Most of what gets called GEO advice online traces back, whether the author realizes it or not, to a single academic paper. Six researchers, Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, and Ameet Deshpande, working across Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi, published "GEO: Generative Engine Optimization" at the ACM SIGKDD Conference in Barcelona in August 2024. It's the paper that coined the term.
I think it's worth actually understanding what the paper did, rather than repeating the "up to 40% visibility lift" statistic that's been recycled across roughly a thousand blog posts since without anyone explaining where it came from. So that's where I want to start.
What Did the Original GEO Research Actually Test?
The researchers built something called GEO-bench, a benchmark of more than 10,000 real user queries spanning nine domains and seven broad categories. They simulated a two-stage generative engine pipeline that mirrors how real systems like Perplexity actually work. First, a traditional search engine retrieves the top five sources for a given query. Then, a language model, GPT-3.5-turbo in the original study, synthesizes those five sources into a single answer with citations.
Against that simulated pipeline, they tested nine distinct content optimization methods, measuring each one against a baseline of unoptimized content. The methods ranged from techniques borrowed directly from traditional SEO to entirely new approaches built specifically for how generative engines process and cite information.
Infographic: GEO-Infographic2.png Alt Text: Bar chart showing Generative Engine Optimization research results from the Princeton GEO-bench study with citation sources statistics, and quotation addition improving visibility while keyword stuffing underperforms
Which GEO Methods Actually Worked, According to the Research?
This is the part that gets flattened into a single stat in most secondary coverage, and it deserves more precision than that.
Keyword stuffing, the traditional SEO tactic of repeating target terms throughout a page, performed poorly. The researchers found it consistently underperformed compared to doing nothing at all, which is a genuinely useful finding, since it directly contradicts the instinct a lot of legacy SEO teams bring into their first GEO project.
Statistics Addition, injecting specific numbers, percentages, counts, and dollar figures into content, was one of the strongest performers across the board. Quotation Addition, including direct, attributable quotes from credible sources, performed similarly well, functioning as what one analysis of the paper called a citation magnet, since models trained to recognize attribution respond specifically to properly attributed quotes. Cite Sources, adding inline references to support your own claims, was the third standout method.
The specific numbers matter here, because they're more nuanced than the single headline figure suggests. The best-performing methods improved Position-Adjusted Word Count, a metric measuring how much of the generated answer draws from a given source, by up to 41%, and improved Subjective Impression, a quality rating of how well a source contributes to the answer, by up to 28%, relative to baseline. When the researchers tested the same methods against Perplexity.ai specifically, rather than their internal simulated pipeline, the improvements were 22% on Position-Adjusted Word Count and 37% on Subjective Impression. Different platform, different exact numbers, same directional finding.
One result stands out as genuinely striking. Sources that ranked fifth, the weakest position in a five-source pipeline, saw a 115.1% relative visibility lift when source citations were added to their content. That's the paper's strongest single finding, and it's rarely the number that gets quoted, probably because 40% is a rounder headline than 115%.
How Is GEO Different From AEO, and Does That Distinction Still Matter?
Here's an honest complication worth naming. The terminology hasn't fully settled, and different sources use GEO and AEO to mean slightly different things.
Some treat AEO as the older, narrower term, originally associated with voice search and featured snippets, that's since been largely absorbed into the broader GEO framework, since most voice queries now route through the same generative response mechanisms. Others treat GEO as the more academic, research-grounded term specifically referring to the Aggarwal et al. framework and its descendants, while AEO gets used more loosely in industry marketing content.
I don't think the terminology distinction is where anyone should spend much energy. What matters is the underlying discipline, structuring content so generative systems can retrieve, trust, and cite it, regardless of which three-letter acronym a given article happens to use. If you're building a program, build around the mechanics, not the label.
Does GEO Replace SEO, or Work Alongside It?
The research and the industry commentary since both point firmly toward the same answer: alongside, not instead of. Jeremy Moser, CEO of uSERP, is quoted making a version of this argument that I think is exactly right. Roughly 80% of GEO is good, fundamental SEO. Supporting that claim, current data shows 87% of ChatGPT citations match Bing's top ten organic results, and 93.67% of Google AI Overview citations link to at least one page already ranking in the top ten organically.
What that tells you practically: a page with zero organic visibility is very unlikely to earn AI citations either, regardless of how well it applies the specific GEO methods from the Princeton research. GEO methods are optimizing the margin; they help a page that's already competitive earn a disproportionate share of citations, not rescuing a page that traditional search has already buried.
What Does a GEO Program Actually Look Like in Practice?
Start with the content that's already ranking reasonably well in traditional search, since that's the content most likely to be in the retrieval pool a generative engine draws from in the first place. Layer the Princeton-validated methods onto it deliberately, rather than treating them as a checklist applied uniformly across every page.
Add specific statistics wherever a claim is currently vague. "Significant efficiency gains" becomes "a 34% reduction in manual review time, measured across 60 customer deployments." This isn't decoration. It's the single method the original research found most consistently effective.
Add attributable quotations from credible sources, whether that's a named customer, a recognized analyst, or a specific study, rather than an unattributed general claim. The research specifically found that properly attributed quotes function differently than the same information presented without attribution.
Cite your own sources for factual claims you make, even about your own product or industry. This might feel unusual for marketing content, which has historically avoided anything that looks like a citation, but the research is specific. This was one of the three strongest-performing methods tested, and the effect was strongest for content that started in a weaker position.
Avoid keyword density as a strategy entirely. The research is unambiguous that this specific legacy SEO tactic actively underperforms in a generative engine context, which is a meaningful signal for any team still running an old-style content playbook.
How Should GEO Priorities Differ Across Domains?
The original research found something practitioners tend to skip past: the effectiveness of specific GEO methods varied meaningfully by domain. What worked best for one category of query didn't transfer identically to another. This is a genuinely important finding for a B2B technology company to internalize, because it means a generic GEO checklist applied uniformly across fintech content, healthcare IT content, and industrial software content is leaving real performance on the table.
The practical implication: test your specific priority query categories directly rather than assuming a single GEO playbook applies evenly across your entire content library. A statistics-heavy approach that performs exceptionally well for a comparison page might not be the highest-leverage move for a definitional, top-of-funnel page in the same category.
This maps to something I've seen consistently across B2B technology verticals specifically. A fintech company's compliance-focused content responds strongly to cited sources and named regulatory frameworks, since that audience is trained to distrust unsupported claims by default. A developer tools company's technical documentation responds more strongly to specific benchmark statistics than to quotations, since developers evaluating a product tend to weigh a concrete number over a testimonial. Neither pattern was obvious in advance. Both only became clear after running the same content through actual query testing rather than assuming the Princeton paper's average findings would apply identically to a specific, narrow B2B audience.
What Should a B2B Company Actually Prioritize First?
Given everything above, I'd sequence it this way for most B2B technology companies building a GEO program from scratch. First, confirm the underlying SEO foundation is genuinely solid, since 87 to 93% citation overlap with top-ten organic rankings means this is a prerequisite, not an optional parallel track. Second, audit your highest-priority existing content for vague, unquantified claims and replace them with specific statistics, since this was the most consistently effective method in the original research. Third, add genuine attribution, quotes, and cited sources to content making factual claims, prioritizing pages that currently rank in a weaker position, since that's where the research found the largest relative gains. Fourth, resist the temptation to keyword-stuff in the name of "optimizing for AI," since the research specifically found this counterproductive.
None of this requires rebuilding a content program from zero. Most B2B technology companies already have a body of reasonably solid, organically ranking content sitting untouched. The fastest path to measurable GEO improvement is usually revisiting that existing library with the three validated methods above, rather than producing an entirely new wave of content built around GEO from the outset.
Frequently Asked Questions
What is the origin of the term Generative Engine Optimization?
The term was coined in the paper "GEO: Generative Engine Optimization" by Pranjal Aggarwal and five co-authors from Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi, presented at the ACM SIGKDD Conference in 2024. The paper introduced GEO-bench, a benchmark of over 10,000 queries, and was the first controlled study demonstrating that specific content changes could measurably increase visibility in AI-generated answers.
What is the actual visibility lift GEO methods produce, according to the research?
The commonly cited "up to 40%" figure refers to the best-performing methods' improvement on Position-Adjusted Word Count in the original study's simulated pipeline. Against Perplexity.ai specifically, the same methods produced a 22% improvement on that metric and a 37% improvement on Subjective Impression. The most striking single finding was a 115.1% relative visibility lift for weaker-ranked sources that added source citations.
Which specific content changes does the research say actually work?
Three methods consistently outperformed the rest: adding specific statistics to previously vague claims, including direct attributable quotations from credible sources, and citing sources to support factual claims. Traditional keyword stuffing, by contrast, underperformed compared to making no changes at all.
Does GEO work independently of traditional SEO?
No. Current data shows 87% of ChatGPT citations and 93.67% of Google AI Overview citations link to pages already ranking in the top ten organic results for the relevant query. GEO methods appear to optimize the margin for content that already has reasonable organic visibility, rather than functioning as a standalone path to citation for content with no existing search presence.
Should GEO tactics be applied the same way across every content category?
The original research specifically found that the effectiveness of different GEO methods varied by domain, meaning a single uniform approach applied across every topic and vertical is likely leaving performance on the table. Testing which specific methods perform best for your own priority query categories is more reliable than assuming a generic playbook applies evenly everywhere.
References
Pedowitz Group, What Is Answer Engine Optimization (AEO): The B2B Marketer's Practical Guide, 65% B2B buyer AI usage statistic and average AEO score benchmark: https://www.pedowitzgroup.com/blog/what-is-aeo-blog
Hunter and Bard, B2B Answer Engine Optimization (AEO): The Executive Guide, SEO versus AEO comparison framework: https://hunterandbard.com/resources/blog/seo-is-no-longer-enough-b2b-guide-to-aeo
PartnerStack, Answer Engine Optimization: The Ultimate 2026 Guide for B2B SaaS Teams, third-party platform influence on AEO citation: https://partnerstack.com/resources/guides/answer-engine-optimization-the-ultimate-2026-guide-for-b2b-saas-teams
SMA Marketing, Answer Engine Optimization: A Practical Guide, B2B SaaS comparison-shopping content framework: https://www.smamarketing.net/blog/answer-engine-optimization-guide
AEO Ranks, AEO for B2B SaaS: Complete Answer Engine Optimization Guide 2026, brand perception and citation accuracy distinction: https://aeoranks.com/aeo-for-b2b-saas/
