The technical signals that get pages cited by AI
Most advice on getting cited by AI answer engines is about the words. A 2025 study of 1,702 real citations points somewhere else — the technical structure of the page. Pages with clean structured data, semantic HTML, and current metadata were cited far more often than pages that read well but gave a machine little to work with.
The full study and its capsule sit in the AI Citation & GEO research index. This post covers what the framework measured and what a marketing team should take from it.
What the GEO-16 study measured
Kumar and Palkhouski built a scoring framework, GEO-16, that rates a page against sixteen on-page quality signals and rolls them into one normalized score. They analyzed 1,702 citations across three AI answer engines and tested which signals separated cited pages from the rest.
Three groups did most of the work: metadata and freshness, semantic HTML, and structured data. A logistic regression backed the headline finding that overall page quality is a strong predictor of whether a page gets cited at all.
The threshold worth knowing
The study reports a usable line, not just a direction. Pages that scored at least 0.70 on the normalized scale while hitting at least 12 of the 16 signals were cited substantially more often than pages below that mark.
That is a concrete target. Clear the quality bar on most of the signals, not one or two, and citation rates rise. A page that nails a single signal and ignores the rest does not get the same lift.
Why machines reward structure
An answer engine reads a page as data, not as writing. Structured data hands it labeled facts. Semantic HTML — real headings, lists, and sections — gives it a map of what the page is about. Current metadata tells it the page is live and maintained. A wall of good prose offers none of those handholds, which is why prose alone underperformed in this study.
This lines up with earlier GEO work showing that structure improves citation independently of the content itself. The words still matter; they are no longer the only thing that does.
What to change on a marketing page
The fixes are technical, and most are one-time. In rough order of payoff:
- Add structured data (schema.org JSON-LD) for articles, products, and FAQ blocks, so engines get labeled facts instead of guessing from layout.
- Use semantic HTML — real
<h2>and<h3>headings, lists, and tables — instead of styled<div>soup. - Keep metadata current. Publish and updated dates, titles, and descriptions signal a maintained page; refresh anything stale.
- Front-load a self-contained answer the engine can lift without the rest of the page around it.
Copper Sun applies the same structure-over-prose principle inside the product. Brand context is stored as labeled, directive statements an AI can extract, not as a brand guide it has to parse first. See how it works.
Frequently Asked Questions
Do these technical signals matter more than the content?
No. The content still has to answer the query and carry specific, sourced claims, and earlier GEO research is clear on that. What the GEO-16 study adds is that technical structure is a separate, measurable signal on top of good content. A well-written page with no structured data and stale metadata leaves citation signal on the table.
What is structured data, in practice?
Structured data is machine-readable markup, usually schema.org JSON-LD, that labels what is on a page: this is an article, this is its author, this is a FAQ, this is a product and its price. It does not change what a human sees. It gives an answer engine labeled facts instead of leaving it to infer them from the visual layout.
Is a normalized score of 0.70 a guarantee of citation?
No. The 0.70 score and 12-signal threshold come from one study across three answer engines. They mark where citation rates rose substantially, not a promise for any single page. Read it as a quality bar to clear, not a switch that turns citation on.
How is this different from traditional SEO?
There is heavy overlap, since semantic HTML, structured data, and fresh metadata help both. The difference is the goal: SEO earns a ranked link a human clicks, while these signals help an answer engine extract and cite a passage directly. SEO and GEO target different outcomes, but the technical hygiene serves both.