What is generative engine optimization (GEO)?
Generative engine optimization (GEO) is the work of getting your content used and cited inside answers written by AI systems, such as ChatGPT, Perplexity, Gemini and Google's AI Overviews. Traditional SEO aims for a ranked blue link; GEO aims for a mention, quote or citation in the generated answer itself.
For example, a software company that ranks fourth for "best payroll software for small business" might find that the AI Overview above the results names three competitors and cites a review site. GEO is the effort to make the company's own pages, or pages that mention it, part of the material that answer is built from.
The term comes from a 2023 academic paper, which defines GEO as "a flexible black-box optimization framework for optimizing web content visibility for proprietary and closed-source generative engines" [2]. In marketing use the meaning is broader: any tactic aimed at visibility in AI answers.
Who coined generative engine optimization?
Generative engine optimization was coined by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande in the paper "GEO: Generative Engine Optimization", first posted to arXiv on November 16, 2023 [1]. The authors were affiliated with Princeton University and IIT Delhi, among others, and the paper was accepted at KDD 2024 [1] [2].
That academic origin is what sets GEO apart from most marketing jargon. It arrived with a definition, a benchmark and published numbers, which is also why its headline figure is so often quoted out of context.
What did the original GEO paper find?
The original GEO paper found that rewriting a source page could raise its visibility in a generated answer by up to 40% in the authors' test setup, and up to 37% in a test on Perplexity [1] [2]. Three changes did best [2]:
| Method tested | What it means | Gain on the paper's main visibility measure |
|---|---|---|
| Quotation addition | Adding quotes from relevant sources | About 41% |
| Statistics addition | Adding quantitative data | About 39% |
| Cite sources | Adding citations to credible sources | About 27% |
| Keyword stuffing | Repeating query keywords | Slightly worse than doing nothing |
The authors also built GEO-bench, a set of 10,000 queries drawn from nine datasets, and found that what works varies by domain [1] [2].
Is the "GEO increases visibility by 40%" claim accurate?
The 40% figure is accurate for what it measured, but it is widely misquoted. It does not mean 40% more traffic, 40% more citations in ChatGPT, or 40% higher rankings. Three details in the paper matter [2]:
- The engine was a simulation. The main experiments used GPT-3.5-turbo writing answers from the top five Google results for each query, not a live commercial AI product.
- "Visibility" was a share of the answer. The main measure, Position-Adjusted Word Count, scores how many words in the answer come from a source and how early they appear. A second measure rated subjective impression.
- Only the source page changed. The test rewrote one of five already-retrieved pages and measured how much of the answer it captured. It did not test whether a page gets retrieved in the first place.
That last point is the gap most GEO advice skips. In live AI search, a page has to be found during grounding before any rewriting can help it, and for Google that retrieval runs through Search ranking systems [3]. The paper measured the second step, not the first.
What is the difference between GEO and AEO?
GEO and answer engine optimization (AEO) describe the same goal, being the source an AI answer uses, and differ mainly in origin and emphasis. GEO is the academic term from the 2023 paper and leans toward generated, multi-source answers. AEO is the industry term and leans toward being the direct answer to a question, a framing that also covers featured snippets and voice assistants.
| GEO | AEO | |
|---|---|---|
| Origin | Academic paper, November 2023 [1] | SEO industry usage |
| Focus | Share of a generated, synthesized answer | Being the answer to a question |
| Typical surfaces | ChatGPT, Perplexity, Gemini, AI Overviews | AI answers, featured snippets, voice assistants |
| Measured by | Visibility in generated answers [2] | Citations and answer appearances |
In practice the tactics overlap almost entirely, and many practitioners use the terms interchangeably. Search Insider uses AEO as its main term and treats GEO as a close synonym with a specific academic history. Google, for its part, does not treat either as separate from SEO [3].
Is GEO different from SEO?
Google says no. Its May 2026 guide states: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO" [3]. Search Engine Journal summarized the guide as calling AEO and GEO "still SEO" [4].
That is Google's position for Google's own AI features, and it rests on how they work: Google's AI Overviews and AI Mode retrieve pages through its normal ranking systems [3]. It is less clear for products that do not run on Google's index, and GEO practitioners argue that answer engines reward things rankings do not, such as being quoted on third-party sites or having original statistics.
Both views are partly right. Retrieval still depends on being findable and ranking, which is SEO. Which retrieved page gets quoted is a separate question. In Search Insider's study of 17,240 ranking pages, Google's ordinary results rewarded a year in the title, but the AI Overview showed no such preference once page age and ranking position were held equal. The same page can win the ranking and lose the citation.
How has GEO changed since 2023?
GEO moved from an academic proposal in late 2023 into everyday marketing language, and in 2026 Google publicly pushed back on GEO tactics it considers unnecessary [3] [6].
2023: the paper
The GEO paper appeared on arXiv on November 16, 2023, introducing the term, the GEO-bench benchmark and the visibility measures [1].
2024: KDD and a revised paper
Revised versions followed on May 28 and June 28, 2024, and the paper was accepted at KDD 2024 [1].
2026: Google weighs in
On a Search Off the Record episode published January 8, 2026, Google's Danny Sullivan warned against making content "bite-sized" for language models, saying "We don't want you to do that" [6]. Search Engine Journal's Roger Montti framed that as Google downplaying GEO, while arguing that AI search results were themselves surfacing weak sources such as an abandoned Medium blog and a LinkedIn article [5]. On May 15, 2026, Google added a guide on optimizing for generative AI features [7], which says there is no need for special machine-readable files, AI-specific markup or tiny content chunks, and that seeking inauthentic mentions "isn't as helpful as it might seem" [3].
What GEO tactics are supported by evidence?
The GEO tactics with published evidence behind them are adding statistics, adding quotations and citing credible sources, which performed best in the original paper's tests [2]. Everything else sold as GEO, from llms.txt files to "AI-friendly" formatting, rests on reasoning or vendor case studies rather than controlled, published tests.
A practical reading of the evidence:
- Be retrievable first. For Google's AI features, that means ordinary SEO: crawlable pages that rank for the questions and follow-up searches behind an answer [3].
- Add what only you have. Original figures, quotes from named experts and cited sources were the strongest methods in the paper [2], and they are also what makes a page worth quoting.
- Skip keyword repetition. Keyword stuffing did not help in the GEO tests [2].
- Don't fake mentions. Google says inauthentic mentions are less helpful than they seem and that its systems block spam [3].
- Test before you believe. Much GEO advice is correlational. Compare like with like (same query, similar ranking position, same page age) before crediting a tactic.
Should you invest in GEO?
Invest in GEO if AI answers already show up for the questions your customers ask, but treat it as an extension of SEO and content quality rather than a separate channel with separate tricks. The work that holds up is the work the original paper and Google's guidance agree on: be findable, be specific, and give AI systems something worth quoting [2] [3].