Home / Knowledge / What is GEO

Definition

What is generative engine optimisation?

The complete answer is in the first paragraph. Everything after it explains why it reads that way, which systems are meant and how the result can be measured.

On this page

Last updated 5 August 2026 · Definition checked against the current literature

Distinction

What GEO is not.

GEO is not a replacement for search engine optimisation and not a technique for manipulating language models. There is no ranking factor you can set and no guarantee of a position. Anyone promising a placement in AI answers is selling something they do not control. What can be influenced is exclusively the material the systems build their answer from.

Background

Why the discipline exists at all.

Language models answer questions instead of showing result lists. That removes the intermediate step in which a user chose between ten hits themselves. The system makes the selection, based on what it finds about the providers. In Switzerland, 47 per cent of 16- to 74-year-olds use AI, the third-highest figure in Europe, and 73 per cent of those users apply it at work, the highest in Europe. 13 per cent start the search for an unfamiliar provider in a chat already.

Sources: Swiss Federal Statistical Office, survey spring 2025 · localsearch and Lucerne University of Applied Sciences, SME Digital Pulse 2025

Mechanics

How an AI answer is produced.

Simplified into three steps. First, the system breaks the question into several sub-questions and searches for each one separately, a procedure known as query fan-out. Second, it retrieves content from its index or through a search partner — for ChatGPT historically via Bing, for Google AI Overview via the Google index, for Perplexity via its own crawler. Third, it composes an answer from the retrieved passages and, in doing so, chooses which sources get named. That selection in step three is what GEO is actually about.

The three levers

What GEO actually works on.

Entity. The system has to recognise beyond doubt which company is meant. That requires matching details across the commercial register, Schema.org markup, Wikidata, industry directories and the Google Business Profile. Contradictory addresses or company names lead to no recommendation being made at all.

Source base. Language models weight sources a company controls itself far lower than independent ones. A study by Ahrefs across 75,000 brands found that the strongest correlation with visibility in AI answers was the number of brand mentions across the web, well ahead of backlinks. Ahrefs itself points out that these are correlations.

Machine readability. Content has to be retrievable, structured and self-contained. A paragraph that only makes sense in the context of the whole page is rarely cited. On top of that come technical prerequisites such as permitted crawlers and indexed pages.

Source: Ahrefs, correlation study across 75,000 brands, May 2025, with an explicit note that this is correlation rather than causation

Systems

Five systems, five mechanics.

Five are relevant in Switzerland: ChatGPT as the most used assistant; Google AI Overview as the system with the widest reach, because it appears inside normal search results; Perplexity as the system with the highest citation density; Microsoft Copilot because of the Bing index and its spread on company computers; and Claude, which is used above average in B2B settings. Each system sources its content differently, which is why there is no single measure that serves them all at once.

Measurement

How GEO can be measured.

With a fixed prompt set: an unchanged list of questions phrased the way customers actually ask them, measured monthly against the same systems. That yields the mention rate, meaning the share of answers in which a company appears at all; the citation rate, meaning the share in which it is linked as a source; and the position within the answer. Without a fixed prompt set, no statement about change is possible, because every change in wording shifts the result.

Described in full in our methodology.

Read on

GEO versus SEO

The difference in detail, and why the two belong together.

Our methodology

How we measure AI visibility, fully disclosed.

Free analysis

What the systems say about your company today.

Frequently asked

Common questions about this.

What does the abbreviation GEO stand for?

GEO stands for generative engine optimisation — optimising for generative answer systems. The term was coined in an academic paper in 2023 and has since become the established name for the discipline. The terms AEO, for answer engine optimisation, and AI SEO are sometimes used as well; they mean essentially the same thing.

No, but the two overlap heavily in technical terms. SEO targets a position in a result list, GEO inclusion in a generated answer. The overlap is measurable: Perplexity draws 91 per cent of its cited domains from Google’s top 10, Google’s AI Overviews 86 per cent. With ChatGPT, and with very specific questions, the overlap is considerably smaller. In detail in the comparison of GEO and SEO.

The first measurable changes in the mention rate typically appear after six to twelve weeks, substantial results after four to six months. The reason for the delay lies less with the models than with the sources: a specialist article first has to be published, indexed and retrieved before it shows up in answers.

Attempts exist, such as hidden instructions in the source code. They work unreliably, are actively fought by the providers and, in the worst case, put your indexing at risk. The commercial angle matters more: a recommendation built on manipulation does not survive the customer’s first follow-up question. Provable substance is the only approach that scales.

It depends on whether the offering gets researched before purchase. For services that need explaining, are expensive or depend on trust, it pays off early, because that is where the recommendation tips the decision. For pure impulse purchases nobody checks in advance, it barely matters.

Very few providers publish prices, and where they do, the packages are hardly comparable. The question only becomes meaningful with findings: how many questions, how many languages, what starting position. That is exactly what the free first analysis delivers.

Merag Shahzad

Merag Shahzad

Founder of meit, in digital marketing since 2006. Forbes Agency Council, LinkedIn Top Voice. Reachable directly at [email protected] – no account team in between, Monday to Friday, 09:00 to 18:00.

Next step

Test it on your own case.

The first analysis shows how often ChatGPT, Claude, Perplexity and Google AI Overview name your company, where you stand in Google and Bing and what the gap costs in francs. Free, reply within two working days.