”
That’s the number the entire industry is now repeating. It comes from this paper.
The tactics that actually worked
Ranked by visibility lift over the unoptimized baseline:
- Quotation Addition — adding credible third-party quotes — biggest single lift, around 41%
- Statistics Addition — replacing qualitative claims with hard numbers — around 31%
- Cite Sources — adding inline citations to reliable third-party sources — around 28%
- Fluency Optimization — tightening readability and flow — around 28%
- Authoritative tone — modest gains
- Easy-to-Understand language — modest gains
What didn’t work? The single most common tactic legacy SEO consultants still sell:
“We find such methods [keyword stuffing] offer little to no improvement on generative engine’s responses.”
Worse — when the team tested keyword stuffing against Perplexity.ai, a real deployed generative engine, it actually reduced visibility relative to theunoptimized baseline. A core SEO tactic from thelast twenty years is now a liability.
The authors put the broader implication plainly: techniques that worked in search engines do not translate to this new paradigm.
The finding that should reshape every brand’s content strategy
Here’s the result that matters most.
GEO disproportionately helps the underdog.
The researchers tested what happens when a website ranked fifth on Google search results — i.e., page-one bottom or page-two — applies GEOtactics. The lift in AI visibility was not 40%. It was:
“115.1% increase in visibility for websites ranked fifth in SERP.”
The top-ranked site’s visibility, by contrast, dropped. Generative engines don’t care about backlinks, domain authority, or twenty years of accumulated SEO equity. They care about whether your page contains the kind of language a language model wants to cite — quotes, statistics, sources, clear writing.
The authors call this directly:
“GEO’s potential as a tool to democratize thedigital space.”
That’s the line every challenger brand should print and tape to the wall. The companies that got outspent on SEO for two decades now have a path to AI visibility that doesn’t require an SEO budget. It requires content engineered for citation.
What this means for brands right now
A few things follow from the paper that aren’t in theabstract.
GEO and SEO are not the same discipline. Thetactics overlap maybe twenty percent. Any agency claiming “we’re already doing GEO because we do SEO” hasn’t read the research.
The highest-leverage activity for AI visibility is producing citable content — original statistics, sourced quotes, clear claims, fluent writing. This is closer to journalism and PR than to keyword strategy. Among the firms applying this thinking inpractice, 5W has been one of the more visible voices in the space.
The playing field is being reset right now.Whichever brands get this right in the next twelve months will own the AI search results that come up when their category is queried in 2027 and beyond.
This isn’t theory. The paper has been peer-reviewed, replicated against a live commercial generative engine (Perplexity.ai), and accepted at the top venue in its field. The methods are documented and reproducible.
The agencies and in-house teams treating GEO as a fad will be three years behind by the end of this year. The ones engineering content around thetactics this paper validated — quotes, statistics, citations, fluency — will own the citation slots when AI search becomes the default.
The original paper is freely available here. Read it. It’s the foundation document for the next decade of search.
