# What is GEO (Generative Engine Optimization): definition, origin and what changes from SEO

Rodrigo Beckmann · Published 2026-10-27 · Category: [GEO](https://rodrigobeckmann.com/en/blog/category/geo/) · HTML version: https://rodrigobeckmann.com/en/blog/what-is-geo/

GEO is the work of getting a brand cited and recommended in AI answers. Where the term came from, what Google says about it and what actually changes from SEO.

## In short

- GEO (Generative Engine Optimization) is the work of getting a brand found, cited and recommended in answers from AI systems such as ChatGPT, Gemini, Perplexity and Google's AI answers.
- The term comes from a paper by researchers at Princeton and IIT Delhi, published in November 2023 and accepted at KDD 2024. In its tests, adding quotations raised a source's visibility by 41%, while keyword stuffing lowered it by 8%.
- For Google, optimizing for AI search is still SEO, according to the guide it published in May 2026. What changes is the target, the weight of third-party sources and the way you measure.

GEO (Generative Engine Optimization) is the work of getting a brand found, cited and recommended in AI answers, such as those from ChatGPT, Gemini, Perplexity and Google's AI Overviews and AI Mode. The term was coined in November 2023 by researchers at Princeton and IIT Delhi, in a [paper](https://arxiv.org/abs/2311.09735) accepted at KDD 2024, one of the main data mining conferences. Instead of competing for a spot in a list of links, GEO competes for space inside the text the AI writes.

The foundation is the same as SEO, and Google itself says so. But three things change: the target, which becomes the citation; where the brand gets decided, which moves from your own site to third-party sources; and the way you measure, which trades position for sampling. This guide covers where the term came from, what changes, what stays the same and what is off limits.

## Where the term GEO came from

The paper "GEO: Generative Engine Optimization" is by Pranjal Aggarwal (IIT Delhi), Vishvak Murahari, Karthik Narasimhan and Ameet Deshpande (Princeton), Tanmay Rajpurohit and Ashwin Kalyan. The starting point was simple: when a generative engine synthesizes an answer from several pages, the people who wrote those pages lose control over how they show up. The authors gave a name to the work of winning back some of that control, and a way to measure it.

To test it, they built GEO-bench: 10,000 queries drawn from nine public datasets across 25 domains. The engine was simulated. It fetched the top five Google results and asked gpt-3.5-turbo for an answer with citations. They then applied one of nine text changes to one of the sources and measured how much of the answer came from it, using a metric that counts words and weights them by position in the answer.

The results, against the unoptimized text, from Table 1 of the paper:

- **Quotations** from relevant sources: +41%, the best result.
- **Statistics** added to the text: +31%.
- **Fluency** and **citing sources**: about +28% each.
- **Authoritative tone**, more persuasive: +10%, well below what the authors expected.
- **Keyword stuffing**, the oldest SEO trick: −8%.

![Bar chart of visibility gain by method in the GEO paper: quotations +40.9%, statistics +30.6%, fluency +28.0%, citing sources +27.5%, technical terms +17.6%, easy to understand +14.0%, authoritative tone +10.4%, unique words +6.2% and keyword stuffing −8.3%](https://rodrigobeckmann.com/blog/what-is-geo-methods.en.svg)

The most interesting finding is about who gains. For a source ranked fifth in search, citing sources raised visibility by 115.1%, and quotations by 99.7%. For the source ranked first, the same method cut visibility by 30.3%. Space in the answer is shared: when one source gains, another loses.

The paper has real limits. It is a 2023 lab experiment, with gpt-3.5-turbo and an engine that read only five results. The authors also tested on Perplexity by uploading the texts as files: quotations gained 22%, and keyword stuffing lost 10%. The numbers point a direction; they are not a recipe. Engines in 2026 run several searches per question, read more sources and change behavior without notice.

## GEO, AEO, LLMO: is it all the same thing?

In practice, yes. AEO (answer engine optimization), LLMO (LLM optimization) and "AI SEO" are other names for the same work. Each vendor prefers one, and the choice says more about their marketing than about their method.

Google joined the debate on May 15, 2026, when it published a [guide to optimizing for generative AI features in Search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide), last updated on July 10. In its section on AEO and GEO, it defines both terms and concludes that, from Search's perspective, optimizing for generative AI search is optimizing for the search experience, "and thus still SEO." For anyone considering "AEO" or "GEO" services, the guide recommends the same filter you'd apply to any SEO advice.

Two caveats. First, the guide speaks only for Google Search. ChatGPT, Perplexity and the Gemini app run their own systems and have published nothing equivalent. Second, not everyone in the field agrees. Ahrefs, which sells SEO software, published [five differences between SEO and GEO](https://ahrefs.com/blog/seo-vs-geo) in September 2025, despite the similarities. The fight over the name matters less than the underlying question: what changes in the work.

## What changes from SEO

| | SEO | GEO |
|---|---|---|
| What you win | A ranking and a click on a link | A mention or citation inside the answer |
| Where the brand gets decided | On your pages and in the links pointing to them | Also in third-party press, reviews, forums and videos |
| What the user sees | A list of pages | Text synthesized from several sources |
| How you measure | Position on a results page | Frequency across many answers, which vary with every query |

### The target is the citation, not the click

In GEO, the brand shows up inside the answer. According to Ahrefs, the user may never visit the site, at least not directly from the AI answer. So what counts is being mentioned and, ideally, cited with a link. It's the shift I described in [why citations are the new backlinks](https://rodrigobeckmann.com/en/blog/why-citations-matter): backlinks moved the ranking, citations move the answer.

### Third-party sources outweigh your own site

In a study of 75,000 brands published in May 2025, [Ahrefs](https://ahrefs.com/blog/ai-overview-brand-correlation/) measured how several factors correlate with a brand's presence in Google AI Overviews. Branded web mentions had a correlation of 0.664. Backlink count, the classic SEO metric, came in at 0.218. Correlation isn't causation, but the gap is wide.

A University of Toronto study by [Chen et al.](https://arxiv.org/abs/2509.08919) (September 2025) reached a similar reading: AI search shows a "systematic and overwhelming bias" toward earned media, such as press and independent reviews, over brand-owned and social content. Google, the authors say, has a more balanced mix.

And those sources shift fast. In the [weekly Reddit in AI search index](https://rodrigobeckmann.com/en/research/indices/reddit-in-ai-search/), Reddit's presence in ChatGPT answers fell from 21.1% in the first week of 2026 to 0.5% in week 42, while in Google AI Overviews it rose from 12.3% to 28.6%. A source strategy that was right in January can be wrong by October.

### Measuring takes a sample, not a screenshot

Ask the same AI the same question twice and you may get different brands. That's why a single answer can't tell you whether a brand "shows up in ChatGPT." So how many questions does it take? First Answer tested that in a methodological study published in April 2026, [FA-MET-01, "How to Measure What an AI Recommends"](https://rodrigobeckmann.com/en/research/how-to-measure-what-an-ai-recommends): three experiments and 7,000 calls to the ChatGPT interface, designed to tell an isolated observation from a real pattern. Even the initial diagnosis, which only tells you whether the market is concentrated and who leads, takes 100 to 200 calls, with a margin of ±7 to 10 points. Measuring each brand's presence within ±3 points takes 1,000.

Where you measure changes the result too. The second study in the series, [FA-MET-02, "API or Interface: That Is the Question"](https://rodrigobeckmann.com/en/research/api-or-interface), asked the same question 1,000 times through the ChatGPT API, the programmatic access developers use, and 1,000 times through the interface consumers use, with the same search tool. Of the 15 most relevant brands, 12 differed significantly between the two environments. In SEO, a ranking was reasonably stable. In GEO, every measurement has to state how many answers it collected, in which environment and with what margin.

## What doesn't change

Google's guide is most useful for what it tells you to ignore:

- **llms.txt and "special" markup.** You don't need to create new machine-readable files, AI text files, markup or Markdown to appear in Search, including its AI features, because Search doesn't use them.
- **Schema as a lever.** Structured data isn't required, and there's no special schema.org markup for AI search. It still matters for rich results.
- **"Chunking."** Splitting content into small pieces for the AI is one of the tactics the guide says you can ignore.

What the guide asks for instead is "non-commodity" content: original, made by people who know the subject, adding something beyond common knowledge. A first-hand analysis is worth more than a summary of what's already everywhere.

Here the guide and the paper point the same way. Quotations, statistics and sources, the methods that worked best in the 2023 experiment, are exactly what separates a text that adds from a text that repeats. The SEO foundation stays: a crawlable site, useful pages and a reputation built off-site.

## What is off limits

On May 15, 2026, the same day as the guide, Google noted in its [documentation changelog](https://developers.google.com/search/updates) that its spam policies also apply to generative AI responses in Search. Its [spam policies page](https://developers.google.com/search/docs/essentials/spam-policies) now lists attempting to manipulate those responses as spam. The guide makes the same point and tells you to ignore pursuing inauthentic mentions.

In practice, buying mentions, planting content in forums and generating pages at scale to show up in answers sit under the same rule as bought links. China went further and turned a list of "black hat GEO" tactics into an [industry-written standard](https://rodrigobeckmann.com/en/blog/china-geo-standard).

## GEO in one sentence

GEO isn't a new trick or a replacement for SEO. It's the same old work of useful content and verifiable reputation, with a new target and a new yardstick: getting cited in an answer that changes with every query, and proving it with a sample, not a screenshot.

## Frequently asked questions

### What is GEO (Generative Engine Optimization)?

GEO is the work of getting a brand found, cited and recommended in answers from AI systems such as ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode. The term was coined in 2023 by researchers at Princeton and IIT Delhi, in a paper accepted at KDD 2024.

### Is GEO just SEO?

For Google, yes: the guide it published in May 2026 says optimizing for AI search is still SEO. Ahrefs points to differences: in GEO the goal is to be mentioned inside the answer, users often don't click, and most brand mentions come from third-party sites.

### What is the difference between GEO and AEO?

In practice, none. AEO (answer engine optimization) and GEO (generative engine optimization) are two names for the work of improving visibility in AI search experiences. Google's guide treats both terms together and says both are SEO.

### Do I need llms.txt or special schema to appear in Google's AI answers?

No. Google's guide says you don't need to create new machine-readable files, AI text files, markup or Markdown to appear in Search, including its AI features, because Search doesn't use them. Structured data isn't required either, though it still helps with rich results.

### Can Google penalize GEO?

Yes, when it turns into manipulation. Since May 15, 2026, Google has stated that its spam policies apply to generative AI responses in Search, and its spam policies page lists attempting to manipulate those responses as spam.

## Sources

1. [GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735) — Aggarwal et al., KDD 2024 (arXiv), 28 Jun 2024 · primary source
2. [Google's Guide to Optimizing for Generative AI Features on Google Search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) — Google Search Central, 10 Jul 2026
3. [Latest Google Search Documentation Updates (May 15, 2026)](https://developers.google.com/search/updates) — Google Search Central, 15 May 2026
4. [Spam policies for Google web search](https://developers.google.com/search/docs/essentials/spam-policies) — Google Search Central, 28 Aug 2026
5. [SEO vs. GEO: 5 Key Differences Despite the Similarities](https://ahrefs.com/blog/seo-vs-geo) — Ahrefs, 10 Sep 2025
6. [An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied)](https://ahrefs.com/blog/ai-overview-brand-correlation/) — Ahrefs, 26 May 2025
7. [Generative Engine Optimization: How to Dominate AI Search](https://arxiv.org/abs/2509.08919) — Chen et al. (arXiv), 10 Sep 2025

**Rodrigo Beckmann** — GEO specialist, co-founder and CTO of First Answer. [More about the author](https://rodrigobeckmann.com/en/about/)