Generative engine optimization: how AI answers pick sources and what to change on your site

What generative engine optimization means and where the term comes from

Generative engine optimization is a name for a new job: making your content visible inside answers written by AI. The term was introduced by researchers Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, in a paper called GEO: Generative Engine Optimization on arXiv. It was first submitted in November 2023 and later accepted to KDD 2024.

The paper describes generative engines as search engines built on language models that gather information and summarise it into one answer. Its authors say creators have little control over whether and how their content appears there. They propose GEO as a way to improve visibility, and a benchmark called GEO-bench to test it. Their abstract reports a visibility boost of up to 40 percent in their experiments and notes that what works varies by domain. That is a laboratory result, not a promise for your site.

How AI answers choose which pages to show

Google’s page about AI features and your website explains the mechanism on its own platform. AI Overviews and AI Mode may use what Google calls query fan-out: several related searches run across subtopics, and the results feed one answer with supporting links. To be shown as a supporting link, a page must be indexed and eligible to appear in Google Search with a snippet. Google adds that indexing and serving are never guaranteed, even for pages that meet every requirement.

OpenAI documents its own crawlers. OAI-SearchBot is the one that surfaces websites in ChatGPT’s search features, and a site that blocks it will not appear in ChatGPT search answers, though it can still show up as a navigational link. GPTBot is a separate crawler that collects content that may be used to train models, and ChatGPT-User acts when a person asks ChatGPT to visit a page. OpenAI says a change to the search rules in robots.txt can take about 24 hours to take effect.

Generative engine optimization against SEO: what stays and what changes

Much stays the same. Google states that there are no additional requirements to appear in AI Overviews or AI Mode, and no special optimizations are necessary. It lists the ordinary practices instead: allow crawling in robots.txt, link your pages internally and keep structured data consistent with the visible text. It also says no new machine-readable files or special schema markup are needed. If a page cannot be crawled or indexed, no assistant that relies on search will quote it.

Google’s guidance on helpful content asks for pages written mainly for people, and it suggests three questions: who created the content, how it was made, and why it exists. Those questions are close to what a careful assistant needs, because it must decide whether a page can be trusted before it repeats a claim from it. Bylines, an About page and honest sourcing are old SEO habits that now do double duty.

What changes is the shape of the page. An assistant quotes passages, not whole articles, so a passage must make sense alone: the question, a plain answer in the first sentence or two, then the detail. Names matter too. A company, a product and a city should be written the same way everywhere, so the system can tell they are one thing. And a fact with a source next to it is easier to repeat than a fact that only claims to be true.

The site changes a small business makes first

Start with answer blocks. For each question a buyer asks before ordering, such as what the service includes, how long it takes, who it is for and what affects the price, put a heading in the buyer’s words and answer in the first two sentences. Keep real numbers only where you can stand behind them, and say what the price depends on where you cannot give a number. This is also what a human visitor prefers.

Add a question section to the pages that need it, written from real customer questions rather than invented ones. Mark the page up with structured data where it fits. Google’s introduction says structured data is a standardised format for describing a page, recommends JSON-LD, and requires that the markup describe content visible on the page. It does not guarantee a special result in search. Treat it as a clear label, not a lever.

Make the company facts easy to find and consistent: the full name, what you sell, the countries or cities you serve, opening hours, contact details, and who writes the content. Put them on an About page and on every service page where they matter. Check that the site loads fast, the text is real HTML rather than an image, and the pages you want quoted are not blocked by robots.txt or a noindex tag.

Off the site: the same facts in profiles and directories

Assistants do not rely on your website alone. They also meet your company in maps listings, marketplace profiles, social pages, directories and articles by other people. If the name is spelt three ways, the address differs and the service list is out of date, the picture they build is blurry. Pick one official spelling of the name, one description of what you do and one set of contact details.

Then update every profile you control to match, and correct the outdated ones first. Ask partners and customers who mention you to use the same name. Reviews should come only from real customers who agreed to leave them. This is slow, ordinary work, and it also helps people who find you through maps or social networks.

How to check whether assistants mention you

The simplest test is to ask. Write twenty questions in the words your buyers use, mixing broad ones such as the best option for a need, narrow ones about your service and local ones with a city. Put each into ChatGPT, Gemini, Perplexity and Google, and record whether you are mentioned, whether the facts are right and which sources are linked.

Answers change from one day to the next, and one run proves little. Repeat the same list monthly, in the same order, and compare. For Google, Search Console includes traffic from AI features in its Performance report under the Web search type, according to Google’s documentation. Treat the whole exercise as a way to find wrong facts and missing pages, not as a ranking.

What does not work, and why

Google’s spam policies name the old tricks. Keyword stuffing means filling pages with keywords or numbers to manipulate rankings. Hidden text means putting content on a page only for search engines, for example invisible text or text placed off-screen. Cloaking means showing search engines different content than people see. Scaled content abuse covers generating many pages mainly to manipulate rankings, including with generative AI and without adding value.

None of this becomes safer because the reader is an assistant. Google’s helpful content guidance also warns that invented authors or false credentials undermine trust. A short, accurate page that answers one question well will beat a long, inflated one.

Do it yourself or order it: what to prepare before a GEO project

You can do the first round alone. Write the twenty buyer questions, run them through the assistants, list the wrong facts, rewrite five answer blocks and check robots.txt. What takes more time is the second round: structured data on many pages, consistent profiles, regular re-checks and rewriting pages that assistants misread. If nobody on your side will own that cycle, ordering it is reasonable.

Prepare a list of your services with plain descriptions, the questions customers actually ask, the official name and contact details, links to every profile, and the competitors you want to be compared with. Decide who approves the final text, since an assistant will quote it. A site that is slow, blocked or out of date should be fixed before any of this.

VITON13 Studio offers this as a set of packages and shows a free draft first: three of the twenty answers, or one rewritten answer block. The AI visibility check costs $13 and takes 2 to 3 working days. It runs 20 buyer questions across ChatGPT, Gemini, Perplexity and Google AI Overviews. GEO and AEO optimisation costs $113 and takes 7 to 12 working days, and monthly monitoring costs $33 a month.