SEO and GEO: how to get cited when an AI answers the question

For twenty years the job was to rank. You wrote a page, Google put it in a list, and someone clicked. The click was the point. Today a large share of questions get answered on the results page itself, or inside ChatGPT, or inside Perplexity, and the person never reaches a list at all. They read an answer, and under it they see 3 or 4 sources.

That is the whole shift. You are no longer competing for position 1. You are competing to be one of the sources the answer is built from. The trade calls this GEO, generative engine optimisation, and it sits on top of SEO rather than replacing it.

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

SEO is optimising to be ranked. GEO is optimising to be quoted. SEO asks whether a search engine can find your page and judge it relevant. GEO asks whether a language model, having found your page, can lift one paragraph out of it and use that paragraph as an answer without breaking it.

The two overlap heavily. An answer engine still has to crawl you, still reads your titles and headings, still cares whether the page loads. What GEO adds is a writing constraint: the useful unit is no longer the page, it is the paragraph.

Why did the rules change?

Google began rolling out AI Overviews in the United States in May 2024, and has expanded them since. ChatGPT, Perplexity and Copilot answer with inline citations. In all of these, the interface shows a written answer first and links second.

The practical consequence for a small business: fewer people land on your homepage after a general question, and more people arrive already knowing what you do, ready to book. Your traffic number can fall while your enquiry number holds. Judge the site on enquiries, not on sessions.

How should a page be written for an answer engine?

Write so that any single paragraph still makes sense with the paragraphs around it deleted. That is the entire technique, and it is harder than it sounds, because good prose usually leans on what came before.

Three habits get you most of the way:

• Name the subject in every paragraph. Not "it takes three days" but "a NewWebsite.builders build takes three business days". A model quoting the second sentence carries your name with it. A model quoting the first has nothing to attribute.
• Ask the question in the heading, answer it in the first sentence underneath. No wind-up. The first sentence after a heading is the sentence most likely to be extracted.
• Put a number in the answer. Numbers are what makes a passage worth quoting instead of paraphrasing. "Fast" is not quotable. "Under 2.5 seconds on a 4G connection" is.

If a paragraph needs the paragraph above it to make sense, an answer engine will either skip it or quote it wrong. Both are your problem, not the engine's.

What structured data does a small business site need?

Structured data is a block of JSON in the page that states, in a fixed vocabulary, what the page is about. Crawlers read it before they read your design. Four types cover almost every small business site:

This article carries BlogPosting and BreadcrumbList. Here is the shape, trimmed to the fields that matter:

{
  "@context": "https://schema.org",
  "@type": "BlogPosting",
  "headline": "SEO and GEO: how to get cited when an AI answers the question",
  "datePublished": "2026-08-18",
  "inLanguage": "en",
  "author": { "@type": "Organization", "name": "NewWebsite.builders" },
  "publisher": { "@type": "Organization", "name": "NewWebsite.builders" },
  "mainEntityOfPage": "https://newwebsite.builders/blog/en/..."
}

Structured data does not make a weak page rank. It removes ambiguity from a good one, which is a different and smaller job. Do it because it costs 20 lines, not because it is magic.

Does classic SEO still count?

Yes, and it is still the larger half. An answer engine cannot cite a page it cannot fetch, cannot render, or cannot finish loading. Every one of the old fundamentals is now a prerequisite for the new game:

• One <h1> per page that says what the page is.
• A real <title> and meta description written for a human, not stuffed.
• A canonical URL on every page, pointing at itself.
• hreflang alternates if you publish in more than one language, in both directions.
• A sitemap that lists every page you want found.
• Core Web Vitals inside Google's thresholds, which is easier on a hand-coded site than on a page builder: Largest Contentful Paint under 2.5 seconds, Interaction to Next Paint under 200 milliseconds, Cumulative Layout Shift under 0.1.

What to change on your site this month

Seven changes, in the order they pay off:

1. Rewrite your 5 most important pages so each heading is a question a customer actually asks.
2. Make the first sentence under each heading a complete answer, with your business name in it.
3. Put a number in every claim. Replace "quickly" with the actual figure.
4. Add Organization and ProfessionalService structured data to the site, and FAQPage to your questions page.
5. Publish a real FAQ, with the questions written the way people say them out loud. The one sentence that describes your business belongs at the top of it.
6. Check your Core Web Vitals in Google Search Console and fix anything outside the three thresholds above.
7. Make sure your robots.txt does not block the crawlers you want citing you, and that your sitemap is submitted.

How do you know whether it is working?

Ask the engines directly. Once a month, put your own 10 buying questions into ChatGPT, Perplexity and Google, and write down whether you were named. That list, kept over 6 months, tells you more than any rank tracker, because it measures the thing you now care about: being the source.

Then keep one number from the old world. Track enquiries per month, not sessions per month. If citations go up and enquiries go up while traffic goes flat, the site is doing its job.