GEO: how to get AI to cite your company

GEO (Generative Engine Optimization) is the practice of structuring a site’s content so that generative engines —ChatGPT, Perplexity, Gemini and Google’s AI answers— cite it as a source. It doesn’t replace SEO: it covers a new channel, the answers AI writes instead of showing a list of links.

How does it differ from SEO and AEO?

The three overlap, but they chase different units of success:

Discipline	What it targets	Where it shows up
SEO	Ranking in the results list	Google’s links
AEO (Answer Engine Optimization)	Winning the direct answer	Featured snippet, answer box
GEO (Generative Engine Optimization)	Being cited in the answer AI writes	ChatGPT, Perplexity, Google AI Overviews

The same well-built page helps all three. The difference is what counts as success: in SEO it’s the position; in GEO, the mention.

Why does it matter now?

More and more queries end in an AI-written answer, not a click. In that format there aren’t ten links: there’s a paragraph with two or three sources cited. If your company isn’t one of those sources, it drops out of the conversation.

And the format rewards structure — and that part is measured. The paper that coined the term GEO —Aggarwal et al., presented at KDD 2024— tested content changes against GEO-bench, a benchmark of 10,000 queries, and found that including citations, quotations from relevant sources and statistics raises visibility by over 40%. On Perplexity.ai —a real engine, not a benchmark— the improvement reached 37%.

How do you optimize so AI cites your content?

1. Write the answer first. A 40-to-60-word paragraph that answers the question, high on the page. It’s the unit the model extracts and cites.
2. Content in the HTML, not only in JavaScript. Many AI crawlers don’t run JS: whatever lives only in a canvas or is painted by scripts doesn’t exist for them.
3. Structure that can be parsed. Headings that are questions, lists, tables and definitions. The original GEO paper (Aggarwal et al., 2024) measured that adding statistics, direct quotes and sources lifts visibility by over 40%.
4. Schema that stacks. Article + FAQPage together, with stable @ids pointing at the same entity. It helps the engine understand what each thing is, but don’t count on it lifting citations by itself: Ahrefs tracked 1,885 pages that added schema against 4,000 controls and measured no uplift on any surface. With the caveat Ahrefs itself flags: those pages were already heavily cited, so the study says nothing about a page nobody sees today.
5. Authorship and evidence. A clear author, visible dates, figures with a source. AI uses attribution as a trust signal.
6. A clear entity and outside consensus. Make it unambiguous what your company is —name, what it does, where it operates— and have other sources confirm it. AI cites entities it recognizes, not orphan pages.
7. Freshness. A visible date, an “updated 2026” note, a changelog. Models prefer what looks current.
8. Don’t block AI crawlers. Before anything else: check that your firewall or CDN (Cloudflare, for example) isn’t returning 403 to ClaudeBot, PerplexityBot or GPTBot. It’s the most common mistake and the most expensive, because it cancels out everything else.

What doesn’t move the needle?

• Schema on its own. Without authority or clear content, tagging everything with JSON-LD wins no citations.
• Repeating keywords. Models understand meaning, not density.
• A magic llms.txt. It helps site hygiene, but it isn’t a citation lever by itself.

How we apply it

This very site is built this way: the citable content lives in the HTML, every page opens with its answer, there’s product and FAQ schema, and a Sonika entity with a stable identity. This article practices what it describes —look at the answer up top and the questions block at the end.

If you want AI to cite your company when someone asks about what you do, message us on WhatsApp. We build sites made, from day one, for a human to read clearly and an AI to cite well.

---

Corrected on August 25, 2026. This post said that “about 74% of citations in AI answers come from structured content”, attributed to “industry analysis”. That figure comes from no study. The 74% in circulation measures something else —the share of newly created web pages that carry AI-generated content, from an Ahrefs analysis of 900,000 pages created in April 2025, where the exact figure is 74.2%— and had been attached to a different claim. It was replaced with what is actually measured, with the source linked.

In the same pass a second claim was corrected: this post said that Article + FAQPage together get “roughly 1.8× more citations”. That figure comes from no study either, and there is a measurement pointing the other way. It was replaced with what that measurement shows, with its caveat.