From content to citation.

Two studies, two different questions

Most advice about "AI-friendly content" blurs together two claims that are actually separate and were tested separately. One is about what you say — the words, the claims, the sources you cite. The other is about how it's arranged — heading structure, paragraph length, whether a claim sits inside a block an engine can lift on its own. They're worth keeping apart, because a business can be strong on one and weak on the other, and the fix for each looks nothing like the fix for the other.

The first line of research is a benchmark study that tested concrete content modifications — adding quotes, adding statistics, citing sources, writing more confidently, stuffing in keywords — against a large set of real queries, then measured how often and how prominently each version showed up in generated answers. The second is a more recent study, from March 2026, that took a different approach entirely: it held the wording, claims, and sources of a page completely constant and varied only the structure around them, then measured citation rate across six answer engines. Same underlying content, different shell. Both studies point the same direction — content that's specific and well-organized gets quoted more — but they isolate different variables, so it's worth looking at each on its own terms before combining them.

What actually moves the needle

Ranked by effect size, the content-level benchmark found a clear hierarchy. Adding relevant quotations from credible sources was the strongest single tactic tested, producing roughly a 41% lift in a position-adjusted visibility metric and a 28% lift in a separate subjective-impression score. Adding statistics and other quantitative figures came next, at roughly 31% and 23% on the same two metrics. Citing identifiable sources by name performed close behind quotations. Confident, authoritative phrasing helped, but by a smaller margin than any of the above. And the tactic most businesses reach for reflexively — keyword stuffing — was among the weakest performers tested, sometimes barely distinguishable from doing nothing at all.

Tactic	Relative effect	Where it ranked
Quotations from credible sources	~41% / ~28%	Strongest
Statistics & quantitative figures	~31% / ~23%	Second
Citing identifiable sources	Close behind	Third
Confident, authoritative phrasing	Smaller positive effect	Fourth
Keyword stuffing	Weak to negligible	Weakest

Across the full benchmark, well-executed changes reached up to roughly 40% improvement in visibility, and a separate real-world test against live engines showed gains up to roughly 37%. The pattern underneath all of it is consistent: engines favor content that reads like it could stand on its own in a footnote — specific, attributable, checkable — over content that reads like it's trying to rank. The research is explicit that these are benchmark-level results, not a guarantee for any individual page, and that effectiveness varies by domain. A weak page doesn't get rescued by bolting on a statistic.

Structure alone changed citation rates

The second study is the more surprising one, because it isolated structure from everything else. Researchers took existing content, left the words, claims, and sources exactly as they were, and changed only the formatting — heading hierarchy, how information was chunked into sections, how it was visually organized on the page. Nothing new was written. Measured across six answer engines, that alone produced a consistent 17.3% improvement in citation rate, plus an 18.5% average improvement in how the content was rated for quality — a proxy metric, not a claim about the words themselves, which hadn't changed.

Not every layer of structure mattered equally. Macro-structure — the overall heading hierarchy and document architecture — had the broadest effect, because it shapes how an engine parses the page as a whole before it ever gets to individual sentences. Meso-structure — breaking information into self-contained chunks, each one a unit that makes sense read in isolation — had the strongest effect on whether an individual passage extracted cleanly. Micro-structure, meaning visual emphasis like bold text or italics, mattered least of the three.

The practical implication is worth sitting with: the highest-return first move for a lot of businesses isn't writing more, it's reorganizing what they've already published. Google's own guidance is consistent with this — no special markup is required for AI features, structured data should simply match what's visibly on the page, and the quality guidance centers on demonstrating real experience, expertise, authoritativeness, and trust, with trust called out as the most important of the four. Google frames it as three plain questions for any page that matters: who made this, how was it made, and why does it exist.

Where the evidence stops

None of this adds up to a formula. No tactic on either list guarantees a citation, on this page or any other. Answer engines are black boxes with behavior that shifts over time and varies by query, by domain, and by what else is available to cite instead. Structured data doesn't force an engine's hand — it makes a page easier to parse correctly, nothing more. And piling statistics or quotes onto a page that's substantively weak doesn't rescue it; the underlying content still has to hold up. These are high-probability improvements grounded in published research, not outcomes anyone can promise you in advance. Treat every figure here the way the source research treats it: directionally real, not a contract.

Three things worth auditing on any page you care about

Both studies point back to the same practical framework. It applies to a page you already have, not just to something new you'd write.

• Understandability. Could a stranger — or an engine — describe what your business does, who it serves, and why, from the headings and visible structure alone, without inferring? A blunt test: strip a page down to just its headings and see if it still tells a coherent story. If it doesn't, an engine skimming that same structure won't either.
• Credibility. Does the page show real evidence of who's behind it — a named author with actual credentials rather than an anonymous "admin" byline, specific case studies instead of vague success claims, visible contact information, claims that are transparent and checkable rather than asserted? This is Google's own test again: who made this, how, and why.
• Citation-worthiness. Could any given paragraph be lifted on its own and still make sense, with something to attribute it to? Answer-first blocks, clear definitions, comparison tables, FAQ content that answers a real question, original data or proof points a competitor can't just copy — these are the shapes an engine can actually extract.

What this looks like in practice

The failures that show up most often are less about effort than about where the effort goes. A few patterns worth checking for directly:

• Length standing in for quotability. A long page with no liftable, attributable claim is not more citable than a short one — it's just longer. Volume was never the variable that moved the needle in either study.
• Keyword stuffing, still. It tested as one of the weakest tactics in the content-level benchmark, and it's still the first instinct for a lot of sites. Replacing it with a specific, sourced sentence is usually a net time savings, not an added cost.
• Generic assertion instead of checkable evidence. "Leading provider of..." carries no information an engine — or a buyer — can verify. A specific number, a named source, or a real example does the same rhetorical job and is the thing that actually gets quoted.
• Headings sized for looks, not for hierarchy, and no named author. A heading that's just bold, bigger text tells a reader nothing structural; a heading phrased as the actual question a buyer would ask does. And publishing without a named, credentialed author undercuts the trust signal Google calls out as most important — it also gives an engine nothing to corroborate you against.

None of this requires a content overhaul to start. Given the structure study's result, the first pass on most existing pages should be rearrangement, not rewriting — the same words, organized so a stranger and an engine can both follow them without inferring.

Frequently asked

Do I need to write more content to be cited more often?

Not necessarily. The March 2026 structure study found a 17.3% citation-rate improvement from reorganizing existing content alone, with no new words added. For most businesses, restructuring what's already published is the cheaper and faster first move; writing more only helps if the new content is specific and sourced, not just longer.

Does adding statistics or quotes guarantee my page gets cited?

No. Both tactics measurably improved visibility in benchmark testing — quotations from credible sources were the strongest single lever, statistics the second strongest — but the research is explicit that these are benchmark-level probabilities, not guarantees for any individual page, and results vary by domain. A page that's substantively weak isn't rescued by adding a statistic to it.

Is keyword stuffing still worth doing for AI visibility?

The evidence says no. In the content-level benchmark, keyword stuffing was among the weakest-performing tactics tested, well behind quotations, statistics, and named sources. Replacing repeated keywords with a specific, attributable claim tends to help more, not less.