• AI visibility

Guide · AI search

What is GEO, and how is it different from SEO?

Generative engine optimization (GEO) is the work of getting your information found, used and correctly cited by AI systems that write answers instead of listing links. SEO competes for a position on a ranked list. GEO competes for a share of one written answer, chosen by an engine that often reads different sources from Google.

By Underneath Team Guide Updated

The short version

  1. The term comes from a 2023 paper by Princeton and IIT Delhi researchers, which reported that content changes could raise a page’s share of AI answers by up to 40% in a simulated AI search engine (Aggarwal and colleagues (opens in a new tab)).
  2. Across 55,936 queries, only 38% of the domains cited by six AI search engines also appeared in Google or Bing results (Zhang and colleagues (opens in a new tab)).
  3. In US car questions, 81.9 percent of the sources an AI engine cited were independent “earned” media, against 45.1 percent for Google (Chen and colleagues (opens in a new tab)).
  4. Keyword stuffing, a classic SEO trick, did 10% worse than doing nothing when tested on Perplexity (Aggarwal and colleagues (opens in a new tab)).
  5. Asked the same question five times, ChatGPT named only 25.2% of its brands every time, in our consistency study.

What is generative engine optimization?

GEO means shaping your content and reputation so AI answer engines find it, use it and cite it accurately. The answer engines in question include ChatGPT, Perplexity, Gemini and Google’s AI Overviews, the AI summary at the top of Google’s results.

The name was coined by Aggarwal and colleagues (opens in a new tab), researchers at Princeton, IIT Delhi and elsewhere, in a paper first posted in 2023. They framed it as a way to help content creators be seen in AI answers when the engines are closed boxes. Their headline result was that GEO “can boost visibility by up to 40% in generative engine responses.”

That 40% needs care. Martinez’s 2026 survey (opens in a new tab) of 45 studies explains that it came from a simulator in which five pages had already been handed to the AI. The best method raised a page’s weighted share of the answer from 19.3 to 27.2. It does not mean 40% more customers or 40% more citations in the wild. We cover this in detail in our article on the 40% claim.

What happens when a search engine writes the answer instead of listing pages?

The prize changes from a place on a list to a share of one answer. A classic search engine shows ten links and leaves the choice to the reader. An AI engine reads sources for the reader, writes a summary and attaches a few citations.

The foundational paper puts the difference plainly: generative engines “combine information from multiple sources in a single response,” so simple ranking metrics “are not applicable” to their answers. Visibility becomes relative. When one source gains a larger share of the answer, another loses some.

The engine also does its own searching. In our hidden-searches study, ChatGPT ran a mean of 3.7 searches per buyer question. None of the 509 searches we saw repeated the user’s question word for word. You are optimizing for searches you never see.

And AI answers favor certain kinds of question. In a 40-day audit of 55,393 trending searches in spring 2026, Xu, Iqbal and Montgomery (opens in a new tab) found that question-form searches triggered an AI Overview 64.7% of the time, against 9.5% for other searches.

Do AI search engines find the same sources as Google?

Mostly not: they overlap with Google only partly, and they lean toward third-party sources. That is the main reason ranking well on Google is not the same as appearing in AI answers.

Zhang and colleagues (opens in a new tab) compared six AI search engines with Google and Bing across 55,936 queries in 2025. Only 38% of domains appeared in both kinds of results, and 37% were unique to the AI engines. Even Google’s own AI Overviews look past its rankings: in the spring 2026 audit by Xu and colleagues, 29.8% of domains cited in AI Overviews did not appear in the first page of results at all. Our guide on how AI and Google sources overlap compares more of these studies.

The type of source shifts too. Chen and colleagues (opens in a new tab) at the University of Toronto sorted sources into brand-owned, social and “earned” media, meaning independent reviews and publications.

US car questions, 2025Earned mediaBrand-ownedSocial
Google45.1 percent39.5 percent15.4 percent
AI search (web-enabled GPT)81.9 percent18.1 percentnone

So GEO reaches beyond your website. What reviewers, publishers and directories say about you often matters more than your own pages. The ranking link still matters, as we explain in our article on whether SEO still matters.

Why is AI visibility harder to pin down than a Google ranking?

Because AI answers change from run to run and from engine to engine. A ranking is one number; AI visibility is a range of outcomes.

In our consistency study, we asked ChatGPT, Gemini and Perplexity the same 20 buyer questions five times each. Of the brands ChatGPT named for a question, 25.2% appeared in all five runs. A single ChatGPT answer showed only 57.8% of the brands its five answers named between them.

Engines also disagree with each other. Across 80 buyer questions in our brand agreement study, ChatGPT, Gemini, Perplexity and Claude all chose the same first pick for 10.0% of questions. A small 15-prompt audit by Tannenbaum (opens in a new tab), an author at a private company, found that 96.4% of cited addresses appeared in only one of four engines on the same day.

Being seen also yields fewer visits. Huang and colleagues (opens in a new tab) cite Pew Research data showing that people clicked a source inside Google’s AI summary in only 1% of visits. So the goal shifts from winning clicks alone to shaping what the answer says. We look at clicks in our article on answer engines and website visits.

What should you do about it?

Treat GEO as an extension of search work, measured on AI answers rather than rankings. Five steps cover most of it.

  1. Keep your SEO foundations. Pages that cannot be crawled or that lack authority rarely get into the pool AI engines choose from.
  2. Invest in third-party coverage. Reviews, trade press and comparison sites carry more weight in AI answers than in Google results.
  3. Write pages an engine can quote. State facts, definitions, comparisons and figures clearly, with sources.
  4. Measure AI answers directly. Ask your buyers’ questions on several engines, several times, and track how often you appear and what is said.
  5. Judge results over months, not single snapshots, because answers vary from run to run.

If you want help setting this up, see our generative engine optimization service.

What does the research not tell us yet?

The research describes how AI search differs from Google far better than it shows what reliably moves results. Several gaps remain.

  • Most GEO experiments test pages already handed to the AI. Few show that a change makes a live engine find a page in the first place.
  • No study we reviewed links GEO work to sales. Martinez’s survey notes that effects on conversions are almost never established.
  • Engines change quickly. Most findings are snapshots from one period, often in English and the US.
  • Several key studies are preprints, small, or written by companies that sell GEO tools.

Frequently asked questions

Is GEO replacing SEO?

No: GEO builds on SEO rather than replacing it. Authority and crawlability still decide whether AI engines consider a page, but AI answers draw on a different mix of sources, so SEO alone covers only part of the picture.

Who invented the term generative engine optimization?

Aggarwal and colleagues, researchers at Princeton, IIT Delhi and elsewhere, introduced it in a paper first posted in November 2023. It reported gains of up to 40% in a simulated AI search engine.

Does a top Google ranking get you into ChatGPT answers?

Not reliably. In our rankings study, only 8.3% of the pages ChatGPT cited ranked in Google’s top 10 for the question, against 25.7% for Claude.

Is GEO the same on every AI engine?

No. Engines cite different sources and different numbers of them, so a page that does well on one may be absent from another. Our brand agreement study found all four assistants shared a first pick for only 10.0% of questions.

Sources

Free strategy call

Some questions are easier to answer about your own business.

Bring the one that matters most. On a free 30-minute call we’ll take a first look at it and send you a short written read afterward.

Ask about my businessHow we work