What the statistic and citation density scorer measures
This tool counts the hard evidence in your writing. It scans your content and tallies the concrete, checkable elements that turn an opinion into a substantiated claim: numbers and percentages, dates and years, named studies or reports, dollar figures and quantities, and references to authoritative sources. It then expresses that as a density, typically how many such elements appear per hundred words, so you can compare a thin, assertion-heavy page against a dense, evidence-rich one on a level footing. The output is a score plus a breakdown of which kinds of evidence you have and which you are missing, paragraph by paragraph.
The reason density, rather than raw count, is the right measure is that a long page will naturally contain more numbers than a short one without being any more substantiated. What you actually want to know is whether your writing leans on evidence consistently or only sprinkles a statistic here and there between long stretches of unsupported claims. A page can have one impressive figure in the intro and then five hundred words of vague assertion, and a density view exposes exactly that imbalance. The scorer is built to surface the difference between a page that cites as it argues and a page that argues and hopes you will take its word for it.
Why statistics and citations have become a ranking and citation signal
Search engines and AI engines are both, in different ways, trying to surface trustworthy content, and concrete evidence is one of the clearest machine-readable trust signals there is. A claim attached to a number, a date, and a source is verifiable; a bare assertion is not. AI engines in particular are biased toward content they can quote with a fact embedded in it, because a sentence that carries a statistic is both more useful to the user and safer for the engine to attribute. A page dense with checkable facts gives these systems exactly the kind of material they prefer to lift, while a page of pure opinion gives them little to grab that will not read as ungrounded.
This connects directly to E-E-A-T and to how generative engines weigh sources. Expertise and trustworthiness are hard for a machine to assess from tone alone, but they correlate strongly with whether the writing is backed by data and references. Research into how AI engines choose what to cite has repeatedly pointed at statistics, quotations, and source references as among the factors that most increase the odds of being included in a generated answer. The density scorer turns that insight into a measurement you can act on, instead of a vague instruction to add more proof.
What kinds of evidence the scorer counts
The tool looks for several distinct categories. It detects numeric statistics, including percentages, ratios, and counts, because these are the backbone of any data-driven claim. It picks out dates and years, which both anchor a claim in time and signal freshness, since a 2026 figure reads as more current and trustworthy than an undated one. It flags monetary amounts and concrete quantities, which make claims specific rather than hand-wavy. And it watches for the language of citation, the phrases that introduce a study, a report, a survey, or a named authority, because these signal that a claim is sourced rather than invented.
Crucially, the scorer distinguishes evidence from noise. A page number, a phone number, a navigation count, or a random figure that does not support a claim is not the same as a meaningful statistic, and a good density read tries to weight toward numbers and references that are actually doing argumentative work. The aim is not to reward you for cramming digits into the text, but to reward you for backing your claims. That distinction is what keeps the density score honest: it is measuring substantiation, not the mere presence of characters that happen to be numeric.
How to read your density score
A high density score means your writing consistently pairs claims with evidence, and the page reads as researched rather than asserted. A low score means the page is running on opinion, with long passages that make claims no reader or engine can verify. The breakdown is where the value is: it shows you which sections are evidence-rich and which are evidence-deserts, so you can target the thin ones rather than uniformly stuffing numbers everywhere. A common and useful finding is that the introduction is well-sourced while the body, where the real argument happens, is bare.
Do not chase the highest possible number blindly. There is such a thing as too dense, where a paragraph becomes an unreadable hail of figures that no longer persuades because the reader cannot follow the thread. The goal is a healthy, even distribution where most substantive claims carry some form of proof and the evidence supports the argument rather than smothering it. Read the score as a guide to balance, not a target to max out, and use the per-section view to find the specific places where a claim is floating without anything to stand on.
Common mistakes around statistic density
The most common mistake is the unsupported superlative: words like best, fastest, most popular, or leading, deployed with no number or source behind them. They feel persuasive to write but read as empty to both a skeptical human and an extracting engine, and the scorer flags how much of your page leans on this kind of unbacked language. A related mistake is the orphaned statistic, a striking number quoted with no source and no date, which is actually worse than no statistic because it invites distrust the moment a reader wonders where it came from.
Another frequent issue is stale or undated data, where the figures may be real but the lack of a year makes them impossible to trust as current, which especially hurts in topics that change fast. Some pages over-correct into number stuffing, packing in figures that do not actually support any claim, which inflates the raw count without improving substantiation and can read as padding. And many pages cite vaguely, with phrases like studies show or experts agree that never name the study or the expert, which signals citation without delivering it. The scorer is designed to separate genuine evidence from these hollow gestures toward it.
One more pattern worth naming is front-loading, where all the evidence clusters in the opening section to make a strong first impression and then thins to nothing through the body, exactly where the detailed argument is being made. A reader who skims the intro feels reassured, but a reader who actually engages, or an engine that retrieves a passage from deep in the page, finds bare assertion. The per-section breakdown is built to catch this, because a healthy density read is not just a high average; it is an even spread, so the passage an engine happens to lift, wherever on the page it falls, still carries a fact it can stand behind. Aim to substantiate the body as diligently as the introduction, not to spend all your evidence up top.
How density fits generative engine optimization in 2026
As more search activity resolves inside AI-generated answers, the competition has shifted toward being the source a model chooses to cite, and evidence density is one of the most reliable ways to win that competition. When a generative engine assembles an answer, it is looking for passages it can quote that carry a concrete, attributable fact, because those passages make its answer more useful and more defensible. A page that supplies a steady stream of sourced statistics is, in effect, supplying the engine with a steady stream of quotable material, which raises the odds that your domain is the one credited in the response.
This is why statistic density has become a deliberate optimization target rather than a writing-quality afterthought. It is not about gaming the model; it is that the same trait which makes content trustworthy to a careful human, namely that it backs its claims, also makes it preferable to an engine choosing what to cite. Improving density tends to help across every surface at once, classic rankings included, because verifiable, well-sourced content is what both algorithmic ranking and generative citation are ultimately trying to reward. Adding real evidence is one of the few changes that genuinely improves a page for humans and machines simultaneously.
What to do after you run the scorer
Start with the lowest-density sections the breakdown identifies, because that is where claims are floating unsupported. For each major assertion in those sections, attach a real number, a date, and a named source, replacing soft superlatives with specifics, so best in class becomes a concrete metric and many users becomes a cited figure. Where you quote a statistic, give it a year and an origin so it reads as current and verifiable rather than orphaned. The goal is that a reader could fact-check almost any substantive claim on the page without leaving it.
Then re-balance: if a section became an overdense block of figures, break it up so the evidence supports a readable argument rather than burying it. Refresh stale numbers to the most recent data you can stand behind, since dated and current evidence reads very differently to both readers and engines. Re-run the scorer to confirm the density rose evenly across the page rather than spiking in one spot, and pair it with checks for quotability and citation formatting so the facts you added are also presented in a way an engine can lift and attribute cleanly. Substantiation, distributed well, is the durable win here.