The Rule of Specific Numbers: Why “37%” Reads More Human Than “Many”

Read two sentences back to back. “Many students struggled with the assignment.” “Nineteen of the twenty-six students lost points on the same question.” One of those was written by someone who was there. The other could have been generated by anything, about anyone, on any topic — and that’s the problem.

Vague quantifiers are the quiet tell that gives writing away. They feel safe, they’re grammatically fine, and they say almost nothing. This piece is about one small, high-leverage edit: trading the word “many” for the number it’s standing in for.

Key takeaways

• Vague quantifiers (“many,” “several,” “a significant number”) are the safest, most predictable words a writer can reach for — which is exactly why they read as machine-generated.
• A real number signals that a real person checked something. “37% of the class” implies you counted; “many students” implies you guessed.
• Specific numbers raise the unpredictability of your text, the same statistical quality AI detectors associate with human writing.
• The rule isn’t “add numbers everywhere.” It’s “replace the vague quantifier that’s hiding a fact you actually have.”
• Don’t fabricate a number to sound precise. A made-up “37%” is worse than an honest “many” because it’s checkable and wrong.

Why “many” sounds like a machine wrote it

Think about what a word like “many” actually does. It points at a quantity without committing to one. It’s the linguistic equivalent of a shrug.

Language models love that shrug. When a model predicts the next word after “the study found that,” words like “many,” “several,” and “significant” sit near the top of the probability list. They fit almost any sentence, so they’re statistically cheap. That’s the same reason they leak into human writing when we’re tired, rushed, or padding a word count — we reach for the safe word instead of the true one.

A specific number does the opposite. “37%” after “the study found that” is a low-probability continuation. It’s surprising, in the technical sense. The model didn’t expect it, and neither did the reader, because it carries information the sentence couldn’t have produced on its own. Someone had to go get that number.

This connects to the actual statistics behind detection. If you want the full picture of why unpredictable word choices read as more human, it’s worth a read — but the short version is that detectors measure how predictable your text is, and vague quantifiers are about as predictable as language gets.

A number is a promise that you checked

Here’s the deeper reason this works, and it has nothing to do with detectors.

When you write “37% of respondents,” you’re making an implicit claim: I counted, or someone did, and I’m reporting it. The reader feels that. The number is a tiny credential. It says there’s a real thing behind this sentence — a survey, a tally, a measurement.

“Many respondents” makes no such promise. It could be 51%. It could be 9%. The writer either doesn’t know or doesn’t want to say, and the reader can’t tell which. So they discount the whole sentence a little. Do that across a page and the writing starts to feel like it’s floating above the facts instead of standing on them.

This is why specificity reads as human even when the human is wrong. A friend who says “the bus was 12 minutes late” sounds more trustworthy than one who says “the bus was super late,” even though “super late” might be the bigger problem. Precision signals attention. Attention signals a person.

The “how many, exactly?” test

When you edit, run every vague quantifier through one question: how many, exactly?

• “We got a lot of complaints.” → How many? “We got 14 complaints in three days.”
• “The page loads slowly.” → How slowly? “The page takes 6 seconds to load.”
• “Most people skipped the tutorial.” → How many? “Nine of the first ten users skipped the tutorial.”

If you know the answer, swap it in. If you don’t know but it matters, that’s a signal to go find out before you publish. And if you don’t know and it genuinely doesn’t matter — sometimes “a lot” is fine — at least you decided on purpose instead of defaulting.

A mini-scenario: the same paragraph, twice

Say you’re writing a recap of a product launch. Here’s the vague draft:

> The launch went well. We saw significant traffic and a lot of signups. Many users came from social media, and engagement was strong across the board. Overall, the campaign exceeded expectations.

Now read it again with the quantifiers replaced by numbers you’d plausibly have on hand:

> The launch beat our target. We got 4,200 visitors on day one — triple a normal Tuesday — and 318 signups. About 60% came from a single Reddit thread, not the paid ads we’d budgeted for. Average session ran 2 minutes 40 seconds, up from the usual 50 seconds.

The second version is roughly the same length, but it tells you something. “Significant traffic” could mean anything; “4,200 visitors, triple a normal Tuesday” gives you a baseline and a result. Notice also that the numbers force honesty — the moment you write “60% came from one Reddit thread,” you’ve admitted the paid campaign underperformed. Vague writing lets you hide that. Specific writing makes you face it, and readers respect the writer who did.

That tension is the whole point. Vague quantifiers are comfortable partly because they let us avoid committing. The discomfort of picking a real number is the same discomfort that makes the writing honest.

The rule has a hard limit: don’t fabricate

This is where the advice can go wrong, so read this part twice.

The goal is to replace a vague word with a fact you actually have — not to manufacture precision. If you write “37%” because it sounds more credible than “many,” but you never counted anything, you haven’t humanized your writing. You’ve lied in a way that’s easy to catch. A reader, a teacher, or an editor can check a number. They can’t check “many.”

A fabricated statistic is the worst of both worlds: it has the confidence of data and the reliability of a guess. And the people who build these tools know how slippery measurement is. OpenAI pulled its own AI-writing classifier in July 2023, saying its accuracy was too low to be useful. Around the same time, a Stanford study (Liang et al., 2023) found that detectors frequently flagged essays by non-native English writers as AI-generated. Precision without grounding is its own failure mode. Don’t add to it.

So when you don’t have the number:

• Find it. Pull the real figure from your data, the source, the analytics dashboard.
• Label the estimate. “Roughly a third,” “about 200 words,” “somewhere north of 50” are honest. They’re still more concrete than “many,” and they don’t pretend to a precision you lack.
• Leave it vague on purpose. If the exact count truly doesn’t change anything, “a few” is fine. Just make it a choice.

Where this fits in the bigger humanizing picture

Specific numbers are one move in a larger habit: writing from what you actually know rather than from what’s easy to type. It pairs with varying your sentence length, using contractions, and cutting filler — small edits that, together, shift writing from generic to grounded. If you want more of these, browse the rest of our writing guides. And if you’re weighing whether a tool fits your workflow, here’s what a humanizer pass actually costs.

Frequently asked questions

Does using specific numbers really make writing pass AI detectors?

It helps, but it’s not a trick. Detectors measure how predictable your text is. Vague quantifiers like “many” and “significant” are exactly what a language model reaches for by default, so they read as low-surprise and machine-like. A concrete, slightly odd number (“41 minutes,” “3 of the 7 reviewers”) is harder to predict, which nudges the statistics toward human. But chase clarity, not the score — write with real numbers because readers trust them, and detection tends to follow.

What if I don’t have an exact number?

Then don’t invent one. You have a few honest options: find the real figure from a source, give an estimate and label it (“roughly a third,” “about 200 words”), or keep the vague word if the precise count genuinely doesn’t matter. A fabricated “37%” is worse than “many” because anyone can check it and catch you. Specificity is retrieval, not invention.

Won’t too many numbers make my writing dense and unreadable?

Yes, if you stuff one into every clause. The rule is selection, not saturation. Pick the one number that proves your point and let the rest breathe. A paragraph with a single well-placed figure lands harder than a paragraph that reads like a spreadsheet. If a number doesn’t change what the reader concludes, it’s clutter.

Are round numbers as good as precise ones?

Often less convincing. “About 100 people” reads like an estimate; “94 people” reads like a count. Oddly specific numbers carry an implicit claim that you actually measured. That said, a fake-precise number you can’t back up is worse than an honest round one — don’t write “94” if you only know it was “around 100.”

Where should I look for vague quantifiers when I edit?

Scan for a short list of usual suspects: many, several, a number of, significant, substantial, various, numerous, a lot of, the majority, countless. Each one is often a fact in disguise. Ask “how many, exactly?” If you know, swap it in. If you don’t and it matters, go find out.

The takeaway

“Many” is a word you can write without looking up from your desk. “37%” means you got up and counted. Readers feel that difference even when they can’t name it, and detectors measure it whether you meant them to or not. The fix is small and it’s mostly subtraction: find the vague quantifier, ask how many, and tell the truth you already know.

Want to see which of your sentences are reading as predictable before you hit publish? Run the draft through PaperBleach and start with the ones it flags — odds are they’re the vaguest sentences on the page.