What Structured Data Actually Does for AI Visibility
Having markup on a page made no measurable difference to where it appeared in an answer or how long it stayed there. One specific type was the exception.
Key Takeaways
- Pages with structured data averaged 4.49 days quoted. Pages without averaged 4.31. No meaningful difference.
- On position in the answer, pages with no markup did marginally better.
- The one exception is FAQ markup: 5.37 days, and 11.9% still quoted after fifteen days.
- Even then it does not lift you higher. It helps you stay in the answer.
- Most structured data describes the furniture of a page, not its answer, which is why it correlates with nothing.
Why we looked into this
Structured data is one of the most consistently recommended technical investments in this field. Add markup describing what your page is, and machines will understand it better. It appears in nearly every AEO checklist published in the last two years, usually near the top.
It is also real work. Implementing it properly across a large site takes engineering time, and keeping it accurate as pages change takes ongoing attention. That is a reasonable investment if it moves something. It is a waste if it does not.
We had the data to check, so we did. Rather than asking whether structured data is a good idea in principle, we asked what it measurably changes about how AI treats a page.
How we tested it
We looked at two different outcomes, because they answer different questions.
The first is position. When an engine lists its sources, they come in an order, and that order is the closest thing to a ranking AI gives us. If markup helps an engine understand a page, better understood pages might appear earlier.
The second is staying power. Because we asked every question daily for a month, we can measure how consistently a page holds its place across 31 days. Markup might not lift a page up the list but could help an engine recognise the same page reliably each time it answers.
We tested both against whether a page carried structured data at all, and then against which specific types it carried, since the types describe very different things.
What we found
Having structured data changes nothing measurable
On position, pages carrying structured data averaged a citation rank of 7.40. Pages carrying none averaged 7.24. The pages without markup came out marginally ahead, which is another way of saying there is no effect here at all.
On staying power the result is the same. Pages with markup held a citation for an average of 4.49 days. Pages without held it for 4.31. A difference too small to act on.
One type is the exception
Pages carrying FAQ markup held their citations for 5.37 days on average, about a quarter longer than pages with no markup at all. They were also the only group whose typical page was quoted on more than one day rather than exactly one.
Looked at another way, 11.9% of pages with FAQ markup were still being quoted after fifteen days or more, against 8.7% of pages with no markup.
Notice what the effect is and is not. FAQ markup did not lift pages higher in the answer. Position was unaffected. What it appears to help with is staying in the answer once you are there.
What most markup actually describes
There is a clue in what sites are implementing. Of the pages carrying structured data, the most common types by a wide margin describe the organisation, the images on the page and the navigation path. Only about a fifth carry markup that describes the substance of the content itself.
Most structured data on the web is describing the furniture of the page rather than its answer. That helps explain why having it, in general, correlates with nothing.
Why this happens
Structured data was designed for a particular job: helping a search engine display a result correctly, as a review star rating, an event date or a recipe time. It tells a machine how to present something it has already decided to show.
An AI engine has a different problem. It reads the page and decides whether the answer is in there. It does not need a label telling it the page is an article, because it can see that. Markup describing the organisation or the navigation adds nothing to that judgment.
FAQ markup is different in kind. It does not describe the page, it structures the content into explicit question and answer pairs. That is the same shape as the task the engine is performing, which is a plausible reason it is the one type that moves anything.
What this means for your business
Treat generic structured data as hygiene rather than as an AI visibility lever. It has good reasons to exist, including how your results appear in traditional search, and none of this suggests removing it. But if it is on your AEO roadmap as a growth initiative, the evidence here does not support the position it usually holds on those lists.
If you are going to invest in markup specifically for AI visibility, FAQ markup is the type with measurable support, and it makes sense to add it where a page genuinely answers discrete questions rather than applying it everywhere. Marking up content as questions and answers when it is not shaped that way is unlikely to help and may misrepresent the page.
The broader lesson is about where technical effort goes. Markup is appealing because it is concrete, delegable and easy to mark complete. The things this study found that do separate cited pages from ignored ones, the format of the page, its fit to the question and whether it works at all, are less tidy and considerably more valuable.
One honest note on the FAQ result. It is a real difference in a large sample, but pages carrying FAQ markup may differ from other pages in ways beyond the markup itself, since the kind of page that has an FAQ section tends to be built for a particular purpose. We are treating it as a promising signal worth acting on rather than a settled mechanism.
We build technical roadmaps from measurement rather than checklists at Novastacks.
Methodology
89 questions agreed with chief executives and business owners as ones their companies need to monitor, put to ChatGPT, Google AI Overview and Perplexity every day through August 2026, producing 61,536 citations across 9,702 pages, each of which was visited and stored exactly as its server delivered it.
Structured data was read directly from the stored pages rather than inferred. Staying power is measured per combination of page, question and engine across 31 days, covering 8,280 such combinations, which controls for how many opportunities each page had to be quoted. Position is the order in which the engine listed its sources. All brands are anonymised.
Frequently Asked Questions
Does structured data help you get cited by AI?
Not generically. Pages carrying structured data held a citation for 4.49 days on average against 4.31 for pages carrying none, and on position in the answer the pages without markup did marginally better.
Which structured data type works best for AI?
FAQ markup was the only type that separated. Pages carrying it held citations for 5.37 days on average, and 11.9% were still cited after fifteen days against 8.7% of pages with no markup.
Does schema improve your position in an AI answer?
No. Average citation rank was 7.40 for pages with structured data and 7.24 for pages without, a difference within noise. Where markup helps, it helps you stay in the answer rather than rank higher.
Should I remove structured data from my site?
No. It has good reasons to exist, including how your results appear in traditional search. The finding is that it should not sit near the top of an AI visibility roadmap as a growth initiative.
How was this measured?
Structured data was read directly from 9,702 stored pages. Staying power is measured per combination of page, question and engine across 31 days, covering 8,280 combinations.
What Is Actually Moving Your AI Visibility?
We measure which technical and content factors correspond to citations for your brand, so your roadmap is built on evidence rather than on a checklist.