What Is Generative Engine Optimization (GEO) in SEO?

Generative engine optimization (GEO) is the practice of making your content and brand more likely to be retrieved, used and cited by generative engines: AI systems such as ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews that search for sources and then write one synthesized answer. In SEO terms, GEO is the part of search optimization aimed at being included in the AI answer rather than ranked in a list of links. The term comes from a 2023 research paper by Aggarwal et al., which found that specific content changes, like adding sources, quotations and statistics, raised a page's visibility in generated answers by up to 40%, while classic keyword stuffing did not help.

This guide explains what GEO means for an SEO team, what the original research did and didn't show, how generative engines choose sources, and how to put it into practice.

What does GEO mean in SEO?

In traditional SEO, a search engine returns a ranked list and the user picks a result. Visibility equals position.

A generative engine works differently. It takes the user's question, runs one or more searches, reads a handful of sources, and writes an answer that mentions some brands and links some pages. Visibility is no longer a position; it's whether your content shaped the answer, whether your brand is named, how prominently, and whether you're linked.

So "GEO in SEO" is not a separate discipline that replaces SEO. It's an extension with a different target:

• SEO target: the ranked results page
• GEO target: the generated answer, and the sources it draws on

Most GEO work depends on SEO foundations. An engine can only cite pages it can crawl and retrieve, and most retrieval still runs through search indexes. For a side-by-side view, see GEO vs SEO.

Where does the term GEO come from?

The term was introduced in GEO: Generative Engine Optimization by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, with authors from Princeton University, IIT Delhi and independent researchers. It was first posted on arXiv on November 16, 2023, revised on June 28, 2024, and accepted to KDD 2024.

The paper's framing: generative engines improve things for users and for the engines, but leave the third party, website and content creators, with "little to no control" over when and how their content appears. GEO is proposed as a way for creators to improve their visibility in generative engine responses.

How the study was set up

• A benchmark, GEO-bench: 10,000 queries drawn from nine sources and 25 domains, split into 8,000 train, 1,000 validation and 1,000 test queries. About 80% are informational, with 10% each transactional and navigational.
• A research generative engine: for each query, the top 5 Google results were fetched and passed to gpt-3.5-turbo, which wrote an answer with citations. Five responses were sampled per query to reduce noise.
• Visibility metrics: a position-adjusted word count (how many words in the answer are attributed to a source, weighted toward earlier positions) and a subjective impression score (an LLM-judged score across seven aspects, including relevance to the query, how much the answer relies on the source, uniqueness, perceived prominence, and likelihood of a click).
• Nine optimization methods, each applied by prompting an LLM to rewrite a source page: Authoritative (more persuasive tone), Statistics Addition, Keyword Stuffing, Cite Sources, Quotation Addition, Easy-to-Understand, Fluency Optimization, Unique Words and Technical Terms.
• A real-world check on Perplexity.ai, using 200 test queries with the sources supplied as file uploads.

What did the GEO paper find?

Method	What it changes	Result reported in the paper
Cite Sources	Adds citations to credible sources	Among the top methods: 30–40% relative improvement on position-adjusted word count and 15–30% on subjective impression
Quotation Addition	Adds relevant quotes from credible sources	Among the top methods; best single method on Perplexity at +22% position-adjusted word count
Statistics Addition	Replaces qualitative claims with numbers	Among the top methods; the combination with Fluency Optimization performed best overall
Fluency Optimization, Easy-to-Understand	Improves readability	15–30% visibility gains
Authoritative	More persuasive, confident tone	No significant improvement overall; helped on debate-style and historical questions
Keyword Stuffing	Adds more query keywords	Little to no improvement; 10% worse than baseline on Perplexity

The headline figure in the abstract: GEO "can boost visibility by up to 40%" in generative engine responses. The best methods improved on the baseline by 41% on position-adjusted word count and 28% on subjective impression in the main benchmark.

Three further findings matter for practitioners:

1. Effects depend on the domain. Cite Sources helped most on factual questions. Statistics Addition helped in areas like law and government and on opinion questions. Quotation Addition worked best for people and society, explanation and history queries. Authoritative tone helped on debate-style and historical questions.
2. Lower-ranked pages gained the most. When all sources were optimized, Cite Sources raised visibility for the source ranked fifth in search by 115.1%, while the top-ranked source's visibility fell by 30.3% on average. The authors read this as GEO giving smaller sites a way to compete on content rather than backlinks.
3. SEO habits don't transfer automatically. Keyword stuffing, a classic SEO tactic, did not help.

What the GEO paper does not show

Read the paper as evidence of mechanism, not as a playbook with guaranteed numbers.

• The engine is a 2023 research setup. It used gpt-3.5-turbo reading the top 5 Google results. Today's ChatGPT, Gemini and Perplexity run different models and different retrieval systems.
• Retrieval was held fixed. The study measured how much of the answer a source earned once it was already among the retrieved sources. It says nothing about how to get retrieved in the first place, which for real engines depends on search indexes, crawler access and rankings.
• Rewrites were done by an LLM under controlled conditions. On a live site, content changes interact with rankings, freshness and everything else.
• Brand mentions weren't the metric. The study measured word share and impression of sources, not whether an engine recommends a company for a commercial query.

That's why GEO in practice has two halves: get retrieved (a search and crawling problem), then be the source the model prefers to use (a content problem the paper speaks to).

How do generative engines choose sources?

Real engines follow a similar pipeline:

1. Query rewriting and fan-out. The engine turns a prompt into one or more search queries. Google's AI features documentation describes AI Overviews and AI Mode issuing multiple related searches across subtopics ("query fan-out"). OpenAI's ChatGPT search help page says ChatGPT typically rewrites queries into one or more targeted queries for its search providers.
2. Retrieval from an index. Google uses its own index. ChatGPT uses third-party search providers plus OpenAI's own crawler; OpenAI's crawler documentation says sites that block OAI-SearchBot will not be shown in ChatGPT search answers. Perplexity's bot documentation describes PerplexityBot as the crawler that surfaces and links websites in Perplexity search results.
3. Reading and selection. The model reads passages from the retrieved pages and decides what to use.
4. Synthesis with citations. It writes the answer, names brands, and links some sources.

Alongside this, the model's training data shapes background knowledge: how it describes your category and your brand even before it searches.

GEO strategies that follow from the evidence

1. Make sure you can be retrieved. Allow the search crawlers of each engine you care about, fix indexing in Google and Bing, and check your CDN isn't blocking AI bots. The free AI crawler checker covers the robots.txt side.
2. Cover the sub-queries. Map the questions an engine will fan out into for your core topics and make sure you have a page or section for each. The query fan-out tool generates a starting list.
3. Write answer-first passages. Open each section with a direct two to four sentence answer. Easy-to-understand, fluent text scored well in the paper, and it's easier for any model to lift.
4. Add evidence. Cite credible sources, include specific numbers with their origin, and quote experts or primary documents where it fits. These were the paper's strongest methods.
5. Adapt to the domain. For factual topics, citations. For opinion or policy topics, data. For explanations, quotes. Don't apply one template to everything.
6. Work on third-party coverage. For commercial prompts, engines often cite comparison articles, reviews and forums. Find which domains are cited for your prompts (AI citation tracking) and work to be present and accurately described there.
7. Keep facts consistent. Same category, pricing and positioning on your site, profiles and listings, so the model doesn't blend conflicting versions.
8. Skip the tactics that didn't work. Keyword stuffing and a more "authoritative" tone showed no reliable benefit.

How do you measure GEO?

Search Console shows rankings. There's no equivalent for most AI engines, and answers change between runs. GEO measurement means running a fixed set of prompts on a schedule and tracking:

• Mention rate: share of answers that name your brand
• Position: where in the answer you appear
• Citation share: how often your pages are linked as sources
• Share of voice: your mentions against named competitors
• Sentiment and accuracy: whether the description is positive and correct

aeotime is our product and tracks these across ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews, AI Mode, Grok and DeepSeek. Answers are collected through each engine's official API with web search on (Google surfaces via SERP data), and API answers can differ slightly from what a logged-in user sees. More on how we approach it on the generative engine optimization page.

Getting started checklist

• Read the GEO paper abstract and note which query types match your market
• Confirm AI search crawlers can reach your key pages
• Write 20–50 buyer-style prompts and record a baseline across engines
• Rewrite your top pages answer-first, with cited sources, numbers and quotes
• List the third-party domains cited for your prompts and plan outreach
• Re-measure monthly against competitors

To get a first baseline for free, run the AI visibility checker.