GEO Rankings
← Blog
Published

The Fact Unit Standard: The 4-Part Anatomy of a Claim AI Engines Can Quote (Claim + Number + Named Source + Date)

The fact unit standard: the 4-part anatomy of a citable claim (claim, number, named source, date), with an 8-row pass/fail rubric AI engines can quote.

Bottom line

A fact unit is a citable claim built from four required parts: a plain claim, a specific number, a named source stated inline, and a real date. Miss one part and AI engines paraphrase you instead of quoting you. Named, dated sourcing can lift AI visibility by up to 41%.

Last updated September 2026.

Every content checklist says “add data” or “cite your sources.” Neither instruction is checkable. A writer can follow both and still produce a sentence no AI engine will ever quote.

That gap needs a name. The fact unit is the smallest structure of writing that an AI engine can lift out of a page, quote whole, and attribute correctly. It has exactly four parts. Miss one, and the whole unit fails.

This standard names those four parts, gives you a pass/fail test for each one, and shows the rewrite pattern that turns a vague sentence into a citable one.

Nobody sets out to write an unquotable sentence. It happens by default, because “add data” and “cite your sources” describe a direction, not a shape. A named, four-part shape is checkable. A direction is not.


Anatomy of one fact unit

Here is a complete fact unit, pulled from a real study:

Every word in that sentence does a job. Break it apart and you get the four-part anatomy that makes it quotable:

FragmentPartWhat it does
”pages with a named inline source get cited more often than pages without one”ClaimStates one checkable assertion an engine can quote whole
”2.1 times”NumberTurns the claim into a measurable, specific fact, not a vague trend
”DigitalApplied’s … analysis of 1,000 Google AI Overviews”Named sourceNames who produced the number, inline, so the engine can attribute it
”April 2026”DateAnchors the number to a real point in time, so the engine can judge freshness

Remove any one row and the sentence still reads fine to a human. It stops being quotable to a machine. A claim with no number is an opinion. A number with no source is unverifiable. A source with no date gives the reader no way to judge whether the number is still current.

The four parts do not need to arrive in this exact order. Some of the strongest fact units lead with the number, then name the source and the claim. What has to hold is presence, not sequence: all four parts, in the same sentence or the one right after it.


The four parts, defined

  • Claim. One checkable assertion, stated in a plain sentence. It fails when it hedges (“could,” “may,” “some believe”) or hides inside a longer sentence about something else. A claim that needs two readings to find the point will not survive extraction.
  • Number. An exact, specific figure: a percentage, a count, a dollar amount, or a ratio. It fails when it hides behind a vague quantifier like “many,” “most,” or “a growing number.” A round number is fine. A missing number is not.
  • Named source. The study, organization, or person that produced the number, stated inline in the sentence itself. It fails when the sentence says “a study found” or “researchers say” with no name attached, or when the only attribution is a bare hyperlink with no visible name in the text.
  • Date. The real month or year the data was published or collected, stated inline. It fails when it is missing, or replaced with “recently” or “this year.” A three-year-old number dressed up as current will not hold up once an engine, or a reader, checks it.

The 8-row pass/fail rubric

Score any claim on a page against this table. A claim passes only when it clears the pass row on all four parts. One fail anywhere fails the whole fact unit, because an engine cannot verify what it cannot check.

PartTestVerdict
ClaimStates one checkable assertion in a plain sentencePass
ClaimHedges (“could,” “may,” “some believe”) or buries the assertion in a subordinate clauseFail
NumberGives an exact, specific figure (percentage, count, dollar amount, or ratio)Pass
NumberUses a vague quantifier (“many,” “most,” “several,” “a growing number”)Fail
Named sourceNames the study, organization, or person inline, in the sentence itselfPass
Named sourceSays “a study,” “researchers,” or “experts,” with no name attachedFail
DateStates the real month or year the data was published or collectedPass
DateOmits the date, or uses “recently” or “this year”Fail

Three rewrites, before and after

The rubric only matters if you can see it applied. Here are three real sentences, each failing a different part, rewritten to pass all four.

These three failure patterns account for most of the unquotable sentences on any page you will audit: a claim asserted with no proof behind it, evidence with no name attached to it, and evidence with no date attached to it.

FailsBeforeAfter
Number, source, date”Most companies are seeing better results with AI-optimized content.""DigitalApplied’s April 2026 analysis of 1,000 Google AI Overviews found that pages with a named inline source get cited 2.1 times more often than pages without one.”
Named source”Researchers found that citing your sources boosts your odds of getting quoted by AI engines.""Aggarwal et al.’s GEO study (Princeton, IIT Delhi, Georgia Tech, and the Allen Institute for AI, presented at ACM KDD 2024) found that citing sources lifted AI visibility by up to 40%.”
Date”A recent survey found AI Overviews now show up on almost half of all searches.""BrightEdge’s February 2026 tracking study found Google AI Overviews appeared on approximately 48% of monitored queries, up from about 30% a year earlier.”

Look at what changed in each row. The claim stayed almost the same. What moved was the number, the name, and the date: the three parts a vague sentence leaves out by default.


Where fact units go missing

Most fact units die during drafting, not during review. Surfer’s Content Score flags a paragraph that makes a claim with no number attached to it, before the page ever ships. It checks structure alongside the usual on-page terms, so a writer sees the gap in the editor instead of finding out three months later that the page never earned a single citation. Surfer starts from $99/mo.

A page can clear every row of the rubric above and still not get cited anywhere. Passing the standard and earning the citation are two different questions, and only one of them is visible from inside a text editor. Otterly.AI runs prompt-level citation checks across several AI engines, so you can see whether a specific fact unit is the one an engine actually quotes, not just structurally eligible to be quoted. Otterly.AI starts from $29/mo for a small prompt set, a reasonable first check before committing to a larger rewrite program.

The harder problem is scale. Across 200 published pages, which ones carry a complete fact unit, and which just have a claim floating with no number attached to it? Temso tracks citation share across eight AI engines from $89/mo and surfaces exactly which pages have a citation gap, so your fact unit rewrites target the pages closest to being quoted, not the ones that were never going to be. Its built-in AI SEO agent can turn a flagged gap straight into a drafted rewrite, which matters once the list of pages to fix runs past what one writer can work through by hand.

The full comparison of GEO platforms, with pricing and engine coverage, lives at /rankings/geo-tools. Terms used in this standard are defined at /glossary.


Copy this checklist before you publish

Run every claim on a page against this list before it ships. A claim that fails one line fails the whole fact unit.

  • The claim states one checkable assertion in a plain sentence, not a hedge.
  • The claim carries an exact number: a percentage, a count, a dollar figure, or a ratio.
  • The number’s source is named inline, in the sentence itself, not just linked.
  • The source is a specific study, organization, or person, not “researchers” or “a study.”
  • The date (month or year) the data was published or collected appears in the same sentence.
  • A reader who sees only this one sentence, with no other context, still gets a complete answer.

Start with your five highest-priority pages

You do not need to rewrite a whole site to test this standard. Pick the five pages you most want an AI engine to quote. Run every claim on them through the rubric above. Fix what fails, ship the page, and check citations again in two weeks.

Two weeks is enough time for most engines to re-crawl a changed page and enough time for you to see whether the specific claims you rewrote start getting quoted instead of paraphrased around.

Start by running the rubric against those five pages this week. If you want the gaps surfaced automatically, Temso flags which of your published pages are missing a complete fact unit and tracks whether the fix moves your citation share, all from $89/mo. For how this standard fits the rest of a GEO program, see /methodology.

FAQ

What is a fact unit?

A fact unit is a claim built from four required parts: a plain claim, a specific number, a named source stated inline, and a real date. Each part is checkable on its own. When one part is missing, an AI engine cannot verify the claim, so it paraphrases the page instead of quoting it.

What are the four parts of the fact unit standard?

The four parts are claim, number, named source, and date. The claim states one checkable assertion. The number gives it an exact figure. The named source states, inline, who produced that figure. The date states when the data was published or collected. All four have to appear together, in the same sentence or two.

Why does naming the source matter more than just linking to it?

A bare hyperlink asks a reader, and an engine, to click through to find out who said it. According to DigitalApplied's April 2026 analysis of 1,000 Google AI Overviews, pages with a named inline source get cited 2.1 times more often than pages that cite anonymously or link with no name attached. The name has to live in the sentence itself, not just in the link.

Does adding statistics actually increase AI citations?

Yes, according to controlled testing. Aggarwal et al.'s GEO study (Princeton, IIT Delhi, Georgia Tech, and the Allen Institute for AI, presented at ACM KDD 2024) tested nine content interventions against a retrieval benchmark. Adding statistics lifted visibility by up to 41%, and citing sources lifted it by up to 40%, on the paper's Position-Adjusted Word Count metric.

How do I check whether a page passes the fact unit standard?

Run each claim on the page against the eight-row pass/fail rubric in this standard. A claim passes only when all four parts individually pass: a plain claim, an exact number, an inline named source, and a real date. One failed part fails the whole fact unit, because an engine cannot quote a claim it cannot verify.

What is the difference between a fact unit and a regular statistic?

A regular statistic can stand alone as just a number, like 40% growth. A fact unit requires that number to travel with its claim, its named source, and its date in the same sentence. That is what makes it quotable: an engine can lift the whole unit and still hand the reader a complete, attributable sentence.