What is generative engine optimization?
Generative engine optimization (GEO) is improving how often, and how prominently, your content appears in answers that a generative engine writes: ChatGPT, Perplexity, Google's AI Mode. The term comes from a 2023 paper by Aggarwal and others, which found that adding citations, quotations and statistics raised visibility by up to 40 percent while keyword stuffing did little or nothing.
Part of Glossary: the terms behind MCP, coding agents and shared pages, defined plainly.
Someone asks a generative engine a question. It runs a search, pulls several pages, and writes one answer from them, citing some.
How much of the answer draws on your page, and how early, is the visibility GEO raises. The paper measured position-adjusted word count and a judged impression.
Pages that cited their sources, quoted named people or added concrete statistics were drawn on 30 to 40 percent more than the baseline in the paper's benchmark.
Keyword stuffing, the old SEO move, scored below the baseline. The engine summarises meaning; it does not match strings.
Results varied by domain and by engine, so the paper's own advice is to measure on your queries rather than copy a checklist.
Source: GEO: Generative Engine Optimization (Aggarwal et al., arXiv 2311.09735)
GEO versus SEO and AEO
Classic search returns links and lets the reader choose. A generative engine reads the top results itself and writes one answer, so the reader may never see a list at all. That shifts what you are optimising. SEO wants the rank. Answer engine optimization wants your paragraph shown verbatim. GEO wants your page to be one of the sources the model leans on, and ideally the one it credits with a citation.
The paper that named the field frames this as a loss of control for publishers: a model decides which of several sources to blend, and in what proportion, and the publisher gets no say beyond the content itself. GEO is the attempt to influence that decision through the content.
What the paper found
Aggarwal and colleagues built GEO-bench, 10,000 queries split across domains, with the web sources a search would return for each. They applied nine rewrites to a source and measured how much more of the generated answer came from it. Three methods stood out: citing sources, adding quotations from named people, and adding statistics. Together they improved position-adjusted word count by 30 to 40 percent and the subjective impression score by 15 to 30 percent. Writing more fluently and in plainer language also helped. Keyword stuffing scored below the unmodified baseline.
Read the result for what it is. The engine was a stand-in built on gpt-3.5-turbo over Google's top five results, and the live check on Perplexity used 200 queries. The effects also varied by domain, which the authors flag themselves. The direction is clear: specific, attributed, verifiable content gets used; dense repetition of the query does not. The exact percentages are a 2023 snapshot.
Example: a glossary page written to be cited
The glossary pages on intraface.cc apply the paper's three winners without calling them that. Each term carries a named primary source with a link, in a "source" field that renders on the page and in the JSON-LD. Where a figure exists, such as a token lifetime or a limit, the page states the number rather than "a while". Where a vendor or a paper defines a term, the page attributes the definition to it. The glossary hub at https://intraface.cc/glossary/ lists every term inside a DefinedTermSet so that a model, or the search behind it, can see the whole set and which definitions are related. None of that guarantees a citation. It does make the page the kind of source the paper found engines draw on.
Common mistakes
- Chasing the engine with keywords. The one method the paper found useless is the one most SEO habits reach for first.
- Reading "up to 40 percent" as a promise. It is the best case on a benchmark, with a stand-in engine, in 2023.
- Citing nothing. A page with no sources gives a model nothing to verify and nothing to credit.
- Confusing GEO with geographic SEO. The search phrase "geo seo" returns both. Local ranking advice will not get you cited in a written answer.
- Blocking the search crawlers. A page that OAI-SearchBot or PerplexityBot cannot fetch cannot be cited, whatever its content.
FAQ
- Who coined generative engine optimization?
- Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, in a paper first posted to arXiv in November 2023 and later accepted at KDD 2024. It introduced both the term and GEO-bench, a benchmark of 10,000 queries for measuring visibility in generated answers.
- What does GEO SEO mean?
- Two unrelated things, which is why the phrase is confusing. In AI search it means generative engine optimization, getting cited in AI-written answers. In local marketing it is short for geographic SEO, ranking for searches tied to a place. Check which one a tool or article means before following its advice.
- Did the generative engine optimization paper test real engines?
- Mostly a stand-in. The main experiments used an engine the authors built on gpt-3.5-turbo that retrieved the top five Google results for each query. They also ran a 200-query check on Perplexity.ai, where adding quotations improved position-adjusted word count by 22 percent over the baseline.
Stop sending HTMLover Slack.
Create an account in a minute, connect your coding agent, and make your first board.