The word "most" is where a number was supposed to go. When you write "most professionals use HubSpot," you are describing a population you can reach directly, because the readers in front of that sentence are the professionals it claims to count. The reflexive fix is to go find a source and bolt on a citation, and that move does one thing: it tells your reader which other page to trust instead of yours.
A better option is open to anyone who publishes, and you can turn vague stats into your own data without borrowing anyone else's figure. Poll the audience already on the page, let them report what they actually do, and the number becomes a first-party statistic only you are positioned to cite. Every soft quantifier sitting in your back catalog is a coordinate, and each one points at a measurement your own readers could hand you this week.
What It Means to Turn Vague Stats Into Your Own Data
To turn vague stats into your own data is to take a sentence like "most professionals use HubSpot" and replace the guess with a number you measured. Because your readers are instances of that population, the claim is a poll question in disguise: ask what they actually do, and their answers become a first-party statistic only you can cite.
Underneath "most professionals use HubSpot" sits a percentage you wanted and did not have. The word carries the full weight of a measurement while standing in for one that never happened, and the figure it replaces is closer than it looks.
A Vague Quantifier Is a Measurement You Skipped
You back up claims with data when the data exists. A vague quantifier marks the spot where it did not, and where you shipped the sentence anyway. It is an unmet data requirement wearing prose, and no synonym polishes it out. It sits in published pages where no process ever flags it.
The tooling built to catch weak claims cannot see this one. LiquiChart's claim extractor reads a page looking for statements it can monitor over time, and it requires a number, a date, or a source to do that. A sentence with a figure, "72% of teams invest in content marketing," gives it something to track. A sentence that says "most professionals use HubSpot" gives it nothing, so the extractor drops the sentence on the floor.
The instrument that should flag the problem is structurally blind to it.
The consequence runs deeper than one tool. I keep seeing it in our own decay studies: the sentences worth polling are the ones the machinery throws away before measurement begins, so any freshness audit built the same way undercounts the entire category. This is a different gap from the one in our study of claims that cite no source at all, where the number is present but its backing is missing. Here the number was never there. The claim is value-less by construction, which is exactly why it is the richest thing you could measure.
Every "Most" Is a Number Your Audience Can Supply
The reason a vague quantifier is recoverable is that you already named the population. "Professionals" is the audience already loading the page, and a sentence about what that group does is a sentence its own readers can answer.
The draft already wrote the poll. All that is missing is the asking.
That reframes the back catalog. Each soft quantifier becomes an entry on a worklist, a place where one question would convert an assertion into a measurement. This is the same move that earns original data as a ranking signal: the page stops repeating what is already known and starts contributing what only it can. The mechanics of writing and embedding the question itself are covered in how to create a poll for a blog, so the work here is upstream of that: deciding which sentence is worth the asking.
When the sentence wanted a number and you did not have one, what did you actually do?
Each of those four paths sets a different ceiling on how far a reader can trust the sentence. The population that sentence describes is already here, reading it. No one has put the question to them yet.
The data is one poll away.
What Makes a Vague Stat Convertible
Not every vague claim can become a poll, and the line between the ones that can and the ones that cannot is the whole discipline. Ask a reader which tool they use and you get a measurement. Ask whether "most people" use it and you get a guess about strangers. The first question aggregates into the number your sentence was missing. The second produces an opinion about a population the reader cannot see, which is a different claim entirely and a weaker one.
So the bar is specific: a vague stat is convertible only when the reader can self-report their own behavior, preference, intent, or belief, and when those self-reports add up to the exact subject the sentence asserts. "Do you use a formal content calendar?" clears the bar, because the reader answers from direct experience and the aggregate measures the claim. "Do you think most teams use a formal content calendar?" fails it, because you are now polling a belief about other people.
This is the difference between self-reported data and opinion. No roundup of survey tips teaches it, because the rule only matters once you point a poll at your own published claims rather than fish for a topic. The test is mechanical: rewrite the vague sentence as a question addressed to one reader about themselves. If it reads naturally and its answers would produce your missing figure, the claim converts. If the only natural question asks the reader to estimate what others do, predict the future, or rank one option against another, it does not, and forcing it would hand you a number that means nothing.
Why Borrowing a Stat Keeps Your Page One Hop From the Source
The instinct, when a claim feels thin, is to find a study and cite it. It feels responsible, and it is faster than running anything yourself. It also hands your authority to someone else's domain.
A borrowed stat does not make your page the source. It makes your page the detour.
When you cite, you become one link in a chain that points away from you, toward whoever measured the thing first, which is the same dynamic behind why AI cites third-party sources over the pages that merely repeat them. Polling your own audience inverts the direction. The figure now originates on your page, with the people reading it, and the next writer who needs that number has to point back at you.
The honest objection is that a self-selected audience is not a representative panel. That is fair. A citable poll earns trust through disclosure rather than size: a clearly described self-selected sample is more defensible than an undisclosed large one. You are reporting what your readers told you, bounded to that audience and disclosed as such, which is a thing only you can report.
How the Engine Picks a Question Worth Asking
The workflow starts on a page you own. The originality engine reads it sentence by sentence, and a deterministic detector flags the shape a writer reaches for when the figure does not exist: a vague quantifier governing a population, "most professionals use HubSpot," with no number anywhere in the line. That first pass is tuned for reach, so it forwards more candidates than it should keep.
The judgment lives in the second pass. A precision veto takes one flagged sentence and asks a single thing of it, whether a reader of this page could report the answer from their own behavior, and whether those answers would aggregate into the number the sentence is missing.
The veto starts from yes and looks for a reason to decline. The bar is the self-report test: the claim clears it only when the reader can answer from direct experience, and fails the moment the poll would measure a guess about strangers. Only a line that survives both passes earns a recommendation, and only above a fixed confidence threshold. Anything the model is unsure about resolves to a skip.
Silence Is the Feature
What you see at the end is one recommendation, "Back this claim with your own poll," carrying a proposed question written in the second person, along with a starter set of mutually exclusive answer options. Acting on it opens poll creation seeded with that question, leaving you to set the options and publish. The engine stops there by design. It does not bring the poll into existence, and it does not touch the sentence in your draft. The figure that eventually replaces "most" arrives only after your own audience votes, which is why "76% of professionals use HubSpot" is a worked example here and nothing the platform could hand you in advance.
Restraint is the part that makes the rest worth trusting. A recommendation you can act on every time it appears is worth more than a longer list you have to triage, so a sentence that would yield trivia, or a belief about people the reader has never met, produces nothing at all. The cost of a recommendation that should not have fired is higher than the cost of one the engine let pass. I tuned it around that asymmetry. Feed it a claim it cannot confirm is self-reportable, and it hands you nothing.
How to Turn Your Back Catalog Into a Poll Worklist
Point the engine at a page you already published. It reads that page as one of your Monitored Pages, runs the detect-and-veto pass across every sentence, and hands back the vague quantifiers worth converting: a short worklist of proposed poll questions. You pick one. Open the pre-filled question in poll creation, set it live, embed it back into the post. The moment your readers begin voting, the answer becomes a first-party claim the platform tracks, and your Originality Score moves, because that metric rises only as the share of claims backed by your own data grows.
A faster first look is available before you convert anything. The Content Health Scanner reads any URL with no login and returns its unattributed and stale claims: the hard numbers sitting in your prose with no source behind them, and the citations pointing at pages that have died or aged out of date. Those are the value-bearing cousins of the vague quantifier, the claims that already carry a figure but have lost their backing. The figureless ones are what the originality engine surfaces instead, because a scan built to grade numbers and sources has nothing to grab onto in a sentence that contains neither. The two passes cover different halves of the same problem.
When the Number Moves, the Loop Catches It
The conversion is not a one-time fix. The poll result lives in the post inside a living content block, and when later votes shift the leading answer, the block updates the figure in its own sentence to match. The claim you generated this quarter does not slip back into vagueness next quarter, because the same infrastructure that surfaced the gap keeps the number it produced current. The number you measured stays accurate for as long as your audience keeps answering.
The Page Where the Number Starts
Every soft quantifier you leave in your archive is first-party data you give away on a weekly schedule. The number was sitting there the whole time, one question away from the people already reading the claim, and each week you publish around it is a week you cede the figure to whoever measures it next. The same logic applies before you write the next post, where the stronger plan starts from utility content over keywords: generate the data the topic needs instead of borrowing it.
Replacing "most" with a number you measured is the smallest version of a larger shift, and it is the change I would make first. A page can repeat what is already known, or it can be where something becomes known, and the difference is whether you ran the poll. The page where that number starts is a position, and right now it is open.