Every grammar checker on the market draws a wavy underline beneath "the report was filed" and suggests "the analyst filed the report." Both sentences are grammatical English. One has an agent who acts on a thing, the other puts the thing first and leaves the agent in a trailing prepositional phrase. The checker prefers the first because Microsoft Word in the late nineties trained a generation of editors to prefer the first, and the heuristic stuck.
The trouble is that the heuristic was always context-blind. Lab write-ups, court filings, security incident postmortems, recipe instructions, and a long list of other genres need passive voice to put the right noun in the subject slot. Forcing the active rewrite in those cases changes the meaning, makes the writing harder to scan, or quietly assigns blame to a person the sentence was not actually about. This piece walks through why the rule exists, where it fails, and how to decide which of the checker's passive flags to accept on any given draft.
Photo by Walls.io on Pexels
What a Grammar Checker Actually Looks For
A grammar checker does not understand passive voice the way an English teacher does. It pattern-matches. The checker scans for a form of "to be" followed by a past participle, optionally followed by a "by" prepositional phrase, and flags any match. "Was filed," "is being reviewed," "were caught," "has been approved" all trigger the same rule.
That pattern works as a recall mechanism. It catches almost every passive construction. It also produces a lot of false positives, because the same verb pattern shows up in legitimate stative descriptions and in adjectival passives that no native speaker would call wrong. "The window is broken" is not passive voice in any meaningful sense, but most checkers flag it. The Stanford Encyclopedia of Philosophy entry on passive voice and the descriptions in Geoffrey Pullum's grammar work both note that automated detection conflates true passives with adjectival predicates and stative copular sentences.
The checker compensates for the false positive rate by softening the alert. It says "consider revising" rather than "this is wrong." The user is meant to read each suggestion and judge. In practice, writers either accept everything to clear the underlines, or ignore everything because the suggestions break sentences they wrote on purpose. Neither response is what the tool was designed for.
Where the Active Rewrite Is Just Worse
Three patterns come up over and over where the passive is the right call and the checker is wrong.
The first is the agentless passive. "The system was breached at 02:14 UTC" is the right opening sentence for a security incident postmortem. The active rewrite, "Someone breached the system at 02:14 UTC," foregrounds an unknown actor the writer has no business naming yet. The agentless passive is how English signals "we know the action happened, we do not yet know who did it, and the focus is on the breach." Forcing an active verb in that slot creates either a fake subject like "an attacker" or a vague subject like "someone," both of which mislead. The SANS Institute incident reporting guidelines describe this convention without using the linguistic terminology, but it is the same thing.
The second is the lab-report passive. "The mixture was heated to 80 degrees Celsius for thirty minutes" puts the object of interest first because the reader cares about the mixture, not about which graduate student stood next to the hot plate. Active voice in a methods section reads as a personal essay, which is the wrong genre. The convention is enforced by every major scientific style guide, including the one in Nature's manuscript guidelines.
The third is the policy-document passive. "Applications must be submitted by August 1" is the right sentence for a public notice. The active rewrite, "You must submit your application by August 1," is fine for some audiences and wrong for others. A regulatory notice that addresses thousands of distinct entities, only some of which are individuals, cannot use a second-person subject without misleading. The passive construction is what allows the sentence to apply universally without naming each applicant type.
In all three cases the grammar checker has no way to know what genre you are in. It sees the verb pattern and flags it. The fix is on the writer, not the tool.
Where the Active Rewrite Is Genuinely Better
The flag is right more often than it is wrong, even if the wrong cases are the loud ones. The kinds of passive voice that deserve the underline include:
The blame-deflecting passive. "Mistakes were made" is the canonical example, and it is bad writing for exactly the reasons the checker says. If your draft has "the deadline was missed" in a status report, the writer almost always meant "we missed the deadline" or "the vendor missed the deadline," and the active sentence is more honest and more readable.
The padded-clause passive. "It was decided by the committee that the proposal would be reviewed" can become "The committee decided to review the proposal" and lose nothing. The original adds eight words and three layers of nominalization for no gain.
The lazy-narrative passive. In a story or a blog post, "the car was driven by Sarah" instead of "Sarah drove the car" almost never has a rhetorical reason behind it. It is a draft sentence that was not edited. The checker is right to flag it.
A grammar checker is a recall device, not a judgement device. It surfaces every candidate and asks the writer to decide. The work is in the deciding. - Dennis Traina, founder of 137Foundry
The skill is being able to tell which kind of passive each flag is pointing at, without thinking about it for thirty seconds per sentence. That comes from reading your own draft against the genre conventions you are writing in, which the checker cannot do for you.
Photo by Ron Lach on Pexels
The Style-Guide Wrinkle
Style guides disagree about passive voice, and the disagreement matters more than most writers realize. The Associated Press Stylebook is more lenient about agentless passives in lede paragraphs because news writing routinely needs to describe events before the responsible party is known or confirmed. The Chicago Manual of Style is more flexible about scientific and academic passives but more strict about narrative ones. The MLA guidelines for humanities papers expect a mix and let the writer decide.
Most grammar checkers ship with a single passive-voice rule that does not branch by style. Some let you pick a profile (academic, news, business, creative) and adjust the strictness, but the underlying detection is still the same regex on the same verb pattern. The profile mostly changes how aggressively the suggestion is phrased, not whether the construction gets flagged at all. The Plain Language Action and Information Network guidance for federal writing is one of the few official sources that explicitly allows agentless passives in certain regulatory contexts, and almost no checker is trained against that allowance.
If you are writing inside an organization that has a house style, the most useful workflow is to skim the house guide once for its passive-voice clause and decide in advance which of the checker's suggestions you will systematically ignore. Trying to make decisions sentence by sentence on a long draft is where editing burnout comes from.
How the Better Checkers Are Getting Smarter
Modern checkers, including the AI Grammar Checker on this site, no longer rely on the bare verb-pattern regex. They look at the surrounding sentence to decide whether the passive is agentless, whether the subject is the topical focus of the paragraph, and whether the genre signals (lab report, news lede, narrative, policy) are present in nearby text. The flag is still raised, but the suggestion is hedged or suppressed when the local context fits one of the known legitimate uses.
This is not a solved problem. The genre signal is fuzzy. A sentence pulled from a lab report and dropped into an email reads differently, and the checker cannot know which document type the user is in unless told. The pragmatic compromise most tools settle on is to keep the flag, suppress the auto-rewrite, and let the user decide. That is the right call until the checkers can read genre with the reliability a human editor can.
Three things separate the better checkers from the worse ones in 2026. The first is whether they distinguish true passives from adjectival predicates ("the window is broken"). The second is whether they downgrade agentless passives that have a clear topical subject in the rest of the paragraph. The third is whether they offer a way to suppress passive flags inside fenced blocks like code samples, quoted dialogue, or long quoted regulations. Tools that do all three save the writer an enormous amount of time on technical drafts.
Photo by Putra Kusuma on Pexels
A Practical Workflow for Reading the Flags
Here is the routine that holds up across genres without making editing a slog.
Read the document type first. If it is a lab report, security postmortem, or policy notice, accept that thirty to fifty percent of the passive flags will be wrong. Skim them but do not stop on every one.
Read the paragraph topic next. The passive is fine when the subject of the sentence is also the topic of the paragraph. "The vulnerability was patched in version 4.2.1" is good if the paragraph is about the vulnerability. The active rewrite, "The maintainers patched the vulnerability in version 4.2.1," is also good if the paragraph is about the maintainers. The checker does not know which paragraph you are in.
Reread the sentence with the active rewrite the tool suggests. If the active version forces you to name an agent you do not want to name, or shifts the focus to a noun that does not matter, keep the passive. If the active version reads more naturally and does not change the meaning, take the suggestion.
Accept all the easy wins. The ten-word "it has been decided that" constructions, the buried-verb noun phrases, the "mistakes were made" deflections. Those are unambiguous improvements and the tool catches them reliably.
For the rest, set a budget. Spend a minute per flag, no more. If a flag is taking longer than that, it is probably a legitimate passive and the right answer is to ignore it and move on. The tool's job is to surface candidates. Your job is to decide. If a candidate is taking real thought, the tool already did its job.
When the Checker Should Be Tuned Down Entirely
Some genres should not be checked for passive voice at all. Legal briefs are one. Federal regulatory text is another. Recipe instructions where the agent is the cook and naming the cook is silly, "the onions are sauteed until translucent," is a third. If you are writing in one of these genres for hours at a stretch, the steady stream of bad flags trains you to dismiss the checker entirely, and you start missing the legitimate ones.
Most checkers, including the AI Grammar Checker, let you disable individual rules. Disabling passive voice for a specific document type is not giving up on quality. It is acknowledging that the rule was built for a different genre and is doing more harm than good on this one. Reenable it on the next general draft. Workflow over purity.
The Purdue OWL writing lab maintains a thoughtful summary of when passive voice is appropriate that is worth bookmarking. The advice has not changed in twenty years because the grammar has not changed. What has changed is the tooling around it, and the tooling deserves more skepticism than most writers give it.
The Short Version
Grammar checkers flag passive voice on every match because their detection is pattern-based and their suggestion is hedged for safety. The flag is right most of the time and wrong in specific genres where the passive is the convention. Lab reports, security incident writeups, and regulatory text all rely on agentless passives that the active rewrite damages.
The skill is reading the genre, accepting the suggestions that improve the sentence, and ignoring the ones that would force a fake agent or change the focus. Twenty minutes spent learning your own style guide's passive clause saves hours of flag-by-flag deliberation over the next year of writing. Modern tools, including the AI Grammar Checker, are getting better at suppressing the bad flags, but the writer's judgement is still the final filter.
If you draft inside an editor that flags passives, decide upfront which underlines you will systematically ignore for each genre and stop relitigating the same suggestion in every document. For more on the writing tools that try to balance recall against false positives, browse the EvvyTools writing and content directory or the EvvyTools blog for related guides.
