Generative Engine Optimization: How to Get Cited by AI

This month we moved our own website off a page builder and onto our own server. One reason was simple. We wanted the HTML our server sends to already contain everything an AI crawler needs: the answer, the facts and the labels that say what each page is. No waiting for JavaScript to fill the page in.

That change is a small piece of what people now call generative engine optimization. This post explains what it is, what the research behind it found, and the steps any business can take to get cited by ChatGPT, Claude and other AI answer tools.

What is generative engine optimization?

Generative engine optimization (GEO) is the work of making your content easy for AI answer tools to find, understand and cite. A "generative engine" is any tool that writes an answer from several sources, such as ChatGPT, Perplexity or Google's AI answers. GEO aims to make your page one of those sources, named in the answer.

Classic SEO tries to get your page into a list of ten links. With AI answers, there is often no list. The tool reads several pages, writes one answer and adds a few source links. If your page is not one of them, the person may never visit your site.

So the goal shifts from "rank" to "be quoted". That changes what good content looks like.

What did the original GEO research find?

The term comes from a 2023 paper by researchers at Princeton and other institutions. They tested ways of editing web pages and measured how visible each page became inside AI-written answers. The best methods raised visibility by up to 40%. Adding quotations, statistics and source citations worked best. Keyword stuffing did not help.

The paper is GEO: Generative Engine Optimization by Aggarwal and colleagues, later accepted at KDD 2024. They built a benchmark of real search questions, ran them through a generative engine, then rewrote the source pages in different ways and measured the change.

Three findings matter most for a business:

• Evidence beats keywords. Adding a relevant quote, a statistic, or a citation to a trusted source were among the strongest methods. Repeating the keyword more often made results slightly worse.
• Lower-ranked pages can gain the most. Pages that ranked lower in normal search results saw the biggest gains from these edits, while top-ranked pages sometimes lost visibility.
• It depends on the subject. Methods that helped in one field did less in another, so there is no single trick.

The second point is the encouraging one. A firm that has never ranked first on Google can still be the page an AI tool chooses to quote, if its page is the clearest and best supported.

How do you get cited by ChatGPT and other AI tools?

Answer the question in the first two or three sentences of the page or section, back it with specific facts and sources, and make sure the page loads fully without JavaScript. Then label it with schema markup in the server HTML. AI tools quote what they can read quickly and trust.

Here is what that looks like in practice.

Lead with the answer. If a page is called "How long does probate take?", the first lines should say how long it usually takes and what changes it. Background comes after. AI tools often lift the first clear answer they find.

Use real facts, with sources. A sentence like "most cases we handle take between X and Y months" is quotable. "Every case is different" is not. Where a fact comes from somewhere else, link to the original source.

Write the questions your buyers ask. Use their words as headings. A clinic should have a heading like "Does the treatment hurt?", not "Patient comfort".

Make the page readable without JavaScript. A study by Vercel found that none of the major AI crawlers currently render JavaScript, including OpenAI's GPTBot and Anthropic's ClaudeBot. If your text only appears after scripts run in a browser, those crawlers may see an empty page.

Add schema markup on the server. Schema markup is a set of labels in the page code that say "this is an FAQ", "this is a service" or "this is our company". Put it in the HTML your server sends, so a crawler that does not run scripts still sees it.

What did we change on our own site?

We rebuilt our site so pages are server-rendered, meaning the server sends finished HTML with the text and schema already in place. Our service pages open with a direct answer under each question heading and carry schema for the organization, the service and its FAQs.

A page that is assembled in the browser works fine for a person with a modern browser. It is a gamble for a crawler that only reads the raw HTML.

The changes were not exotic:

• Main text and schema ship in the first HTML response, not after scripts run.
• Each question heading is followed by a short, direct answer before any detail.
• Facts about the company, such as when it started and how many projects we have built, should read the same way everywhere, so tools do not see conflicting versions.

That last point is easy to miss. If your About page says one founding year and your footer says another, an AI tool has to guess. Consistent facts are easier to repeat.

We are not going to claim results we have not measured yet. What we can say is that a crawler reading our pages now gets the full text and the labels on the first request.

How is GEO different from SEO?

SEO and GEO share most of the basics: a fast, crawlable site, useful content and trusted links. The difference is the target. SEO aims to rank a page as a whole. GEO aims to have specific sentences and facts lifted into an answer, so each section must stand on its own and be clearly sourced.

In practice, that means a few habits change.

Long intros hurt more under GEO, because the answer gets buried. Vague claims hurt more, because there is nothing to quote. Pages that mix five topics hurt more, because the tool cannot tell which part answers the question.

Some habits stay exactly the same. You still need pages that load fast, that search engines can crawl, and that other sites link to. GEO builds on SEO. It does not replace it.

How do you measure whether it is working?

Pick 20 to 30 questions your buyers really ask, put them to ChatGPT, Claude, Perplexity and Google's AI answers once a month, and note whether your site is cited. Track the same list over time. It is manual, but it shows real movement better than any single score.

Keep the log simple: the question, the tool, the date, and whether you were cited or named. Add a note on which competitor was cited instead. Their page usually shows you what yours is missing, whether that is a clearer answer, a statistic or a better source.

Also watch your server logs or analytics for AI crawlers visiting, and for visits coming from AI tools. Neither number is perfect, but both trend in a useful direction when the work is right.

Be wary of anyone who promises citations or rankings. AI tools change how they choose sources often. What anyone can honestly promise is the work: clearer pages, better evidence, clean technical setup, and steady checking.

If you would rather not do this by hand, our SEO agent does the repeat work: it builds answer-first pages, adds schema in the page itself, keeps your business details consistent and checks every month whether AI assistants mention you. You review the pages before they go live, and we promise the work, not rankings.