How a passive voice detector actually works (and when to ignore it)

"Avoid the passive voice" is the most-repeated rule in English writing advice, and the most misunderstood. Half the people who quote it cannot reliably identify a passive sentence, and a fair number of the constructions they "fix" were never passive in the first place. A passive voice detector is meant to settle the argument by finding the constructions for you — but to trust its verdict you need to know exactly what it is looking for, and where its eyes are weaker than yours.

This piece opens the hood on the passive voice detector and explains the one pattern it chases, why it sometimes flags sentences that are not passive, why it sometimes misses ones that are, and how to read the percentage it gives you without letting a rule of thumb flatten your writing.

What "passive voice" actually is

A sentence is in the passive voice when the grammatical subject is the thing being acted upon rather than the thing doing the acting. "The committee approved the budget" is active — the committee does the approving. "The budget was approved by the committee" is passive: the budget, which receives the action, has been promoted to subject, and the real actor is demoted to a tacked-on "by" phrase or dropped entirely.

The grammatical signature is consistent: a form of the verb to be — is, are, was, were, be, been, being — followed by a past participle, the verb form that usually ends in -ed ("approved", "measured") or, for irregular verbs, something like "written", "taken", or "built". "Was approved", "is written", "were taken". That two-word fingerprint is what every automatic detector, including this one, goes looking for.

The pattern the detector matches

Under the hood the tool splits your text into sentences, then scans each sentence word by word. When it hits a form of "to be", it looks forward up to three words for a past participle. If it finds one, the span from the "be" form to the participle is marked as a passive construction and highlighted. A sentence that contains at least one such match is counted as a passive sentence.

The three-word reach is deliberate, because real passives rarely sit shoulder to shoulder. English loves to slip words between the "be" and the participle: an adverb ("was quietly approved"), a negation ("was not approved"), or a stacked auxiliary ("has been approved", "will be reviewed"). The detector steps over adverbs ending in -ly, over "not", and over helper verbs like have, has, had, will, would, can, could, should while it searches, so these everyday constructions are caught rather than missed.

Recognising the participle itself is the harder half. The tool treats a word as a likely past participle if it ends in -ed (and is more than three letters long, to skip words like "red"), or if it ends in -en (with a hand-built exclusion list so "then", "open", "seven", and "garden" are not mistaken for verbs), or if it appears in a curated list of common irregular participles — "been", "done", "gone", "written", "built", "bought", "caught", "held", "kept", "lost", "said", "sent", "told", "thought", and dozens more. That three-pronged test covers the vast majority of participles you will ever write.

Why it sometimes flags sentences that are not passive

The detector matches a pattern, not a meaning, and the "be + participle" pattern is not unique to the passive voice. The biggest source of false positives is the adjectival past participle — a participle that has hardened into an adjective describing a state, not an action being done. "I am tired", "the door was closed", "she is interested", "the results were mixed". Grammatically these are predicate adjectives; the detector, seeing "be" plus an -ed word, will flag them anyway.

This is not a flaw you can fully engineer away. "The door was closed" is genuinely ambiguous in isolation — it could describe a state (the door was shut, an adjective) or report an event (someone closed the door, a true passive). Distinguishing the two needs the surrounding context and sometimes the writer's intent, which no surface pattern-matcher has access to. The honest stance is to treat every highlight as a candidate, not a conviction, and to glance at each one yourself.

Why it sometimes misses real passives

The mirror-image limitation: a passive built on an irregular participle that is not on the tool's list, or one where the gap between "be" and the participle is longer than three words, can slip through unflagged. Likewise, the detector keys on forms of "to be", so passives formed with "get" — "the window got broken", "she got promoted", common in speech — are outside its net by design. If your passive count looks suspiciously low for prose that feels sluggish, these are the usual culprits.

None of this makes the tool unreliable; it makes it a fast first pass. On ordinary prose it catches the overwhelming majority of textbook passives and shows you each one in context, which is exactly the work that is tedious to do by hand.

Reading the percentage

The headline number the tool reports is the share of your sentences that contain at least one passive construction — passive sentences divided by total sentences, as a percentage. Note the unit: it counts a sentence once whether it holds one passive or three, so the figure tracks how widespread the passive is across your draft, not its raw density.

There is no universally correct target, and chasing zero is a mistake. Plenty of style guides are happy with passives in the low tens of percent. The passive voice is the right choice when the actor is unknown ("the files were deleted overnight"), irrelevant ("the bridge was completed in 1932"), or deliberately de-emphasised, and it is the backbone of scientific and legal registers where the method matters more than the person. Use the percentage as a flag that says "look here", then decide each case on its merits: would naming the actor make the sentence clearer and more direct? If yes, rewrite; if no, leave it.

Putting it to work

The most useful way to run the detector is not to glance at the percentage and move on, but to read the highlighted sentences one by one. Each highlight is a small question: who is doing this, and would my reader be better served if I said so? Most of the time the answer rewrites itself — "mistakes were made" becomes "we made mistakes", and the sentence gains a spine. Occasionally the answer is "the passive is correct here", and you move on with a clear conscience instead of a vague guilt.

That is the right relationship to have with any writing rule turned into software: let it find the candidates fast, and keep the judgement for yourself. Paste a draft into the passive voice detector, read what it lights up, and treat the number as the start of a conversation about clarity — not a verdict on it. If you are tightening prose more broadly, it pairs naturally with the filler word detector and a quick readability consensus check to see whether your edits are landing.