Generative Engine Optimization (GEO)

Where the term comes from

The term was introduced in a paper titled "GEO: Generative Engine Optimization", submitted to arXiv on 16 November 2023 by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, and last revised on 28 June 2024.

This matters more than it usually would, because GEO is a term with a paper trail. Most marketing acronyms do not have one, and the paper says several things that later usage has quietly dropped.

The authors' framing is about a power imbalance rather than a marketing tactic. They describe content creators as "the third stakeholder", who, "given the black-box and fast-moving nature of generative engines", have "little to no control over when and how their content is displayed." GEO is presented as a response to that: a "flexible black-box optimization framework" for a system nobody outside it can see into.

What the paper actually measured

The headline result is the one that travels: the paper demonstrates that GEO "can boost visibility by up to 40% in generative engine responses."

Three things belong next to that number, and all three are in the same abstract.

"Up to" is a ceiling, not an average. It describes the best observed case, not what a given site should expect.

Effectiveness is domain-dependent. The authors state that "the efficacy of these strategies varies across domains, underscoring the need for domain-specific optimization methods." A method that worked in one field is not established for another.

It was measured on their own benchmark. The evaluation ran on GEO-bench, which the same paper introduces: "a large-scale benchmark of diverse user queries across multiple domains, along with relevant web sources to answer these queries." That is a reasonable way to make results reproducible, and it is not the same as a measurement in the live systems people actually use.

The paper dates from 2023 and 2024. The generative engines it studied have changed since.

What Google says about its own AI features

Google's documentation for site owners takes a different position on its own surfaces, and states it plainly:

"There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."

And, specifically on the file-and-markup approaches often sold under the GEO label:

"You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add."

This is not a contradiction of the paper, and reading it as one gets both wrong. The two statements have different scopes. The paper is about generative engines as a class, including assistants that are not Google. Google is describing what is required to appear in [AI Overviews](/glossary/ai-overview) and AI Mode specifically.

What follows is narrower and more useful than either "GEO works" or "GEO is snake oil": for Google's AI surfaces, the operator of those surfaces says the existing fundamentals are the mechanism. For assistants that are not Google, that statement does not apply, and no equivalent statement from those vendors is quoted here because none was found.

What that leaves worth doing

The fundamentals Google names are unglamorous and apply regardless: pages that can be crawled, content findable through internal links, important information available as text, structured data that matches what a reader sees.

Beyond that, the honest position is that visibility in AI answers is easier to measure than to optimise. Measuring is a matter of asking assistants what they say about a category and recording the answer. Optimising is a matter of claims, and the strongest published claim comes with "up to", a domain caveat and a benchmark from 2024.

Anyone selling a GEO service should be able to say which of those two they are doing.

Common mistakes

Quoting the 40 percent without its conditions. It is an upper bound from a specific benchmark, measured on systems that have since changed.

Treating GEO as a Google technique. Google says its AI features need no special optimisation. GEO as a term covers generative engines broadly, most of which are not Google.

Buying AI-specific files or markup for Google visibility. The documentation addresses this directly and says they are not needed. Whether they help elsewhere is a separate question with a separate answer.

Assuming what worked in one sector transfers. The paper's own finding is that effectiveness varies by domain.

Related terms

• [AI Overview](/glossary/ai-overview): the Google surface most GEO discussion is aimed at
• AI Mode: a separate Google surface with different models and different links
• llms.txt: one of the file-based approaches Google's documentation addresses
• Answer engine: the broader category the paper calls a generative engine