The Specific-Subject Rule: Why ‘Researchers Found’ Beats ‘It Has Been Found’

Open any stack of student papers and you’ll find the same ghostly phrasing over and over: “it has been found,” “it can be concluded,” “the data were analyzed.” Nobody is doing anything. Things just sort of happen, by no one, to no one. That habit makes writing vague — and, as a bonus problem, it makes it read more like a machine wrote it.

There’s a simple fix, and it isn’t “never use passive voice.” It’s narrower and more useful than that. Name the agent. Say who did the thing.

Key takeaways

• Agentless passive (“it has been found”) hides who did the work, which makes prose vaguer and flatter.
• The fix is the specific-subject rule: name a real agent — a researcher, a team, a study, a tool — whenever you reasonably can.
• Naming the subject usually drags a concrete detail in with it, and concrete detail reads as more human to both people and detectors.
• Some passive sentences are correct and even better. This is about defaulting to a named agent, not banning a grammar form.
• A quick “who did this?” pass over your draft catches the sentences worth rewriting.

What the specific-subject rule actually says

Here’s the whole thing in one line: when a sentence describes an action, prefer a version where a specific, named subject performs that action.

Compare these two. (I’m using a real example here — when you write your own, plug in your actual source.)

> It has been found that caffeine improves memory.

> Yassa and his team at Johns Hopkins found that a 200 mg dose of caffeine, taken right after studying images, improved how well people remembered them a day later.

The first sentence is true, grammatical, and almost weightless. Found by whom? When? You can’t tell, and neither can your reader. The second one commits. It puts a real actor in front of the verb, and the moment you do that, you’re forced to supply detail — who, what dose, what they actually measured. That extra detail is the point. (That study is real, by the way — Johns Hopkins, 2014 — but the lesson works with whatever source you’re citing.)

This isn’t a grammar-police thing. The American Psychological Association’s own style guidance tells writers to prefer the active voice and to use first person (“we measured,” “I argue”) when describing their own work, rather than hiding behind constructions like “the authors measured.” The convention you may have absorbed in high school — that real scholars write in a faceless passive voice — is, in a lot of fields, just wrong.

Passive voice vs. agentless writing

Quick distinction, because people blur these. Passive voice is a grammatical structure: the object of the action becomes the subject of the sentence (“the sample was heated”). It’s only a problem when it also drops the agent and leaves you with no one responsible.

“The sample was heated by the technician” is passive but not agentless — you still know who acted. “The sample was heated” is passive and agentless. “It has been argued that…” is the worst case: passive, agentless, and abstract on top. The specific-subject rule targets that last category hardest.

Why this matters for sounding human

You came here, probably, because passive-heavy writing keeps getting flagged. So let’s connect the dots honestly.

AI detectors mostly measure two things: how predictable your text is (perplexity) and how much your sentence rhythm varies (burstiness). Agentless passive sentences hurt you on both. They’re abstract, and abstract phrasing is easier for a language model to predict — low perplexity. They also tend to clump into the same droning shape: “It is widely accepted that… It has further been observed that… It can therefore be concluded that…” That’s flat, low-variation cadence, exactly the fingerprint detectors associate with machine text. If you want the math under the hood, we lay it out in the statistics behind why predictable prose gets flagged.

Naming an agent breaks both patterns at once, and you don’t even have to think about the statistics to get the benefit. Watch:

> It is widely accepted that early intervention reduces relapse. It has further been observed that family involvement improves outcomes.

Now with subjects:

> Most clinicians accept that early intervention reduces relapse. The trial I’m citing went further: in their sample, patients whose families joined therapy sessions relapsed less often than those who went it alone.

The second version is more specific, and it varies its sentence shape — because committing to a subject forced a real claim into the open. (Notice I didn’t invent a precise percentage there; if you have one from your source, use it, and don’t round a number you can’t back up.) That’s not a coincidence. Vagueness and uniformity travel together, and so do specificity and variety.

One caveat worth stating plainly: fixing passive voice alone won’t flip a detector score, and you shouldn’t approach it like a trick. It’s one habit among several. A score is a signal, never proof. The real win is that your writing gets clearer.

A mini-scenario: rewriting a paragraph

Say you’re a grad student and your discussion section opens like this:

> It was hypothesized that sleep deprivation would impair decision-making. The hypothesis was supported. It is suggested that future work be conducted to examine longer durations. These findings are consistent with what has been reported in the literature.

Four sentences, zero people. Run the “who did this?” pass on each one.

• *It was hypothesized* — by you. So: “We hypothesized that…”
• *The hypothesis was supported* — by your results. So: “Our results supported it” or “The data backed this up.”
• *It is suggested that future work be conducted* — you’re suggesting it. So: “We think the next step is testing longer sleep windows.”
• *consistent with what has been reported* — reported by whom? Name them. Drop in the actual paper you’re leaning on instead of waving at “the literature.”

Rewritten:

> We hypothesized that sleep deprivation would impair decision-making, and our data backed that up. The obvious next step is testing longer windows — what happens after a full day awake, not just half of one. That fits the prior work we cite below, which found the same drop-off in judgment.

Shorter, in places. Longer, in others. Specific where it was foggy. And it sounds like a person who actually ran the experiment, because — well, you did. The only rule that survives the rewrite: when you name “the prior work we cite below,” you have to actually cite it. Don’t fill the slot with a study you half-remember.

When passive voice is the right call

The rule is “prefer a named agent,” not “exterminate every passive sentence.” Sometimes the agent genuinely doesn’t matter, and forcing one in is worse:

• The doer is unknown or irrelevant. “The samples were stored at minus 80 degrees.” Nobody needs to know which technician closed the freezer.
• The action is the point, not the actor. “The bill was passed in March.” You’re not writing a legislative thriller.
• You’re deliberately keeping focus on the object. “The patient was discharged after 48 hours.” The patient is your subject, fittingly.
• Field convention requires it. Many journals still want the methods section in passive voice to keep attention on the procedure. Follow your professor or your target journal — that’s not the place to fight this battle.

The skill is choosing. A paper where every sentence is agentless reads like a robot. A paper where you’ve consciously picked passive for a handful of sentences and active for the rest reads like a careful human made decisions. That second thing is what you’re after.

A two-minute pass you can run on any draft

You don’t need software for this, though you can search to speed it up.

1. Find the suspects. Search your draft for forms of “to be” (is, was, were, been, being) next to a word ending in *-ed* or *-en*, plus the giveaway phrase “by.” That catches most passive constructions.
2. Ask “who did this?” For each hit, see if the sentence can name an agent without you inventing one. If it can’t answer without a real name, flag it.
3. Rewrite only what matters. Intros, discussions, conclusions, and topic sentences benefit most. Leave correct, intentional passives alone.
4. Read it out loud. Agentless paragraphs drone. If your jaw goes slack reading a stretch aloud, that stretch needs a subject.

If you’re tightening up your craft more broadly, browse the rest of the writing-craft posts for the companion habits, like varying sentence length and choosing the real word over a thesaurus synonym. Editing a lot of long-form work? The plans cover heavier use.

The short version

“Researchers found” beats “it has been found” because it tells the truth more completely. It names who, invites the when and where, and pulls a concrete fact into a sentence that was happy to stay vague. Do that consistently and your writing gets clearer for human readers and, almost incidentally, harder for detectors to pin as machine-made — because it genuinely is more specific.

Want to check whether your fixes moved the needle? Run a draft through the free checker and see which sentences still read flat.