What generative engine optimisation (GEO) is, and how we measure it

Key facts

About one in five people start a search in an AI assistant; among Gen Z it is roughly one in three (SE Ranking survey, 2026, via Reporter Outreach).
StatCounter measured ChatGPT at 76.9% of AI-assistant usage in April 2026, with Gemini, Perplexity, Copilot and Claude sharing the rest (Sedestral).
Gartner expects roughly a quarter of organic search traffic to shift toward AI assistants by the end of 2026 (Stackmatix).
The large majority of AI answers end without a click, so being named inside the answer is the goal, not the referral (Prowlo).

Generative engine optimisation (GEO) is the practice of making a company the cited answer in AI assistants such as ChatGPT, Gemini, Perplexity and Copilot, and in Google AI Overviews. Where SEO optimises a page to rank in a list of links, GEO optimises an entity and its content to be named, quoted and recommended inside a generated answer. It is measured as share of answer: the percentage of a fixed set of buyer prompts for which the company is mentioned or cited.

Why GEO is not just SEO with a new name

The technical layer overlaps: both need a crawlable site, fast pages, structured data and clear internal links. The difference is in what gets evaluated. A search engine evaluates pages. A generative engine evaluates an entity, the sources it trusts about that entity, and whether the content it finds can be quoted accurately in one or two sentences. Three consequences follow:

Entity consistency matters more than keywords. If your name, description, address and contact details differ across your site, your LinkedIn page, a review platform and a directory, the engine cannot resolve who you are and will name a competitor it can resolve.
Citable beats long. An answer-first paragraph with a definition, a date and a number is quoted. A 2,000-word page that builds up to the point is not.
Off-site sources carry weight. Engines cite review sites, directories and press more readily than a company's own pages. Being present where your category is already cited is part of the work.

The engines that matter, and why you cannot measure just one

Usage is concentrated but not exclusive. StatCounter's April 2026 figures put ChatGPT at 76.9% of AI-assistant usage, with Gemini around 9%, Perplexity around 8%, Copilot around 4% and Claude around 3%. That split changes month to month and differs by audience: B2B buyers skew more towards Perplexity and Copilot than the general population. Measuring ChatGPT alone tells you about ChatGPT. Measuring five engines tells you about your market.

How we measure share of answer

Step	What we do	Why it has to be this way
1. Prompt set	25 questions a real buyer asks before choosing a supplier in your category, agreed with you and then frozen	Changing prompts between months makes the trend meaningless
2. Engines	ChatGPT, Gemini, Perplexity, Copilot and Google AI Overviews	Usage is spread across them and differs by buyer type
3. Run and record	Each prompt run on each engine on the same day each month; answers stored with screenshots	Answers change daily; a dated record is the only reproducible evidence
4. Count	For each answer: are you mentioned, are you cited, which competitors are named, which sources are cited	Mentions and citations behave differently and need different fixes
5. Report	Share of answer per engine and overall, the trend, the sources behind competitor mentions, and the three changes to make next	The number only matters if it tells you what to do

What moves the number

In the order we usually do it: fix entity consistency everywhere engines look; publish an llms.txt and page-level structured data so the entity is machine-readable; rewrite the pages that answer the buyer prompts in a citable format, with definitions, dates and sourced numbers; then earn presence on the off-site sources the engines already cite for your category. Technical and entity fixes show inside weeks; share of answer typically moves over two to four monthly cycles as engines re-crawl and re-cite.

A note on the numbers in this article

Statistics about AI search vary widely by methodology. Survey data (who says they start with AI), usage share (page views across assistants) and referral data (clicks that reach a website) answer different questions. We cite the source for each number and treat them as directions, not decimals. The reproducible number is the one you measure on your own prompt set.

Where ITovio fits

The method above is the AI Visibility Audit: a share-of-answer baseline for 25 prompts across five engines, a technical SEO audit, an entity-consistency check and a prioritised plan, in two weeks. If you want a first look, the free 5-prompt AI visibility check runs five buyer prompts and sends you the screenshots with who gets named and why.