**TL;DR** — Across 26 client sites through May 2026 we audited a writing choice that lives entirely inside a single number: whether a statistic on the page is stated with **exact precision and its unit attached** (for example, "loads in 1.8 seconds" or "cut bounce rate by 34%") or left rounded, hedged or unit-stripped ("loads in about two seconds", "cut bounce rate by roughly a third"), and whether that precision changes how often the AI Overview lifts the number into the answer card. Across 6,940 cited-passage events that contained at least one quantitative claim, we joined each cited number to the precision state of how it was written. The headline is that numeric precision is a real and badly underused citation lever, but only when three things line up. A statistic written with an exact figure and its unit was cited 2.4× more often than a matched passage that stated the same fact as a rounded or vague quantity. The strongest predictor was unit attachment — a number carrying its unit inline (%, seconds, USD, per month) was lifted far more than a bare figure the reader had to infer the unit for. The second was figure exactness — "34%" beat "about a third" and "roughly 30%" by a wide margin, as though the composer reads a precise number as a more citable fact. The third, and the warning, was figure-claim agreement — a precise number that did not actually answer the query, or that contradicted a more authoritative source, was cited no more often and on 6% of pages was cited less. One change — rewriting vague quantities as exact figures with their units, on the sentences that answered the query — lifted cited-passage rate by 27% on the affected sites over a 30-day follow-up.
Why we ran this audit
A number is the most quotable thing on a page. When a query asks "how fast", "how much", "what percentage", the answer the user wants is a figure, and the AI Overview composer has to decide which figure on which page to lift. Writers, though, treat the precision of a number as a style choice — we round for readability, hedge for safety, drop the unit because context makes it obvious to a human reader — and we rarely think the choice has any bearing on whether the number gets cited. But the composer is not a human reader filling in context; it is a model deciding which passage states the answer most cleanly, and a rounded or unit-stripped number is a less complete answer than an exact one with its unit attached. We suspected the composer was treating numeric precision as a citability signal — a precise, unit-carrying figure reads as a fact ready to lift, a vague one as something the composer would have to firm up itself — and we wanted to know whether that was real or whether a number is a number to the model regardless of how it is written.
The second motivation was a pattern we kept hitting on pages that had the right number but did not get cited. Several clients stated a genuinely strong statistic but wrote it loosely — "we cut load time by about half", "most users finish in under a minute" — while a thinner competitor stated the equivalent number exactly, "cut load time by 51%", "median completion is 47 seconds", and the competitor took the chip. The obvious hypothesis was that the precise figure read to the composer as a liftable fact while our hedged version read as an approximation it could not quote cleanly. We needed to know whether precision was doing that work, because if it is, tightening a vague number into an exact one is among the cheapest citation interventions that exists — it changes no claim, adds no content, and rewrites a single quantity.
How we ran the measurement
26 client sites — 10 SaaS, 6 publisher, 6 B2B services, 4 DTC — each with a fixed 200-query basket of its real in-market queries, filtered to the quantitative subset where the answer the user wanted was a number (a speed, a price, a percentage, a count, a duration). Twice daily through May 2026 we captured every AI Overview card, and for cards citing a client page we identified the specific lifted passage and recorded the precision state of the number it carried. We classified each cited quantitative passage on three axes: unit attachment (was the unit stated inline with the figure, or left for the reader to infer), figure exactness (was the number stated precisely, rounded to a soft value, or hedged with "about / roughly / around / most"), and figure-claim agreement (did the number actually answer the query, and did it agree with the more authoritative sources on the same fact). We built a matched control for every cited number: a comparable passage on a similar query whose number carried a different precision state, so the comparison was precise-vs-vague rather than good-page-vs-bad-page. The cited quantitative cohort was 6,940 events.
Two normalisation moves matter. We scored precision state independently of passage quality, because an exact number can sit inside a weak paragraph and a vague one inside a strong paragraph, and we wanted to isolate the number rather than re-measure the prose around it. And we matched on passage citability before comparing precision — we paired each cited number with a control our existing cited-paragraph rubric scored as equally liftable (self-contained, concrete, right length), so the effect we attribute to precision is not just the precise pages being better-written overall. The 2.4× and 2.7× figures are from those matched comparisons, not raw averages across unmatched pages.
The shape of the numeric-precision pattern
The flat headline first. Precise numbers are cited more. A statistic written with an exact figure and its unit was cited 2.4× more often than a matched passage stating the same fact as a rounded or vague quantity. The effect survived the quality match and the citability control: among passages our rubric scored as equally liftable, the ones whose number was precise and unit-carrying were lifted far more than the ones that left it soft. The composer behaves as though it reads a precise figure as a fact ready to quote and a vague one as an approximation it would have to commit to itself — so the precise number is, in effect, the page handing the composer a citation-ready answer.
The most decision-relevant cut was that the gain is about the number, not the page. We had half-expected a page-level effect — "pages that state numbers precisely are rigorous pages and those get cited." That existed but it was weak. The strong effect was local: on the same page, a sentence with an exact figure was lifted while an equally relevant sentence two paragraphs down that hedged its number was skipped, and tightening the second number often brought it into citation without touching the first. Numeric precision is not a page-wide credibility credential; it is a property of the individual figure, which means the fix is targeted — make exact the numbers you want cited, not every sentence on the site.
Driver one: attach the unit to the figure
The single strongest predictor was whether the number carried its unit inline. Holding the passage constant, a figure stated with its unit attached — "1.8 seconds", "34%", "USD 1,200 per month" — was lifted at 2.4× the rate of the same figure with the unit left implicit, where the prose expected the reader to know from context that the number was seconds or dollars or a percentage. The composer appears to treat a unit-carrying figure as a complete, self-describing fact it can lift verbatim, and a bare number as an incomplete one it would have to annotate. A human reader fills the unit in from context; the composer, lifting a single passage out of the page, has no surrounding context to fill it from, so the unit has to travel with the number.
We ran a structural test on 19 passages across 10 clients, each stating a strong number with the unit left implicit — a table column header three rows up, a sentence two paragraphs earlier that established "all times in seconds". We rewrote each so the unit travelled inline with the figure in the answer sentence itself, changing no values. Over the 45 days that followed, 13 of the 19 passages began being lifted on at least one target query where they had previously been skipped. The lever was not a new number; it was making the number self-describing in the one sentence the composer would lift, so the figure carried its own meaning out of the page.
Driver two: state the exact figure, not a soft round
Holding unit attachment constant, the second driver was figure exactness. A precisely stated number — "34%", "47 seconds", "ranked 3rd" — was lifted far more than the same fact rounded to a soft value or hedged with an approximator: "about a third", "under a minute", "one of the top". The reading consistent with the data is that the composer reads an exact figure as a fact the page is committing to and a hedged one as an estimate it would have to firm up, and it prefers to lift the fact it does not have to underwrite. "About a third" forces the composer either to quote a vague phrase into a card that wants a number or to compute the precise value itself; "34%" hands it the value it can quote directly.
We ran a structural test on 16 passages across 8 clients that stated genuinely known numbers in hedged form — the team had the exact figure in their analytics but wrote it loosely for readability. We replaced the approximator with the exact figure, changing the fact to its true precise value rather than inventing precision. Over the 60 days after the change, 11 of the 16 passages improved their cited-passage rate. The two drivers compound: an exact figure with no unit is a half-built signal and a unit-carrying approximation is the other half, and the passages that won carried both — a precise number with its unit attached, stated in the sentence that answered the query.
Driver three: figure-claim agreement, and the precise number that backfires
The third driver was the warning. A precise number that did not actually answer the query — an exact figure for a related-but-different metric, a precise stat that contradicted the more authoritative sources on the same fact — was cited no more often than a vague one, and on 6% of audited pages a page stating a precise but wrong or off-target number was cited measurably less than the same page with no number at all. The reading consistent with the data is that precision is only an asset when the number is the right one: the composer is drawn to the exact figure, checks it against the query and against other sources, and where it finds the precise number answers a different question or disagrees with the consensus, it treats the page as confidently wrong — and a confidently wrong number appears to cost more than a hedged one would. Figure-claim agreement, not precision alone, is what the signal rewards.
We confirmed this on 13 pages across 7 clients where a precise number was either off-target — answering a near-neighbour question rather than the query the passage ranked for — or stale, contradicting a figure the rest of the web had updated. We re-pointed the precision onto the number that actually answered the query, and refreshed the stale figures to agree with the current consensus, changing the values to be correct rather than merely exact. Over the following 45 days, the re-pointed and refreshed pages recovered citation on their target queries, better than both the wrong-precise state and the vague state they could have been left in. The actionable rule is blunt: state the exact number that answers the query and make sure it is right, because a precise wrong number is worse than an honest hedge.
What changed in our content checklist
Three changes. We rewrote our numbers guidance from "round for readability" to "state the exact figure with its unit" on any sentence that answers a quantitative query: the number the user is asking for should be precise and carry its unit inline, not rounded for prose rhythm or left to context. We added a unit-attachment rule to the same pass: a figure in an answer sentence must travel with its unit, never relying on a table header or an earlier sentence to supply it, because the composer lifts the sentence without that surrounding context. And we added a figure-claim agreement gate: the precise number must answer the query and agree with the authoritative sources, because a confidently wrong figure hurts more than a hedge — precision is only worth claiming on a number we can stand behind.
We dropped one habit. For years our house style had rounded numbers for readability as a reflex — "about half", "roughly a third", "under a minute" read more smoothly than the exact values, and we assumed the smoothing was free. The audit removes that default for any sentence answering a quantitative query: rounding spends the one thing the composer most wants to lift, an exact citable figure, in exchange for prose rhythm the composer does not read. So reflexive rounding left our playbook for answer sentences — we now state the exact figure with its unit where a number answers the query, and reserve rounding for narrative passages where no citation is at stake.
- 01State the exact figure with its unit. A precise, unit-carrying number was cited 2.4× more than a matched passage stating the same fact as a rounded or vague quantity — the composer lifts a citation-ready number.
- 02Attach the unit inline. A figure carrying its unit in the answer sentence beat the same figure with the unit left to context; 13 of 19 passages were cited after the unit travelled with the number — the composer lifts the sentence without surrounding context.
- 03Exact beats a soft round. "34%" and "47 seconds" beat "about a third" and "under a minute"; 11 of 16 passages improved after the hedge became the exact figure — the composer prefers a fact it does not have to firm up.
- 04Make the precise number right. A precise but off-target or stale figure was cited no more than a vague one and hurt on 6% of pages — state the exact number that answers the query and that agrees with the sources, never a confidently wrong one.
Where this argument breaks
For navigational, brand or transactional queries there is usually no number the user is asking for — someone searching your brand or a SKU is not after a statistic, and forcing a precise figure into a passage to court a citation that will not come is wasted effort; the precision lever is for quantitative informational queries where the answer is a number, not for every page. For genuinely uncertain quantities, false precision is worse than an honest range — stating "34%" when the real figure is a noisy estimate that could be anywhere from 25% to 45% is exactly the confidently-wrong failure the third driver punishes, and there the honest move is a stated range or a clear confidence qualifier, not invented exactness. For numbers that change often — live prices, current counts, today's figures — the precision has to be paired with freshness, because an exact-but-stale number is the stale-contradiction failure, and the win there is in keeping the precise figure current, not merely in making it precise once. For YMYL topics (health, finance, legal) the figure-claim agreement bar is far higher than our averages suggest — the composer appears to penalise a precise wrong number much more heavily there, so on those topics a figure is only worth stating exactly if it is sourced, current and correct. The 6% wrong-precise penalty is small and noisy; we are confident a precise wrong number does not help and mildly confident it hurts, but that is the weakest finding in the audit and we would not restructure a page on it alone. For Chinese-language AI search (文心一言, 元宝, 通义) the precision effect was present but weaker in our parallel audit, and the engines were more tolerant of a unit left to context, though exactness still helped — a hedged number bought less there than in Google's AI Overview but the gap was narrower. Our window was 60 days and the cohort was 26 sites; the multipliers are point estimates that will move by vertical and query type. Outside those carve-outs the lesson holds: in 2026 the AI Overview lifts a number that is precise, carries its unit, and answers the query correctly far more readily than a rounded, unit-stripped or off-target one, the unit is the individual figure rather than the page, and the cheapest citation win on a quantitative answer is to state the exact number with its unit in the sentence that answers the query — and to make sure that number is right.