What is generative engine optimization (GEO), and how is it different from SEO

Generative engine optimization (GEO) is the work of making a business or a page more likely to be shown and cited when an AI assistant such as ChatGPT, Gemini, Perplexity or Claude writes an answer. The term comes from a 2024 study by researchers at Princeton and IIT Delhi, which found that editing content can raise its visibility in AI answers by up to 40%. GEO builds on SEO rather than replacing it. The main difference is the goal: SEO aims for a position in a list of links, while GEO aims to be named, described correctly and cited in a single written answer.

Where the term comes from

The name was introduced in "GEO: Generative Engine Optimization" by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande of Princeton and IIT Delhi. It was accepted at KDD 2024, a peer-reviewed data science conference. The authors use "generative engines" for search systems that read several sources and write one answer with citations, as opposed to a list of ten links.

They built a benchmark, GEO-bench, of roughly 10,000 queries drawn from nine datasets, and ran a controlled experiment: change the text of a source page in specific ways, then measure how visible that source is in the generated answer. Their abstract states:

"GEO can boost visibility by up to 40% in generative engine responses."

The methods that worked best were adding citations to credible sources, adding quotations, and adding concrete statistics. The authors also found that what works differs by topic, and they argue for domain-specific methods. One limitation matters for local businesses: the queries were general questions, not searches like "dentist near me". The study shows that content changes can shift AI visibility. It does not measure local recommendations.

The Toronto preprint: AI search leans on third-party sources

A second paper, "Generative Engine Optimization: How to Dominate AI Search" by Mahe Chen, Xiaoxuan Wang, Kaiwen Chen and Nick Koudas of the University of Toronto, was posted to arXiv on September 10, 2025. It is a preprint and has not been peer reviewed. It is also an observational comparison, not an experiment.

The authors compared the sources cited by AI search (ChatGPT, Perplexity, Gemini) with Google results. They report that AI search shows a systematic bias toward what they call earned media, meaning third-party, authoritative sources, and gives comparatively little weight to brand-owned pages and social media. Google's results showed a more balanced mix. They also describe a "big brand bias" that smaller companies have to overcome, and differences between engines in source diversity, freshness, stability across languages and sensitivity to how a question is phrased.

Why it seems to disagree with Yext

Yext's study of 6.8 million citations in ChatGPT, Gemini and Perplexity answers (October 2025) points the other way: 44% of citations went to businesses' own websites and 42% to directory listings, which Yext counts together as 86% from sources a business manages. The two studies asked different kinds of questions. Yext ran 1.6 million queries per model about locations and services in retail, finance, healthcare and food service, where the natural answer is a business's own page or listing. The Toronto team compared broader queries across engines. Yext also sells listing management, and the Toronto paper is unreviewed, so neither settles the question.

A reasonable reading of both: for questions about a specific place or service, the business's own website and listings are the main material. For broader "which is best" questions, outside sources such as articles, guides and review sites carry more weight. In Whitespark's 2026 survey of 47 local search specialists, presence in expert "best of" lists was rated the top factor for AI visibility. That is expert opinion, not a measurement, but it fits the same pattern.

What stays the same as SEO

AI assistants depend on search indexes, so most SEO basics still apply.

• Being crawlable and indexed. Google says there are no additional requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible for a snippet. ChatGPT search relies largely on the Bing index (BrightLocal, December 2024). A page that search engines cannot read is a page assistants cannot cite.
• Allowing the right crawlers. OpenAI says sites that block OAI-SearchBot are not shown in ChatGPT search answers.
• Accurate business listings. Google's local ranking guidance lists relevance, distance and prominence, and says more reviews and positive ratings can help. The same listing data now feeds AI answers.
• Clear, specific content. Pages that state facts plainly are easier to rank and easier to quote.

What is different

The goal is being named, not ranking

A search results page shows ten links and a map shows a list. An AI answer often names a handful of businesses, or one. SOCi's 2026 Local Visibility Index, covering more than 350,000 locations of 2,751 chain brands, found that ChatGPT recommended 1.2% of them while 35.9% appeared in the Google Maps top three. The locations ChatGPT recommended averaged a 4.3 rating. This is a vendor report on chain locations, so it shows correlation, not cause, but it shows how narrow the answer is.

Each assistant reads different data

There is no single AI index. Gemini answers place questions with Google Maps data, including ratings and reviews (Google, October 2025). ChatGPT has licensed Yelp reviews, ratings and photos since July 23, 2026 (PYMNTS). Perplexity licenses Yelp data (since March 2024) and Tripadvisor data (since January 2025). In SEO, one Google ranking often stood for the whole picture. In GEO, a business can look strong in one assistant and absent in another.

Listings and reviews feed the answer directly

In classic SEO, a Yelp page or a Google Business Profile is a separate result. In AI answers, the assistant reads those listings and repeats their facts: the rating, the hours, the price level, what reviewers say. BrightLocal's 2025 comparison found Yelp was a source in 33% of queries across four AI tools, and that Google's AI shows Business Profile data even when it cites another source. Listing accuracy and review content are now part of the answer text, not only ranking signals.

Quotable content

The Princeton and IIT Delhi results suggest that assistants favor content they can quote and credit: a figure, a short quote, a cited source. For a business website that means stating the facts a customer asks about in plain sentences, with numbers where they exist: what you offer, where, for whom, price range, hours, and how to book.

How to measure GEO

Classic rank tracking measures positions in a list of results, not whether an assistant named you. Measurement has to be built by hand or with tools, and it works best when it is boring and repeatable.

1. Write a fixed question set. Ten to twenty questions a real customer would ask, in their words, for example "best family dentist in Glendale that takes walk-ins". Include questions with and without your business name.
2. Run every question in every assistant. ChatGPT, Gemini, Perplexity, Claude and Copilot at minimum, since each reads different data.
3. Keep the wording fixed. The Toronto preprint reports that engines differ in sensitivity to phrasing, so changing words between runs mixes up the comparison.
4. Save every answer with its date. Screenshots or text exports. Record whether you were named, in what position, how you were described, whether any facts were wrong, and which sources were cited.
5. Repeat weekly. Assistants and their data deals change, as the July 2026 Yelp and OpenAI license showed. A weekly record shows whether a change in your listings or website was followed by a change in answers.

The cited sources are often the most useful column. They show which website, listing or article an assistant trusts for your category, which tells you where to work next.

That is also how AIReputa approaches GEO for local businesses: a fixed question set across the main assistants, saved answers, fixes to the sources each assistant reads, and the same questions again. We commit to the work and the measurement. No one can promise what an assistant will say.

The audience is growing. In BrightLocal's 2026 survey of 1,002 US adults, 45% said they use AI tools for local recommendations, up from 6% a year earlier. That is what people say rather than measured behavior, but it is the reason the measurement is worth doing now.